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

   1 /*
   2  * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
   3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
   4  *
   5  * This library is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU Lesser General Public
   7  * License as published by the Free Software Foundation; either
   8  * version 2 of the License, or (at your option) any later version.
   9  *
  10  * This library is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * Lesser General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU Lesser General Public
  16  * License along with this library; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18  *
  19  */
  20
  21 /**
  22  * @file h264.c
  23  * H.264 / AVC / MPEG4 part10 codec.
  24  * @author Michael Niedermayer <michaelni@gmx.at>
  25  */
  26
  27 #include "common.h"
  28 #include "dsputil.h"
  29 #include "avcodec.h"
  30 #include "mpegvideo.h"
  31 #include "h264data.h"
  32 #include "golomb.h"
  33
  34 #undef NDEBUG
  35 #include <assert.h>
  36
  37 #define interlaced_dct interlaced_dct_is_a_bad_name
  38 #define mb_intra mb_intra_isnt_initalized_see_mb_type
  39
  40 #define LUMA_DC_BLOCK_INDEX   25
  41 #define CHROMA_DC_BLOCK_INDEX 26
  42
  43 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
  44 #define COEFF_TOKEN_VLC_BITS           8
  45 #define TOTAL_ZEROS_VLC_BITS           9
  46 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
  47 #define RUN_VLC_BITS                   3
  48 #define RUN7_VLC_BITS                  6
  49
  50 #define MAX_SPS_COUNT 32
  51 #define MAX_PPS_COUNT 256
  52
  53 #define MAX_MMCO_COUNT 66
  54
  55 /**
  56  * Sequence parameter set
  57  */
  58 typedef struct SPS{
  59
  60     int profile_idc;
  61     int level_idc;
  62     int multiple_slice_groups;         ///< more_than_one_slice_group_allowed_flag
  63     int arbitrary_slice_order;         ///< arbitrary_slice_order_allowed_flag
  64     int redundant_slices;              ///< redundant_slices_allowed_flag
  65     int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
  66     int poc_type;                      ///< pic_order_cnt_type
  67     int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
  68     int delta_pic_order_always_zero_flag;
  69     int offset_for_non_ref_pic;
  70     int offset_for_top_to_bottom_field;
  71     int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
  72     int ref_frame_count;               ///< num_ref_frames
  73     int required_frame_num_update_behaviour_flag;
  74     int mb_width;                      ///< frame_width_in_mbs_minus1 + 1
  75     int mb_height;                     ///< frame_height_in_mbs_minus1 + 1
  76     int frame_mbs_only_flag;
  77     int mb_aff;                        ///<mb_adaptive_frame_field_flag
  78     int direct_8x8_inference_flag;
  79     int vui_parameters_present_flag;
  80     int sar_width;
  81     int sar_height;
  82     short offset_for_ref_frame[256]; //FIXME dyn aloc?
  83 }SPS;
  84
  85 /**
  86  * Picture parameter set
  87  */
  88 typedef struct PPS{
  89     int sps_id;
  90     int cabac;                  ///< entropy_coding_mode_flag
  91     int pic_order_present;      ///< pic_order_present_flag
  92     int slice_group_count;      ///< num_slice_groups_minus1 + 1
  93     int mb_slice_group_map_type;
  94     int ref_count[2];           ///< num_ref_idx_l0/1_active_minus1 + 1
  95     int weighted_pred;          ///< weighted_pred_flag
  96     int weighted_bipred_idc;
  97     int init_qp;                ///< pic_init_qp_minus26 + 26
  98     int init_qs;                ///< pic_init_qs_minus26 + 26
  99     int chroma_qp_index_offset;
 100     int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
 101     int constrained_intra_pred; ///< constrained_intra_pred_flag
 102     int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
 103     int crop;                   ///< frame_cropping_flag
 104     int crop_left;              ///< frame_cropping_rect_left_offset
 105     int crop_right;             ///< frame_cropping_rect_right_offset
 106     int crop_top;               ///< frame_cropping_rect_top_offset
 107     int crop_bottom;            ///< frame_cropping_rect_bottom_offset
 108 }PPS;
 109
 110 /**
 111  * Memory management control operation opcode.
 112  */
 113 typedef enum MMCOOpcode{
 114     MMCO_END=0,
 115     MMCO_SHORT2UNUSED,
 116     MMCO_LONG2UNUSED,
 117     MMCO_SHORT2LONG,
 118     MMCO_SET_MAX_LONG,
 119     MMCO_RESET,
 120     MMCO_LONG,
 121 } MMCOOpcode;
 122
 123 /**
 124  * Memory management control operation.
 125  */
 126 typedef struct MMCO{
 127     MMCOOpcode opcode;
 128     int short_frame_num;
 129     int long_index;
 130 } MMCO;
 131
 132 /**
 133  * H264Context
 134  */
 135 typedef struct H264Context{
 136     MpegEncContext s;
 137     int nal_ref_idc;
 138     int nal_unit_type;
 139 #define NAL_SLICE               1
 140 #define NAL_DPA                 2
 141 #define NAL_DPB                 3
 142 #define NAL_DPC                 4
 143 #define NAL_IDR_SLICE           5
 144 #define NAL_SEI                 6
 145 #define NAL_SPS                 7
 146 #define NAL_PPS                 8
 147 #define NAL_PICTURE_DELIMITER   9
 148 #define NAL_FILTER_DATA         10
 149     uint8_t *rbsp_buffer;
 150     int rbsp_buffer_size;
 151
 152     int chroma_qp; //QPc
 153
 154     int prev_mb_skiped; //FIXME remove (IMHO not used)
 155
 156     //prediction stuff
 157     int chroma_pred_mode;
 158     int intra16x16_pred_mode;
 159
 160     int8_t intra4x4_pred_mode_cache[5*8];
 161     int8_t (*intra4x4_pred_mode)[8];
 162     void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
 163     void (*pred8x8  [4+3])(uint8_t *src, int stride);
 164     void (*pred16x16[4+3])(uint8_t *src, int stride);
 165     unsigned int topleft_samples_available;
 166     unsigned int top_samples_available;
 167     unsigned int topright_samples_available;
 168     unsigned int left_samples_available;
 169
 170     /**
 171      * non zero coeff count cache.
 172      * is 64 if not available.
 173      */
 174     uint8_t non_zero_count_cache[6*8];
 175     uint8_t (*non_zero_count)[16];
 176
 177     /**
 178      * Motion vector cache.
 179      */
 180     int16_t mv_cache[2][5*8][2];
 181     int8_t ref_cache[2][5*8];
 182 #define LIST_NOT_USED -1 //FIXME rename?
 183 #define PART_NOT_AVAILABLE -2
 184
 185     /**
 186      * is 1 if the specific list MV&references are set to 0,0,-2.
 187      */
 188     int mv_cache_clean[2];
 189
 190     int block_offset[16+8];
 191     int chroma_subblock_offset[16]; //FIXME remove
 192
 193     uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
 194     uint16_t *mb2b8_xy;
 195     int b_stride;
 196     int b8_stride;
 197
 198     int halfpel_flag;
 199     int thirdpel_flag;
 200
 201     SPS sps_buffer[MAX_SPS_COUNT];
 202     SPS sps; ///< current sps
 203
 204     PPS pps_buffer[MAX_PPS_COUNT];
 205     /**
 206      * current pps
 207      */
 208     PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
 209
 210     int slice_num;
 211     uint8_t *slice_table_base;
 212     uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
 213     int slice_type;
 214     int slice_type_fixed;
 215
 216     //interlacing specific flags
 217     int mb_field_decoding_flag;
 218
 219     int sub_mb_type[4];
 220
 221     //POC stuff
 222     int poc_lsb;
 223     int poc_msb;
 224     int delta_poc_bottom;
 225     int delta_poc[2];
 226     int frame_num;
 227     int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
 228     int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
 229     int frame_num_offset;         ///< for POC type 2
 230     int prev_frame_num_offset;    ///< for POC type 2
 231     int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
 232
 233     /**
 234      * frame_num for frames or 2*frame_num for field pics.
 235      */
 236     int curr_pic_num;
 237
 238     /**
 239      * max_frame_num or 2*max_frame_num for field pics.
 240      */
 241     int max_pic_num;
 242
 243     //Weighted pred stuff
 244     int luma_log2_weight_denom;
 245     int chroma_log2_weight_denom;
 246     int luma_weight[2][16];
 247     int luma_offset[2][16];
 248     int chroma_weight[2][16][2];
 249     int chroma_offset[2][16][2];
 250
 251     //deblock
 252     int disable_deblocking_filter_idc;
 253     int slice_alpha_c0_offset_div2;
 254     int slice_beta_offset_div2;
 255
 256     int redundant_pic_count;
 257
 258     int direct_spatial_mv_pred;
 259
 260     /**
 261      * num_ref_idx_l0/1_active_minus1 + 1
 262      */
 263     int ref_count[2];// FIXME split for AFF
 264     Picture *short_ref[16];
 265     Picture *long_ref[16];
 266     Picture default_ref_list[2][32];
 267     Picture ref_list[2][32]; //FIXME size?
 268     Picture field_ref_list[2][32]; //FIXME size?
 269
 270     /**
 271      * memory management control operations buffer.
 272      */
 273     MMCO mmco[MAX_MMCO_COUNT];
 274     int mmco_index;
 275
 276     int long_ref_count;  ///< number of actual long term references
 277     int short_ref_count; ///< number of actual short term references
 278
 279     //data partitioning
 280     GetBitContext intra_gb;
 281     GetBitContext inter_gb;
 282     GetBitContext *intra_gb_ptr;
 283     GetBitContext *inter_gb_ptr;
 284
 285     DCTELEM mb[16*24] __align8;
 286 }H264Context;
 287
 288 static VLC coeff_token_vlc[4];
 289 static VLC chroma_dc_coeff_token_vlc;
 290
 291 static VLC total_zeros_vlc[15];
 292 static VLC chroma_dc_total_zeros_vlc[3];
 293
 294 static VLC run_vlc[6];
 295 static VLC run7_vlc;
 296
 297 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
 298 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
 299
 300 static inline uint32_t pack16to32(int a, int b){
 301 #ifdef WORDS_BIGENDIAN
 302    return (b&0xFFFF) + (a<<16);
 303 #else
 304    return (a&0xFFFF) + (b<<16);
 305 #endif
 306 }
 307
 308 /**
 309  * fill a rectangle.
 310  * @param h height of the recatangle, should be a constant
 311  * @param w width of the recatangle, should be a constant
 312  * @param size the size of val (1 or 4), should be a constant
 313  */
 314 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
 315     uint8_t *p= (uint8_t*)vp;
 316     assert(size==1 || size==4);
 317
 318     w      *= size;
 319     stride *= size;
 320
 321 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
 322     if(w==2 && h==2){
 323         *(uint16_t*)(p + 0)=
 324         *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
 325     }else if(w==2 && h==4){
 326         *(uint16_t*)(p + 0*stride)=
 327         *(uint16_t*)(p + 1*stride)=
 328         *(uint16_t*)(p + 2*stride)=
 329         *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
 330     }else if(w==4 && h==1){
 331         *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
 332     }else if(w==4 && h==2){
 333         *(uint32_t*)(p + 0*stride)=
 334         *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
 335     }else if(w==4 && h==4){
 336         *(uint32_t*)(p + 0*stride)=
 337         *(uint32_t*)(p + 1*stride)=
 338         *(uint32_t*)(p + 2*stride)=
 339         *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
 340     }else if(w==8 && h==1){
 341         *(uint32_t*)(p + 0)=
 342         *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
 343     }else if(w==8 && h==2){
 344         *(uint32_t*)(p + 0 + 0*stride)=
 345         *(uint32_t*)(p + 4 + 0*stride)=
 346         *(uint32_t*)(p + 0 + 1*stride)=
 347         *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
 348     }else if(w==8 && h==4){
 349         *(uint64_t*)(p + 0*stride)=
 350         *(uint64_t*)(p + 1*stride)=
 351         *(uint64_t*)(p + 2*stride)=
 352         *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
 353     }else if(w==16 && h==2){
 354         *(uint64_t*)(p + 0+0*stride)=
 355         *(uint64_t*)(p + 8+0*stride)=
 356         *(uint64_t*)(p + 0+1*stride)=
 357         *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
 358     }else if(w==16 && h==4){
 359         *(uint64_t*)(p + 0+0*stride)=
 360         *(uint64_t*)(p + 8+0*stride)=
 361         *(uint64_t*)(p + 0+1*stride)=
 362         *(uint64_t*)(p + 8+1*stride)=
 363         *(uint64_t*)(p + 0+2*stride)=
 364         *(uint64_t*)(p + 8+2*stride)=
 365         *(uint64_t*)(p + 0+3*stride)=
 366         *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
 367     }else
 368         assert(0);
 369 }
 370
 371 static inline void fill_caches(H264Context *h, int mb_type){
 372     MpegEncContext * const s = &h->s;
 373     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
 374     int topleft_xy, top_xy, topright_xy, left_xy[2];
 375     int topleft_type, top_type, topright_type, left_type[2];
 376     int left_block[4];
 377     int i;
 378
 379     //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
 380
 381     if(h->sps.mb_aff){
 382     //FIXME
 383     }else{
 384         topleft_xy = mb_xy-1 - s->mb_stride;
 385         top_xy     = mb_xy   - s->mb_stride;
 386         topright_xy= mb_xy+1 - s->mb_stride;
 387         left_xy[0]   = mb_xy-1;
 388         left_xy[1]   = mb_xy-1;
 389         left_block[0]= 0;
 390         left_block[1]= 1;
 391         left_block[2]= 2;
 392         left_block[3]= 3;
 393     }
 394
 395     topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
 396     top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
 397     topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
 398     left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
 399     left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
 400
 401     if(IS_INTRA(mb_type)){
 402         h->topleft_samples_available=
 403         h->top_samples_available=
 404         h->left_samples_available= 0xFFFF;
 405         h->topright_samples_available= 0xEEEA;
 406
 407         if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
 408             h->topleft_samples_available= 0xB3FF;
 409             h->top_samples_available= 0x33FF;
 410             h->topright_samples_available= 0x26EA;
 411         }
 412         for(i=0; i<2; i++){
 413             if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
 414                 h->topleft_samples_available&= 0xDF5F;
 415                 h->left_samples_available&= 0x5F5F;
 416             }
 417         }
 418
 419         if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
 420             h->topleft_samples_available&= 0x7FFF;
 421
 422         if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
 423             h->topright_samples_available&= 0xFBFF;
 424
 425         if(IS_INTRA4x4(mb_type)){
 426             if(IS_INTRA4x4(top_type)){
 427                 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
 428                 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
 429                 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
 430                 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
 431             }else{
 432                 int pred;
 433                 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
 434                     pred= 2;
 435                 else{
 436                     pred= -1;
 437                 }
 438                 h->intra4x4_pred_mode_cache[4+8*0]=
 439                 h->intra4x4_pred_mode_cache[5+8*0]=
 440                 h->intra4x4_pred_mode_cache[6+8*0]=
 441                 h->intra4x4_pred_mode_cache[7+8*0]= pred;
 442             }
 443             for(i=0; i<2; i++){
 444                 if(IS_INTRA4x4(left_type[i])){
 445                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
 446                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
 447                 }else{
 448                     int pred;
 449                     if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
 450                         pred= 2;
 451                     else{
 452                         pred= -1;
 453                     }
 454                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
 455                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
 456                 }
 457             }
 458         }
 459     }
 460
 461
 462 /*
 463 0 . T T. T T T T
 464 1 L . .L . . . .
 465 2 L . .L . . . .
 466 3 . T TL . . . .
 467 4 L . .L . . . .
 468 5 L . .. . . . .
 469 */
 470 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
 471     if(top_type){
 472         h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
 473         h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
 474         h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
 475         h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
 476
 477         h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
 478         h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
 479
 480         h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
 481         h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
 482     }else{
 483         h->non_zero_count_cache[4+8*0]=
 484         h->non_zero_count_cache[5+8*0]=
 485         h->non_zero_count_cache[6+8*0]=
 486         h->non_zero_count_cache[7+8*0]=
 487
 488         h->non_zero_count_cache[1+8*0]=
 489         h->non_zero_count_cache[2+8*0]=
 490
 491         h->non_zero_count_cache[1+8*3]=
 492         h->non_zero_count_cache[2+8*3]= 64;
 493     }
 494
 495     if(left_type[0]){
 496         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
 497         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
 498         h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
 499         h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
 500     }else{
 501         h->non_zero_count_cache[3+8*1]=
 502         h->non_zero_count_cache[3+8*2]=
 503         h->non_zero_count_cache[0+8*1]=
 504         h->non_zero_count_cache[0+8*4]= 64;
 505     }
 506
 507     if(left_type[1]){
 508         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
 509         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
 510         h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
 511         h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
 512     }else{
 513         h->non_zero_count_cache[3+8*3]=
 514         h->non_zero_count_cache[3+8*4]=
 515         h->non_zero_count_cache[0+8*2]=
 516         h->non_zero_count_cache[0+8*5]= 64;
 517     }
 518
 519 #if 1
 520     if(IS_INTER(mb_type)){
 521         int list;
 522         for(list=0; list<2; list++){
 523             if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
 524                 /*if(!h->mv_cache_clean[list]){
 525                     memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
 526                     memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
 527                     h->mv_cache_clean[list]= 1;
 528                 }*/
 529                 continue; //FIXME direct mode ...
 530             }
 531             h->mv_cache_clean[list]= 0;
 532
 533             if(IS_INTER(topleft_type)){
 534                 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
 535                 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
 536                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
 537                 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
 538             }else{
 539                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
 540                 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
 541             }
 542
 543             if(IS_INTER(top_type)){
 544                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
 545                 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
 546                 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
 547                 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
 548                 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
 549                 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
 550                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
 551                 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
 552                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
 553                 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
 554             }else{
 555                 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
 556                 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
 557                 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
 558                 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
 559                 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
 560             }
 561
 562             if(IS_INTER(topright_type)){
 563                 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
 564                 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
 565                 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
 566                 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
 567             }else{
 568                 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
 569                 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
 570             }
 571
 572             //FIXME unify cleanup or sth
 573             if(IS_INTER(left_type[0])){
 574                 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
 575                 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
 576                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]];
 577                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1]];
 578                 h->ref_cache[list][scan8[0] - 1 + 0*8]=
 579                 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
 580             }else{
 581                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
 582                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
 583                 h->ref_cache[list][scan8[0] - 1 + 0*8]=
 584                 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
 585             }
 586
 587             if(IS_INTER(left_type[1])){
 588                 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
 589                 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
 590                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[2]];
 591                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[3]];
 592                 h->ref_cache[list][scan8[0] - 1 + 2*8]=
 593                 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
 594             }else{
 595                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
 596                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
 597                 h->ref_cache[list][scan8[0] - 1 + 2*8]=
 598                 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
 599             }
 600
 601             h->ref_cache[list][scan8[5 ]+1] =
 602             h->ref_cache[list][scan8[7 ]+1] =
 603             h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewher else)
 604             h->ref_cache[list][scan8[4 ]] =
 605             h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
 606             *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
 607             *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
 608             *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
 609             *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
 610             *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
 611         }
 612 //FIXME
 613
 614     }
 615 #endif
 616 }
 617
 618 static inline void write_back_intra_pred_mode(H264Context *h){
 619     MpegEncContext * const s = &h->s;
 620     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
 621
 622     h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
 623     h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
 624     h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
 625     h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
 626     h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
 627     h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
 628     h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
 629 }
 630
 631 /**
 632  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 633  */
 634 static inline int check_intra4x4_pred_mode(H264Context *h){
 635     MpegEncContext * const s = &h->s;
 636     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
 637     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
 638     int i;
 639
 640     if(!(h->top_samples_available&0x8000)){
 641         for(i=0; i<4; i++){
 642             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
 643             if(status<0){
 644                 fprintf(stderr, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
 645                 return -1;
 646             } else if(status){
 647                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
 648             }
 649         }
 650     }
 651
 652     if(!(h->left_samples_available&0x8000)){
 653         for(i=0; i<4; i++){
 654             int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
 655             if(status<0){
 656                 fprintf(stderr, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
 657                 return -1;
 658             } else if(status){
 659                 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
 660             }
 661         }
 662     }
 663
 664     return 0;
 665 } //FIXME cleanup like next
 666
 667 /**
 668  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 669  */
 670 static inline int check_intra_pred_mode(H264Context *h, int mode){
 671     MpegEncContext * const s = &h->s;
 672     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
 673     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
 674
 675     if(!(h->top_samples_available&0x8000)){
 676         mode= top[ mode ];
 677         if(mode<0){
 678             fprintf(stderr, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 679             return -1;
 680         }
 681     }
 682
 683     if(!(h->left_samples_available&0x8000)){
 684         mode= left[ mode ];
 685         if(mode<0){
 686             fprintf(stderr, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 687             return -1;
 688         }
 689     }
 690
 691     return mode;
 692 }
 693
 694 /**
 695  * gets the predicted intra4x4 prediction mode.
 696  */
 697 static inline int pred_intra_mode(H264Context *h, int n){
 698     const int index8= scan8[n];
 699     const int left= h->intra4x4_pred_mode_cache[index8 - 1];
 700     const int top = h->intra4x4_pred_mode_cache[index8 - 8];
 701     const int min= FFMIN(left, top);
 702
 703     tprintf("mode:%d %d min:%d\n", left ,top, min);
 704
 705     if(min<0) return DC_PRED;
 706     else      return min;
 707 }
 708
 709 static inline void write_back_non_zero_count(H264Context *h){
 710     MpegEncContext * const s = &h->s;
 711     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
 712
 713     h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
 714     h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
 715     h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
 716     h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
 717     h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
 718     h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
 719     h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
 720
 721     h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
 722     h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
 723     h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
 724
 725     h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
 726     h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
 727     h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
 728 }
 729
 730 /**
 731  * gets the predicted number of non zero coefficients.
 732  * @param n block index
 733  */
 734 static inline int pred_non_zero_count(H264Context *h, int n){
 735     const int index8= scan8[n];
 736     const int left= h->non_zero_count_cache[index8 - 1];
 737     const int top = h->non_zero_count_cache[index8 - 8];
 738     int i= left + top;
 739
 740     if(i<64) i= (i+1)>>1;
 741
 742     tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
 743
 744     return i&31;
 745 }
 746
 747 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
 748     const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
 749
 750     if(topright_ref != PART_NOT_AVAILABLE){
 751         *C= h->mv_cache[list][ i - 8 + part_width ];
 752         return topright_ref;
 753     }else{
 754         tprintf("topright MV not available\n");
 755
 756         *C= h->mv_cache[list][ i - 8 - 1 ];
 757         return h->ref_cache[list][ i - 8 - 1 ];
 758     }
 759 }
 760
 761 /**
 762  * gets the predicted MV.
 763  * @param n the block index
 764  * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
 765  * @param mx the x component of the predicted motion vector
 766  * @param my the y component of the predicted motion vector
 767  */
 768 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
 769     const int index8= scan8[n];
 770     const int top_ref=      h->ref_cache[list][ index8 - 8 ];
 771     const int left_ref=     h->ref_cache[list][ index8 - 1 ];
 772     const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
 773     const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
 774     const int16_t * C;
 775     int diagonal_ref, match_count;
 776
 777     assert(part_width==1 || part_width==2 || part_width==4);
 778
 779 /* mv_cache
 780   B . . A T T T T
 781   U . . L . . , .
 782   U . . L . . . .
 783   U . . L . . , .
 784   . . . L . . . .
 785 */
 786
 787     diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
 788     match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
 789
 790     if(match_count > 1){ //most common
 791         *mx= mid_pred(A[0], B[0], C[0]);
 792         *my= mid_pred(A[1], B[1], C[1]);
 793     }else if(match_count==1){
 794         if(left_ref==ref){
 795             *mx= A[0];
 796             *my= A[1];
 797         }else if(top_ref==ref){
 798             *mx= B[0];
 799             *my= B[1];
 800         }else{
 801             *mx= C[0];
 802             *my= C[1];
 803         }
 804     }else{
 805         if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
 806             *mx= A[0];
 807             *my= A[1];
 808         }else{
 809             *mx= mid_pred(A[0], B[0], C[0]);
 810             *my= mid_pred(A[1], B[1], C[1]);
 811         }
 812     }
 813
 814     tprintf("pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1],                    diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);
 815 }
 816
 817 /**
 818  * gets the directionally predicted 16x8 MV.
 819  * @param n the block index
 820  * @param mx the x component of the predicted motion vector
 821  * @param my the y component of the predicted motion vector
 822  */
 823 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
 824     if(n==0){
 825         const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
 826         const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
 827
 828         tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
 829
 830         if(top_ref == ref){
 831             *mx= B[0];
 832             *my= B[1];
 833             return;
 834         }
 835     }else{
 836         const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
 837         const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
 838
 839         tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
 840
 841         if(left_ref == ref){
 842             *mx= A[0];
 843             *my= A[1];
 844             return;
 845         }
 846     }
 847
 848     //RARE
 849     pred_motion(h, n, 4, list, ref, mx, my);
 850 }
 851
 852 /**
 853  * gets the directionally predicted 8x16 MV.
 854  * @param n the block index
 855  * @param mx the x component of the predicted motion vector
 856  * @param my the y component of the predicted motion vector
 857  */
 858 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
 859     if(n==0){
 860         const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
 861         const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
 862
 863         tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
 864
 865         if(left_ref == ref){
 866             *mx= A[0];
 867             *my= A[1];
 868             return;
 869         }
 870     }else{
 871         const int16_t * C;
 872         int diagonal_ref;
 873
 874         diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
 875
 876         tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
 877
 878         if(diagonal_ref == ref){
 879             *mx= C[0];
 880             *my= C[1];
 881             return;
 882         }
 883     }
 884
 885     //RARE
 886     pred_motion(h, n, 2, list, ref, mx, my);
 887 }
 888
 889 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
 890     const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
 891     const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
 892
 893     tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
 894
 895     if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
 896        || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
 897        || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
 898
 899         *mx = *my = 0;
 900         return;
 901     }
 902
 903     pred_motion(h, 0, 4, 0, 0, mx, my);
 904
 905     return;
 906 }
 907
 908 static inline void write_back_motion(H264Context *h, int mb_type){
 909     MpegEncContext * const s = &h->s;
 910     const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
 911     const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
 912     int list;
 913
 914     for(list=0; list<2; list++){
 915         int y;
 916         if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
 917             if(1){ //FIXME skip or never read if mb_type doesnt use it
 918                 for(y=0; y<4; y++){
 919                     *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
 920                     *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
 921                 }
 922                 for(y=0; y<2; y++){
 923                     *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
 924                 }
 925             }
 926             continue; //FIXME direct mode ...
 927         }
 928
 929         for(y=0; y<4; y++){
 930             *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y];
 931             *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];
 932         }
 933         for(y=0; y<2; y++){
 934             s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
 935             s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
 936         }
 937     }
 938 }
 939
 940 /**
 941  * Decodes a network abstraction layer unit.
 942  * @param consumed is the number of bytes used as input
 943  * @param length is the length of the array
 944  * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
 945  * @returns decoded bytes, might be src+1 if no escapes
 946  */
 947 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
 948     int i, si, di;
 949     uint8_t *dst;
 950
 951 //    src[0]&0x80;              //forbidden bit
 952     h->nal_ref_idc= src[0]>>5;
 953     h->nal_unit_type= src[0]&0x1F;
 954
 955     src++; length--;
 956 #if 0
 957     for(i=0; i<length; i++)
 958         printf("%2X ", src[i]);
 959 #endif
 960     for(i=0; i+1<length; i+=2){
 961         if(src[i]) continue;
 962         if(i>0 && src[i-1]==0) i--;
 963         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
 964             if(src[i+2]!=3){
 965                 /* startcode, so we must be past the end */
 966                 length=i;
 967             }
 968             break;
 969         }
 970     }
 971
 972     if(i>=length-1){ //no escaped 0
 973         *dst_length= length;
 974         *consumed= length+1; //+1 for the header
 975         return src;
 976     }
 977
 978     h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
 979     dst= h->rbsp_buffer;
 980
 981 //printf("deoding esc\n");
 982     si=di=0;
 983     while(si<length){
 984         //remove escapes (very rare 1:2^22)
 985         if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
 986             if(src[si+2]==3){ //escape
 987                 dst[di++]= 0;
 988                 dst[di++]= 0;
 989                 si+=3;
 990             }else //next start code
 991                 break;
 992         }
 993
 994         dst[di++]= src[si++];
 995     }
 996
 997     *dst_length= di;
 998     *consumed= si + 1;//+1 for the header
 999 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1000     return dst;
1001 }
1002
1003 /**
1004  * @param src the data which should be escaped
1005  * @param dst the target buffer, dst+1 == src is allowed as a special case
1006  * @param length the length of the src data
1007  * @param dst_length the length of the dst array
1008  * @returns length of escaped data in bytes or -1 if an error occured
1009  */
1010 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1011     int i, escape_count, si, di;
1012     uint8_t *temp;
1013
1014     assert(length>=0);
1015     assert(dst_length>0);
1016
1017     dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1018
1019     if(length==0) return 1;
1020
1021     escape_count= 0;
1022     for(i=0; i<length; i+=2){
1023         if(src[i]) continue;
1024         if(i>0 && src[i-1]==0)
1025             i--;
1026         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1027             escape_count++;
1028             i+=2;
1029         }
1030     }
1031
1032     if(escape_count==0){
1033         if(dst+1 != src)
1034             memcpy(dst+1, src, length);
1035         return length + 1;
1036     }
1037
1038     if(length + escape_count + 1> dst_length)
1039         return -1;
1040
1041     //this should be damn rare (hopefully)
1042
1043     h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1044     temp= h->rbsp_buffer;
1045 //printf("encoding esc\n");
1046
1047     si= 0;
1048     di= 0;
1049     while(si < length){
1050         if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1051             temp[di++]= 0; si++;
1052             temp[di++]= 0; si++;
1053             temp[di++]= 3;
1054             temp[di++]= src[si++];
1055         }
1056         else
1057             temp[di++]= src[si++];
1058     }
1059     memcpy(dst+1, temp, length+escape_count);
1060
1061     assert(di == length+escape_count);
1062
1063     return di + 1;
1064 }
1065
1066 /**
1067  * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1068  */
1069 static void encode_rbsp_trailing(PutBitContext *pb){
1070     int length;
1071     put_bits(pb, 1, 1);
1072     length= (-get_bit_count(pb))&7;
1073     if(length) put_bits(pb, length, 0);
1074 }
1075
1076 /**
1077  * identifies the exact end of the bitstream
1078  * @return the length of the trailing, or 0 if damaged
1079  */
1080 static int decode_rbsp_trailing(uint8_t *src){
1081     int v= *src;
1082     int r;
1083
1084     tprintf("rbsp trailing %X\n", v);
1085
1086     for(r=1; r<9; r++){
1087         if(v&1) return r;
1088         v>>=1;
1089     }
1090     return 0;
1091 }
1092
1093 /**
1094  * idct tranforms the 16 dc values and dequantize them.
1095  * @param qp quantization parameter
1096  */
1097 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1098     const int qmul= dequant_coeff[qp][0];
1099 #define stride 16
1100     int i;
1101     int temp[16]; //FIXME check if this is a good idea
1102     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1103     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1104
1105 //memset(block, 64, 2*256);
1106 //return;
1107     for(i=0; i<4; i++){
1108         const int offset= y_offset[i];
1109         const int z0= block[offset+stride*0] + block[offset+stride*4];
1110         const int z1= block[offset+stride*0] - block[offset+stride*4];
1111         const int z2= block[offset+stride*1] - block[offset+stride*5];
1112         const int z3= block[offset+stride*1] + block[offset+stride*5];
1113
1114         temp[4*i+0]= z0+z3;
1115         temp[4*i+1]= z1+z2;
1116         temp[4*i+2]= z1-z2;
1117         temp[4*i+3]= z0-z3;
1118     }
1119
1120     for(i=0; i<4; i++){
1121         const int offset= x_offset[i];
1122         const int z0= temp[4*0+i] + temp[4*2+i];
1123         const int z1= temp[4*0+i] - temp[4*2+i];
1124         const int z2= temp[4*1+i] - temp[4*3+i];
1125         const int z3= temp[4*1+i] + temp[4*3+i];
1126
1127         block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1128         block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1129         block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1130         block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1131     }
1132 }
1133
1134 /**
1135  * dct tranforms the 16 dc values.
1136  * @param qp quantization parameter ??? FIXME
1137  */
1138 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1139 //    const int qmul= dequant_coeff[qp][0];
1140     int i;
1141     int temp[16]; //FIXME check if this is a good idea
1142     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1143     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1144
1145     for(i=0; i<4; i++){
1146         const int offset= y_offset[i];
1147         const int z0= block[offset+stride*0] + block[offset+stride*4];
1148         const int z1= block[offset+stride*0] - block[offset+stride*4];
1149         const int z2= block[offset+stride*1] - block[offset+stride*5];
1150         const int z3= block[offset+stride*1] + block[offset+stride*5];
1151
1152         temp[4*i+0]= z0+z3;
1153         temp[4*i+1]= z1+z2;
1154         temp[4*i+2]= z1-z2;
1155         temp[4*i+3]= z0-z3;
1156     }
1157
1158     for(i=0; i<4; i++){
1159         const int offset= x_offset[i];
1160         const int z0= temp[4*0+i] + temp[4*2+i];
1161         const int z1= temp[4*0+i] - temp[4*2+i];
1162         const int z2= temp[4*1+i] - temp[4*3+i];
1163         const int z3= temp[4*1+i] + temp[4*3+i];
1164
1165         block[stride*0 +offset]= (z0 + z3)>>1;
1166         block[stride*2 +offset]= (z1 + z2)>>1;
1167         block[stride*8 +offset]= (z1 - z2)>>1;
1168         block[stride*10+offset]= (z0 - z3)>>1;
1169     }
1170 }
1171 #undef xStride
1172 #undef stride
1173
1174 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1175     const int qmul= dequant_coeff[qp][0];
1176     const int stride= 16*2;
1177     const int xStride= 16;
1178     int a,b,c,d,e;
1179
1180     a= block[stride*0 + xStride*0];
1181     b= block[stride*0 + xStride*1];
1182     c= block[stride*1 + xStride*0];
1183     d= block[stride*1 + xStride*1];
1184
1185     e= a-b;
1186     a= a+b;
1187     b= c-d;
1188     c= c+d;
1189
1190     block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1191     block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1192     block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1193     block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1194 }
1195
1196 static void chroma_dc_dct_c(DCTELEM *block){
1197     const int stride= 16*2;
1198     const int xStride= 16;
1199     int a,b,c,d,e;
1200
1201     a= block[stride*0 + xStride*0];
1202     b= block[stride*0 + xStride*1];
1203     c= block[stride*1 + xStride*0];
1204     d= block[stride*1 + xStride*1];
1205
1206     e= a-b;
1207     a= a+b;
1208     b= c-d;
1209     c= c+d;
1210
1211     block[stride*0 + xStride*0]= (a+c);
1212     block[stride*0 + xStride*1]= (e+b);
1213     block[stride*1 + xStride*0]= (a-c);
1214     block[stride*1 + xStride*1]= (e-b);
1215 }
1216
1217 /**
1218  * gets the chroma qp.
1219  */
1220 static inline int get_chroma_qp(H264Context *h, int qscale){
1221
1222     return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1223 }
1224
1225
1226 /**
1227  *
1228  */
1229 static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1230     int i;
1231     uint8_t *cm = cropTbl + MAX_NEG_CROP;
1232
1233     block[0] += 32;
1234 #if 1
1235     for(i=0; i<4; i++){
1236         const int z0=  block[i + 4*0]     +  block[i + 4*2];
1237         const int z1=  block[i + 4*0]     -  block[i + 4*2];
1238         const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
1239         const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
1240
1241         block[i + 4*0]= z0 + z3;
1242         block[i + 4*1]= z1 + z2;
1243         block[i + 4*2]= z1 - z2;
1244         block[i + 4*3]= z0 - z3;
1245     }
1246
1247     for(i=0; i<4; i++){
1248         const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
1249         const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
1250         const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
1251         const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
1252
1253         dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ];
1254         dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ];
1255         dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ];
1256         dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ];
1257     }
1258 #else
1259     for(i=0; i<4; i++){
1260         const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
1261         const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
1262         const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
1263         const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
1264
1265         block[0 + 4*i]= z0 + z3;
1266         block[1 + 4*i]= z1 + z2;
1267         block[2 + 4*i]= z1 - z2;
1268         block[3 + 4*i]= z0 - z3;
1269     }
1270
1271     for(i=0; i<4; i++){
1272         const int z0=  block[i + 4*0]     +  block[i + 4*2];
1273         const int z1=  block[i + 4*0]     -  block[i + 4*2];
1274         const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
1275         const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
1276
1277         dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1278         dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1279         dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1280         dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1281     }
1282 #endif
1283 }
1284
1285 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1286     int i;
1287     //FIXME try int temp instead of block
1288
1289     for(i=0; i<4; i++){
1290         const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1291         const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1292         const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1293         const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1294         const int z0= d0 + d3;
1295         const int z3= d0 - d3;
1296         const int z1= d1 + d2;
1297         const int z2= d1 - d2;
1298
1299         block[0 + 4*i]=   z0 +   z1;
1300         block[1 + 4*i]= 2*z3 +   z2;
1301         block[2 + 4*i]=   z0 -   z1;
1302         block[3 + 4*i]=   z3 - 2*z2;
1303     }
1304
1305     for(i=0; i<4; i++){
1306         const int z0= block[0*4 + i] + block[3*4 + i];
1307         const int z3= block[0*4 + i] - block[3*4 + i];
1308         const int z1= block[1*4 + i] + block[2*4 + i];
1309         const int z2= block[1*4 + i] - block[2*4 + i];
1310
1311         block[0*4 + i]=   z0 +   z1;
1312         block[1*4 + i]= 2*z3 +   z2;
1313         block[2*4 + i]=   z0 -   z1;
1314         block[3*4 + i]=   z3 - 2*z2;
1315     }
1316 }
1317
1318 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1319 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1320 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1321     int i;
1322     const int * const quant_table= quant_coeff[qscale];
1323     const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1324     const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1325     const unsigned int threshold2= (threshold1<<1);
1326     int last_non_zero;
1327
1328     if(seperate_dc){
1329         if(qscale<=18){
1330             //avoid overflows
1331             const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1332             const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1333             const unsigned int dc_threshold2= (dc_threshold1<<1);
1334
1335             int level= block[0]*quant_coeff[qscale+18][0];
1336             if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1337                 if(level>0){
1338                     level= (dc_bias + level)>>(QUANT_SHIFT-2);
1339                     block[0]= level;
1340                 }else{
1341                     level= (dc_bias - level)>>(QUANT_SHIFT-2);
1342                     block[0]= -level;
1343                 }
1344 //                last_non_zero = i;
1345             }else{
1346                 block[0]=0;
1347             }
1348         }else{
1349             const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1350             const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1351             const unsigned int dc_threshold2= (dc_threshold1<<1);
1352
1353             int level= block[0]*quant_table[0];
1354             if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1355                 if(level>0){
1356                     level= (dc_bias + level)>>(QUANT_SHIFT+1);
1357                     block[0]= level;
1358                 }else{
1359                     level= (dc_bias - level)>>(QUANT_SHIFT+1);
1360                     block[0]= -level;
1361                 }
1362 //                last_non_zero = i;
1363             }else{
1364                 block[0]=0;
1365             }
1366         }
1367         last_non_zero= 0;
1368         i=1;
1369     }else{
1370         last_non_zero= -1;
1371         i=0;
1372     }
1373
1374     for(; i<16; i++){
1375         const int j= scantable[i];
1376         int level= block[j]*quant_table[j];
1377
1378 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1379 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1380         if(((unsigned)(level+threshold1))>threshold2){
1381             if(level>0){
1382                 level= (bias + level)>>QUANT_SHIFT;
1383                 block[j]= level;
1384             }else{
1385                 level= (bias - level)>>QUANT_SHIFT;
1386                 block[j]= -level;
1387             }
1388             last_non_zero = i;
1389         }else{
1390             block[j]=0;
1391         }
1392     }
1393
1394     return last_non_zero;
1395 }
1396
1397 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1398     const uint32_t a= ((uint32_t*)(src-stride))[0];
1399     ((uint32_t*)(src+0*stride))[0]= a;
1400     ((uint32_t*)(src+1*stride))[0]= a;
1401     ((uint32_t*)(src+2*stride))[0]= a;
1402     ((uint32_t*)(src+3*stride))[0]= a;
1403 }
1404
1405 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1406     ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1407     ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1408     ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1409     ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1410 }
1411
1412 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1413     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1414                    + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1415
1416     ((uint32_t*)(src+0*stride))[0]=
1417     ((uint32_t*)(src+1*stride))[0]=
1418     ((uint32_t*)(src+2*stride))[0]=
1419     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1420 }
1421
1422 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1423     const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1424
1425     ((uint32_t*)(src+0*stride))[0]=
1426     ((uint32_t*)(src+1*stride))[0]=
1427     ((uint32_t*)(src+2*stride))[0]=
1428     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1429 }
1430
1431 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1432     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1433
1434     ((uint32_t*)(src+0*stride))[0]=
1435     ((uint32_t*)(src+1*stride))[0]=
1436     ((uint32_t*)(src+2*stride))[0]=
1437     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1438 }
1439
1440 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1441     ((uint32_t*)(src+0*stride))[0]=
1442     ((uint32_t*)(src+1*stride))[0]=
1443     ((uint32_t*)(src+2*stride))[0]=
1444     ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1445 }
1446
1447
1448 #define LOAD_TOP_RIGHT_EDGE\
1449     const int t4= topright[0];\
1450     const int t5= topright[1];\
1451     const int t6= topright[2];\
1452     const int t7= topright[3];\
1453
1454 #define LOAD_LEFT_EDGE\
1455     const int l0= src[-1+0*stride];\
1456     const int l1= src[-1+1*stride];\
1457     const int l2= src[-1+2*stride];\
1458     const int l3= src[-1+3*stride];\
1459
1460 #define LOAD_TOP_EDGE\
1461     const int t0= src[ 0-1*stride];\
1462     const int t1= src[ 1-1*stride];\
1463     const int t2= src[ 2-1*stride];\
1464     const int t3= src[ 3-1*stride];\
1465
1466 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1467     const int lt= src[-1-1*stride];
1468     LOAD_TOP_EDGE
1469     LOAD_LEFT_EDGE
1470
1471     src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1472     src[0+2*stride]=
1473     src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1474     src[0+1*stride]=
1475     src[1+2*stride]=
1476     src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1477     src[0+0*stride]=
1478     src[1+1*stride]=
1479     src[2+2*stride]=
1480     src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1481     src[1+0*stride]=
1482     src[2+1*stride]=
1483     src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1484     src[2+0*stride]=
1485     src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1486     src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1487 }
1488
1489 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1490     LOAD_TOP_EDGE
1491     LOAD_TOP_RIGHT_EDGE
1492 //    LOAD_LEFT_EDGE
1493
1494     src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1495     src[1+0*stride]=
1496     src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1497     src[2+0*stride]=
1498     src[1+1*stride]=
1499     src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1500     src[3+0*stride]=
1501     src[2+1*stride]=
1502     src[1+2*stride]=
1503     src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1504     src[3+1*stride]=
1505     src[2+2*stride]=
1506     src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1507     src[3+2*stride]=
1508     src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1509     src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1510 }
1511
1512 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1513     const int lt= src[-1-1*stride];
1514     LOAD_TOP_EDGE
1515     LOAD_LEFT_EDGE
1516     const __attribute__((unused)) int unu= l3;
1517
1518     src[0+0*stride]=
1519     src[1+2*stride]=(lt + t0 + 1)>>1;
1520     src[1+0*stride]=
1521     src[2+2*stride]=(t0 + t1 + 1)>>1;
1522     src[2+0*stride]=
1523     src[3+2*stride]=(t1 + t2 + 1)>>1;
1524     src[3+0*stride]=(t2 + t3 + 1)>>1;
1525     src[0+1*stride]=
1526     src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1527     src[1+1*stride]=
1528     src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1529     src[2+1*stride]=
1530     src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1531     src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1532     src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1533     src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1534 }
1535
1536 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1537     LOAD_TOP_EDGE
1538     LOAD_TOP_RIGHT_EDGE
1539     const __attribute__((unused)) int unu= t7;
1540
1541     src[0+0*stride]=(t0 + t1 + 1)>>1;
1542     src[1+0*stride]=
1543     src[0+2*stride]=(t1 + t2 + 1)>>1;
1544     src[2+0*stride]=
1545     src[1+2*stride]=(t2 + t3 + 1)>>1;
1546     src[3+0*stride]=
1547     src[2+2*stride]=(t3 + t4+ 1)>>1;
1548     src[3+2*stride]=(t4 + t5+ 1)>>1;
1549     src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1550     src[1+1*stride]=
1551     src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1552     src[2+1*stride]=
1553     src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1554     src[3+1*stride]=
1555     src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1556     src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1557 }
1558
1559 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1560     LOAD_LEFT_EDGE
1561
1562     src[0+0*stride]=(l0 + l1 + 1)>>1;
1563     src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1564     src[2+0*stride]=
1565     src[0+1*stride]=(l1 + l2 + 1)>>1;
1566     src[3+0*stride]=
1567     src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1568     src[2+1*stride]=
1569     src[0+2*stride]=(l2 + l3 + 1)>>1;
1570     src[3+1*stride]=
1571     src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1572     src[3+2*stride]=
1573     src[1+3*stride]=
1574     src[0+3*stride]=
1575     src[2+2*stride]=
1576     src[2+3*stride]=
1577     src[3+3*stride]=l3;
1578 }
1579
1580 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1581     const int lt= src[-1-1*stride];
1582     LOAD_TOP_EDGE
1583     LOAD_LEFT_EDGE
1584     const __attribute__((unused)) int unu= t3;
1585
1586     src[0+0*stride]=
1587     src[2+1*stride]=(lt + l0 + 1)>>1;
1588     src[1+0*stride]=
1589     src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1590     src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1591     src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1592     src[0+1*stride]=
1593     src[2+2*stride]=(l0 + l1 + 1)>>1;
1594     src[1+1*stride]=
1595     src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1596     src[0+2*stride]=
1597     src[2+3*stride]=(l1 + l2+ 1)>>1;
1598     src[1+2*stride]=
1599     src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1600     src[0+3*stride]=(l2 + l3 + 1)>>1;
1601     src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1602 }
1603
1604 static void pred16x16_vertical_c(uint8_t *src, int stride){
1605     int i;
1606     const uint32_t a= ((uint32_t*)(src-stride))[0];
1607     const uint32_t b= ((uint32_t*)(src-stride))[1];
1608     const uint32_t c= ((uint32_t*)(src-stride))[2];
1609     const uint32_t d= ((uint32_t*)(src-stride))[3];
1610
1611     for(i=0; i<16; i++){
1612         ((uint32_t*)(src+i*stride))[0]= a;
1613         ((uint32_t*)(src+i*stride))[1]= b;
1614         ((uint32_t*)(src+i*stride))[2]= c;
1615         ((uint32_t*)(src+i*stride))[3]= d;
1616     }
1617 }
1618
1619 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1620     int i;
1621
1622     for(i=0; i<16; i++){
1623         ((uint32_t*)(src+i*stride))[0]=
1624         ((uint32_t*)(src+i*stride))[1]=
1625         ((uint32_t*)(src+i*stride))[2]=
1626         ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1627     }
1628 }
1629
1630 static void pred16x16_dc_c(uint8_t *src, int stride){
1631     int i, dc=0;
1632
1633     for(i=0;i<16; i++){
1634         dc+= src[-1+i*stride];
1635     }
1636
1637     for(i=0;i<16; i++){
1638         dc+= src[i-stride];
1639     }
1640
1641     dc= 0x01010101*((dc + 16)>>5);
1642
1643     for(i=0; i<16; i++){
1644         ((uint32_t*)(src+i*stride))[0]=
1645         ((uint32_t*)(src+i*stride))[1]=
1646         ((uint32_t*)(src+i*stride))[2]=
1647         ((uint32_t*)(src+i*stride))[3]= dc;
1648     }
1649 }
1650
1651 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1652     int i, dc=0;
1653
1654     for(i=0;i<16; i++){
1655         dc+= src[-1+i*stride];
1656     }
1657
1658     dc= 0x01010101*((dc + 8)>>4);
1659
1660     for(i=0; i<16; i++){
1661         ((uint32_t*)(src+i*stride))[0]=
1662         ((uint32_t*)(src+i*stride))[1]=
1663         ((uint32_t*)(src+i*stride))[2]=
1664         ((uint32_t*)(src+i*stride))[3]= dc;
1665     }
1666 }
1667
1668 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1669     int i, dc=0;
1670
1671     for(i=0;i<16; i++){
1672         dc+= src[i-stride];
1673     }
1674     dc= 0x01010101*((dc + 8)>>4);
1675
1676     for(i=0; i<16; i++){
1677         ((uint32_t*)(src+i*stride))[0]=
1678         ((uint32_t*)(src+i*stride))[1]=
1679         ((uint32_t*)(src+i*stride))[2]=
1680         ((uint32_t*)(src+i*stride))[3]= dc;
1681     }
1682 }
1683
1684 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1685     int i;
1686
1687     for(i=0; i<16; i++){
1688         ((uint32_t*)(src+i*stride))[0]=
1689         ((uint32_t*)(src+i*stride))[1]=
1690         ((uint32_t*)(src+i*stride))[2]=
1691         ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1692     }
1693 }
1694
1695 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1696   int i, j, k;
1697   int a;
1698   uint8_t *cm = cropTbl + MAX_NEG_CROP;
1699   const uint8_t * const src0 = src+7-stride;
1700   const uint8_t *src1 = src+8*stride-1;
1701   const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
1702   int H = src0[1] - src0[-1];
1703   int V = src1[0] - src2[ 0];
1704   for(k=2; k<=8; ++k) {
1705     src1 += stride; src2 -= stride;
1706     H += k*(src0[k] - src0[-k]);
1707     V += k*(src1[0] - src2[ 0]);
1708   }
1709   if(svq3){
1710     H = ( 5*(H/4) ) / 16;
1711     V = ( 5*(V/4) ) / 16;
1712   }else{
1713     H = ( 5*H+32 ) >> 6;
1714     V = ( 5*V+32 ) >> 6;
1715   }
1716
1717   a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1718   for(j=16; j>0; --j) {
1719     int b = a;
1720     a += V;
1721     for(i=-16; i<0; i+=4) {
1722       src[16+i] = cm[ (b    ) >> 5 ];
1723       src[17+i] = cm[ (b+  H) >> 5 ];
1724       src[18+i] = cm[ (b+2*H) >> 5 ];
1725       src[19+i] = cm[ (b+3*H) >> 5 ];
1726       b += 4*H;
1727     }
1728     src += stride;
1729   }
1730 }
1731
1732 static void pred16x16_plane_c(uint8_t *src, int stride){
1733     pred16x16_plane_compat_c(src, stride, 0);
1734 }
1735
1736 static void pred8x8_vertical_c(uint8_t *src, int stride){
1737     int i;
1738     const uint32_t a= ((uint32_t*)(src-stride))[0];
1739     const uint32_t b= ((uint32_t*)(src-stride))[1];
1740
1741     for(i=0; i<8; i++){
1742         ((uint32_t*)(src+i*stride))[0]= a;
1743         ((uint32_t*)(src+i*stride))[1]= b;
1744     }
1745 }
1746
1747 static void pred8x8_horizontal_c(uint8_t *src, int stride){
1748     int i;
1749
1750     for(i=0; i<8; i++){
1751         ((uint32_t*)(src+i*stride))[0]=
1752         ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1753     }
1754 }
1755
1756 static void pred8x8_128_dc_c(uint8_t *src, int stride){
1757     int i;
1758
1759     for(i=0; i<4; i++){
1760         ((uint32_t*)(src+i*stride))[0]=
1761         ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1762     }
1763     for(i=4; i<8; i++){
1764         ((uint32_t*)(src+i*stride))[0]=
1765         ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1766     }
1767 }
1768
1769 static void pred8x8_left_dc_c(uint8_t *src, int stride){
1770     int i;
1771     int dc0, dc2;
1772
1773     dc0=dc2=0;
1774     for(i=0;i<4; i++){
1775         dc0+= src[-1+i*stride];
1776         dc2+= src[-1+(i+4)*stride];
1777     }
1778     dc0= 0x01010101*((dc0 + 2)>>2);
1779     dc2= 0x01010101*((dc2 + 2)>>2);
1780
1781     for(i=0; i<4; i++){
1782         ((uint32_t*)(src+i*stride))[0]=
1783         ((uint32_t*)(src+i*stride))[1]= dc0;
1784     }
1785     for(i=4; i<8; i++){
1786         ((uint32_t*)(src+i*stride))[0]=
1787         ((uint32_t*)(src+i*stride))[1]= dc2;
1788     }
1789 }
1790
1791 static void pred8x8_top_dc_c(uint8_t *src, int stride){
1792     int i;
1793     int dc0, dc1;
1794
1795     dc0=dc1=0;
1796     for(i=0;i<4; i++){
1797         dc0+= src[i-stride];
1798         dc1+= src[4+i-stride];
1799     }
1800     dc0= 0x01010101*((dc0 + 2)>>2);
1801     dc1= 0x01010101*((dc1 + 2)>>2);
1802
1803     for(i=0; i<4; i++){
1804         ((uint32_t*)(src+i*stride))[0]= dc0;
1805         ((uint32_t*)(src+i*stride))[1]= dc1;
1806     }
1807     for(i=4; i<8; i++){
1808         ((uint32_t*)(src+i*stride))[0]= dc0;
1809         ((uint32_t*)(src+i*stride))[1]= dc1;
1810     }
1811 }
1812
1813
1814 static void pred8x8_dc_c(uint8_t *src, int stride){
1815     int i;
1816     int dc0, dc1, dc2, dc3;
1817
1818     dc0=dc1=dc2=0;
1819     for(i=0;i<4; i++){
1820         dc0+= src[-1+i*stride] + src[i-stride];
1821         dc1+= src[4+i-stride];
1822         dc2+= src[-1+(i+4)*stride];
1823     }
1824     dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1825     dc0= 0x01010101*((dc0 + 4)>>3);
1826     dc1= 0x01010101*((dc1 + 2)>>2);
1827     dc2= 0x01010101*((dc2 + 2)>>2);
1828
1829     for(i=0; i<4; i++){
1830         ((uint32_t*)(src+i*stride))[0]= dc0;
1831         ((uint32_t*)(src+i*stride))[1]= dc1;
1832     }
1833     for(i=4; i<8; i++){
1834         ((uint32_t*)(src+i*stride))[0]= dc2;
1835         ((uint32_t*)(src+i*stride))[1]= dc3;
1836     }
1837 }
1838
1839 static void pred8x8_plane_c(uint8_t *src, int stride){
1840   int j, k;
1841   int a;
1842   uint8_t *cm = cropTbl + MAX_NEG_CROP;
1843   const uint8_t * const src0 = src+3-stride;
1844   const uint8_t *src1 = src+4*stride-1;
1845   const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
1846   int H = src0[1] - src0[-1];
1847   int V = src1[0] - src2[ 0];
1848   for(k=2; k<=4; ++k) {
1849     src1 += stride; src2 -= stride;
1850     H += k*(src0[k] - src0[-k]);
1851     V += k*(src1[0] - src2[ 0]);
1852   }
1853   H = ( 17*H+16 ) >> 5;
1854   V = ( 17*V+16 ) >> 5;
1855
1856   a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1857   for(j=8; j>0; --j) {
1858     int b = a;
1859     a += V;
1860     src[0] = cm[ (b    ) >> 5 ];
1861     src[1] = cm[ (b+  H) >> 5 ];
1862     src[2] = cm[ (b+2*H) >> 5 ];
1863     src[3] = cm[ (b+3*H) >> 5 ];
1864     src[4] = cm[ (b+4*H) >> 5 ];
1865     src[5] = cm[ (b+5*H) >> 5 ];
1866     src[6] = cm[ (b+6*H) >> 5 ];
1867     src[7] = cm[ (b+7*H) >> 5 ];
1868     src += stride;
1869   }
1870 }
1871
1872 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1873                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1874                            int src_x_offset, int src_y_offset,
1875                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1876     MpegEncContext * const s = &h->s;
1877     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1878     const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1879     const int luma_xy= (mx&3) + ((my&3)<<2);
1880     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1881     uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1882     uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1883     int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1884     int extra_height= extra_width;
1885     int emu=0;
1886     const int full_mx= mx>>2;
1887     const int full_my= my>>2;
1888
1889     assert(pic->data[0]);
1890
1891     if(mx&7) extra_width -= 3;
1892     if(my&7) extra_height -= 3;
1893
1894     if(   full_mx < 0-extra_width
1895        || full_my < 0-extra_height
1896        || full_mx + 16/*FIXME*/ > s->width + extra_width
1897        || full_my + 16/*FIXME*/ > s->height + extra_height){
1898         ff_emulated_edge_mc(s, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, s->width, s->height);
1899             src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1900         emu=1;
1901     }
1902
1903     qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1904     if(!square){
1905         qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1906     }
1907
1908     if(s->flags&CODEC_FLAG_GRAY) return;
1909
1910     if(emu){
1911         ff_emulated_edge_mc(s, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
1912             src_cb= s->edge_emu_buffer;
1913     }
1914     chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1915
1916     if(emu){
1917         ff_emulated_edge_mc(s, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
1918             src_cr= s->edge_emu_buffer;
1919     }
1920     chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1921 }
1922
1923 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1924                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1925                            int x_offset, int y_offset,
1926                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1927                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1928                            int list0, int list1){
1929     MpegEncContext * const s = &h->s;
1930     qpel_mc_func *qpix_op=  qpix_put;
1931     h264_chroma_mc_func chroma_op= chroma_put;
1932
1933     dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
1934     dest_cb +=   x_offset +   y_offset*s->uvlinesize;
1935     dest_cr +=   x_offset +   y_offset*s->uvlinesize;
1936     x_offset += 8*s->mb_x;
1937     y_offset += 8*s->mb_y;
1938
1939     if(list0){
1940         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1941         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1942                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
1943                            qpix_op, chroma_op);
1944
1945         qpix_op=  qpix_avg;
1946         chroma_op= chroma_avg;
1947     }
1948
1949     if(list1){
1950         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1951         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1952                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
1953                            qpix_op, chroma_op);
1954     }
1955 }
1956
1957 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1958                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1959                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
1960     MpegEncContext * const s = &h->s;
1961     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1962     const int mb_type= s->current_picture.mb_type[mb_xy];
1963
1964     assert(IS_INTER(mb_type));
1965
1966     if(IS_16X16(mb_type)){
1967         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1968                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1969                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1970     }else if(IS_16X8(mb_type)){
1971         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1972                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1973                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1974         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1975                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1976                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1977     }else if(IS_8X16(mb_type)){
1978         mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
1979                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1980                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1981         mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
1982                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1983                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1984     }else{
1985         int i;
1986
1987         assert(IS_8X8(mb_type));
1988
1989         for(i=0; i<4; i++){
1990             const int sub_mb_type= h->sub_mb_type[i];
1991             const int n= 4*i;
1992             int x_offset= (i&1)<<2;
1993             int y_offset= (i&2)<<1;
1994
1995             if(IS_SUB_8X8(sub_mb_type)){
1996                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1997                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1998                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1999             }else if(IS_SUB_8X4(sub_mb_type)){
2000                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2001                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2002                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2003                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2004                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2005                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2006             }else if(IS_SUB_4X8(sub_mb_type)){
2007                 mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2008                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2009                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2010                 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2011                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2012                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2013             }else{
2014                 int j;
2015                 assert(IS_SUB_4X4(sub_mb_type));
2016                 for(j=0; j<4; j++){
2017                     int sub_x_offset= x_offset + 2*(j&1);
2018                     int sub_y_offset= y_offset +   (j&2);
2019                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2020                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2021                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2022                 }
2023             }
2024         }
2025     }
2026 }
2027
2028 static void decode_init_vlc(H264Context *h){
2029     static int done = 0;
2030
2031     if (!done) {
2032         int i;
2033         done = 1;
2034
2035         init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2036                  &chroma_dc_coeff_token_len [0], 1, 1,
2037                  &chroma_dc_coeff_token_bits[0], 1, 1);
2038
2039         for(i=0; i<4; i++){
2040             init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2041                      &coeff_token_len [i][0], 1, 1,
2042                      &coeff_token_bits[i][0], 1, 1);
2043         }
2044
2045         for(i=0; i<3; i++){
2046             init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2047                      &chroma_dc_total_zeros_len [i][0], 1, 1,
2048                      &chroma_dc_total_zeros_bits[i][0], 1, 1);
2049         }
2050         for(i=0; i<15; i++){
2051             init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2052                      &total_zeros_len [i][0], 1, 1,
2053                      &total_zeros_bits[i][0], 1, 1);
2054         }
2055
2056         for(i=0; i<6; i++){
2057             init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2058                      &run_len [i][0], 1, 1,
2059                      &run_bits[i][0], 1, 1);
2060         }
2061         init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2062                  &run_len [6][0], 1, 1,
2063                  &run_bits[6][0], 1, 1);
2064     }
2065 }
2066
2067 /**
2068  * Sets the intra prediction function pointers.
2069  */
2070 static void init_pred_ptrs(H264Context *h){
2071 //    MpegEncContext * const s = &h->s;
2072
2073     h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2074     h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2075     h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2076     h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2077     h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2078     h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2079     h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2080     h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2081     h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2082     h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2083     h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2084     h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2085
2086     h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2087     h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2088     h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2089     h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2090     h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2091     h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2092     h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2093
2094     h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2095     h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2096     h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2097     h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2098     h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2099     h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2100     h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2101 }
2102
2103 //FIXME factorize
2104 #define CHECKED_ALLOCZ(p, size)\
2105 {\
2106     p= av_mallocz(size);\
2107     if(p==NULL){\
2108         perror("malloc");\
2109         goto fail;\
2110     }\
2111 }
2112
2113 static void free_tables(H264Context *h){
2114     av_freep(&h->intra4x4_pred_mode);
2115     av_freep(&h->non_zero_count);
2116     av_freep(&h->slice_table_base);
2117     h->slice_table= NULL;
2118
2119     av_freep(&h->mb2b_xy);
2120     av_freep(&h->mb2b8_xy);
2121 }
2122
2123 /**
2124  * allocates tables.
2125  * needs widzh/height
2126  */
2127 static int alloc_tables(H264Context *h){
2128     MpegEncContext * const s = &h->s;
2129     const int big_mb_num= s->mb_stride * (s->mb_height+1);
2130     int x,y;
2131
2132     CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2133     CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2134     CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2135
2136     memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2137     h->slice_table= h->slice_table_base + s->mb_stride + 1;
2138
2139     CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint16_t));
2140     CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2141     for(y=0; y<s->mb_height; y++){
2142         for(x=0; x<s->mb_width; x++){
2143             const int mb_xy= x + y*s->mb_stride;
2144             const int b_xy = 4*x + 4*y*h->b_stride;
2145             const int b8_xy= 2*x + 2*y*h->b8_stride;
2146
2147             h->mb2b_xy [mb_xy]= b_xy;
2148             h->mb2b8_xy[mb_xy]= b8_xy;
2149         }
2150     }
2151
2152     return 0;
2153 fail:
2154     free_tables(h);
2155     return -1;
2156 }
2157
2158 static void common_init(H264Context *h){
2159     MpegEncContext * const s = &h->s;
2160
2161     s->width = s->avctx->width;
2162     s->height = s->avctx->height;
2163     s->codec_id= s->avctx->codec->id;
2164
2165     init_pred_ptrs(h);
2166
2167     s->decode=1; //FIXME
2168 }
2169
2170 static int decode_init(AVCodecContext *avctx){
2171     H264Context *h= avctx->priv_data;
2172     MpegEncContext * const s = &h->s;
2173
2174     s->avctx = avctx;
2175     common_init(h);
2176
2177     s->out_format = FMT_H264;
2178     s->workaround_bugs= avctx->workaround_bugs;
2179
2180     // set defaults
2181     s->progressive_sequence=1;
2182 //    s->decode_mb= ff_h263_decode_mb;
2183     s->low_delay= 1;
2184     avctx->pix_fmt= PIX_FMT_YUV420P;
2185
2186     decode_init_vlc(h);
2187
2188     return 0;
2189 }
2190
2191 static void frame_start(H264Context *h){
2192     MpegEncContext * const s = &h->s;
2193     int i;
2194
2195     MPV_frame_start(s, s->avctx);
2196     ff_er_frame_start(s);
2197     h->mmco_index=0;
2198
2199     assert(s->linesize && s->uvlinesize);
2200
2201     for(i=0; i<16; i++){
2202         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2203         h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2204     }
2205     for(i=0; i<4; i++){
2206         h->block_offset[16+i]=
2207         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2208     }
2209
2210 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2211 }
2212
2213 static void hl_decode_mb(H264Context *h){
2214     MpegEncContext * const s = &h->s;
2215     const int mb_x= s->mb_x;
2216     const int mb_y= s->mb_y;
2217     const int mb_xy= mb_x + mb_y*s->mb_stride;
2218     const int mb_type= s->current_picture.mb_type[mb_xy];
2219     uint8_t  *dest_y, *dest_cb, *dest_cr;
2220     int linesize, uvlinesize /*dct_offset*/;
2221     int i;
2222
2223     if(!s->decode)
2224         return;
2225
2226     if(s->mb_skiped){
2227     }
2228
2229     dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2230     dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2231     dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2232
2233     if (h->mb_field_decoding_flag) {
2234         linesize = s->linesize * 2;
2235         uvlinesize = s->uvlinesize * 2;
2236         if(mb_y&1){ //FIXME move out of this func?
2237             dest_y -= s->linesize*15;
2238             dest_cb-= s->linesize*7;
2239             dest_cr-= s->linesize*7;
2240         }
2241     } else {
2242         linesize = s->linesize;
2243         uvlinesize = s->uvlinesize;
2244 //        dct_offset = s->linesize * 16;
2245     }
2246
2247     if(IS_INTRA(mb_type)){
2248         if(!(s->flags&CODEC_FLAG_GRAY)){
2249             h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2250             h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2251         }
2252
2253         if(IS_INTRA4x4(mb_type)){
2254             if(!s->encoding){
2255                 for(i=0; i<16; i++){
2256                     uint8_t * const ptr= dest_y + h->block_offset[i];
2257                     uint8_t *topright= ptr + 4 - linesize;
2258                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2259                     const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2260                     int tr;
2261
2262                     if(!topright_avail){
2263                         tr= ptr[3 - linesize]*0x01010101;
2264                         topright= (uint8_t*) &tr;
2265                     }
2266
2267                     h->pred4x4[ dir ](ptr, topright, linesize);
2268                     if(h->non_zero_count_cache[ scan8[i] ]){
2269                         if(s->codec_id == CODEC_ID_H264)
2270                             h264_add_idct_c(ptr, h->mb + i*16, linesize);
2271                         else
2272                             svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2273                     }
2274                 }
2275             }
2276         }else{
2277             h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2278             if(s->codec_id == CODEC_ID_H264)
2279                 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2280             else
2281                 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2282         }
2283     }else if(s->codec_id == CODEC_ID_H264){
2284         hl_motion(h, dest_y, dest_cb, dest_cr,
2285                   s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2286                   s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2287     }
2288
2289
2290     if(!IS_INTRA4x4(mb_type)){
2291         if(s->codec_id == CODEC_ID_H264){
2292             for(i=0; i<16; i++){
2293                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2294                     uint8_t * const ptr= dest_y + h->block_offset[i];
2295                     h264_add_idct_c(ptr, h->mb + i*16, linesize);
2296                 }
2297             }
2298         }else{
2299             for(i=0; i<16; i++){
2300                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2301                     uint8_t * const ptr= dest_y + h->block_offset[i];
2302                     svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2303                 }
2304             }
2305         }
2306     }
2307
2308     if(!(s->flags&CODEC_FLAG_GRAY)){
2309         chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2310         chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2311         if(s->codec_id == CODEC_ID_H264){
2312             for(i=16; i<16+4; i++){
2313                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2314                     uint8_t * const ptr= dest_cb + h->block_offset[i];
2315                     h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2316                 }
2317             }
2318             for(i=20; i<20+4; i++){
2319                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2320                     uint8_t * const ptr= dest_cr + h->block_offset[i];
2321                     h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2322                 }
2323             }
2324         }else{
2325             for(i=16; i<16+4; i++){
2326                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2327                     uint8_t * const ptr= dest_cb + h->block_offset[i];
2328                     svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2329                 }
2330             }
2331             for(i=20; i<20+4; i++){
2332                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2333                     uint8_t * const ptr= dest_cr + h->block_offset[i];
2334                     svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2335                 }
2336             }
2337         }
2338     }
2339 }
2340
2341 static void decode_mb_cabac(H264Context *h){
2342 //    MpegEncContext * const s = &h->s;
2343 }
2344
2345 /**
2346  * fills the default_ref_list.
2347  */
2348 static int fill_default_ref_list(H264Context *h){
2349     MpegEncContext * const s = &h->s;
2350     int i;
2351     Picture sorted_short_ref[16];
2352
2353     if(h->slice_type==B_TYPE){
2354         int out_i;
2355         int limit= -1;
2356
2357         for(out_i=0; out_i<h->short_ref_count; out_i++){
2358             int best_i=-1;
2359             int best_poc=-1;
2360
2361             for(i=0; i<h->short_ref_count; i++){
2362                 const int poc= h->short_ref[i]->poc;
2363                 if(poc > limit && poc < best_poc){
2364                     best_poc= poc;
2365                     best_i= i;
2366                 }
2367             }
2368
2369             assert(best_i != -1);
2370
2371             limit= best_poc;
2372             sorted_short_ref[out_i]= *h->short_ref[best_i];
2373         }
2374     }
2375
2376     if(s->picture_structure == PICT_FRAME){
2377         if(h->slice_type==B_TYPE){
2378             const int current_poc= s->current_picture_ptr->poc;
2379             int list;
2380
2381             for(list=0; list<2; list++){
2382                 int index=0;
2383
2384                 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2385                     const int i2= list ? h->short_ref_count - i - 1 : i;
2386                     const int poc= sorted_short_ref[i2].poc;
2387
2388                     if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2389
2390                     if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2391                         h->default_ref_list[list][index  ]= sorted_short_ref[i2];
2392                         h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2393                     }
2394                 }
2395
2396                 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2397                     if(h->long_ref[i]->reference != 3) continue;
2398
2399                     h->default_ref_list[ list ][index  ]= *h->long_ref[i];
2400                     h->default_ref_list[ list ][index++].pic_id= i;;
2401                 }
2402
2403                 if(h->long_ref_count > 1 && h->short_ref_count==0){
2404                     Picture temp= h->default_ref_list[1][0];
2405                     h->default_ref_list[1][0] = h->default_ref_list[1][1];
2406                     h->default_ref_list[1][0] = temp;
2407                 }
2408
2409                 if(index < h->ref_count[ list ])
2410                     memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2411             }
2412         }else{
2413             int index=0;
2414             for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2415                 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2416                 h->default_ref_list[0][index  ]= *h->short_ref[i];
2417                 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2418             }
2419             for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2420                 if(h->long_ref[i]->reference != 3) continue;
2421                 h->default_ref_list[0][index  ]= *h->long_ref[i];
2422                 h->default_ref_list[0][index++].pic_id= i;;
2423             }
2424             if(index < h->ref_count[0])
2425                 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2426         }
2427     }else{ //FIELD
2428         if(h->slice_type==B_TYPE){
2429         }else{
2430             //FIXME second field balh
2431         }
2432     }
2433     return 0;
2434 }
2435
2436 static int decode_ref_pic_list_reordering(H264Context *h){
2437     MpegEncContext * const s = &h->s;
2438     int list;
2439
2440     if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2441
2442     for(list=0; list<2; list++){
2443         memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2444
2445         if(get_bits1(&s->gb)){
2446             int pred= h->curr_pic_num;
2447             int index;
2448
2449             for(index=0; ; index++){
2450                 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2451                 int pic_id;
2452                 int i;
2453
2454
2455                 if(index >= h->ref_count[list]){
2456                     fprintf(stderr, "reference count overflow\n");
2457                     return -1;
2458                 }
2459
2460                 if(reordering_of_pic_nums_idc<3){
2461                     if(reordering_of_pic_nums_idc<2){
2462                         const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2463
2464                         if(abs_diff_pic_num >= h->max_pic_num){
2465                             fprintf(stderr, "abs_diff_pic_num overflow\n");
2466                             return -1;
2467                         }
2468
2469                         if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2470                         else                                pred+= abs_diff_pic_num;
2471                         pred &= h->max_pic_num - 1;
2472
2473                         for(i= h->ref_count[list]-1; i>=index; i--){
2474                             if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2475                                 break;
2476                         }
2477                     }else{
2478                         pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2479
2480                         for(i= h->ref_count[list]-1; i>=index; i--){
2481                             if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2482                                 break;
2483                         }
2484                     }
2485
2486                     if(i < index){
2487                         fprintf(stderr, "reference picture missing during reorder\n");
2488                         memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2489                     }else if(i > index){
2490                         Picture tmp= h->ref_list[list][i];
2491                         for(; i>index; i--){
2492                             h->ref_list[list][i]= h->ref_list[list][i-1];
2493                         }
2494                         h->ref_list[list][index]= tmp;
2495                     }
2496                 }else if(reordering_of_pic_nums_idc==3)
2497                     break;
2498                 else{
2499                     fprintf(stderr, "illegal reordering_of_pic_nums_idc\n");
2500                     return -1;
2501                 }
2502             }
2503         }
2504
2505         if(h->slice_type!=B_TYPE) break;
2506     }
2507     return 0;
2508 }
2509
2510 static int pred_weight_table(H264Context *h){
2511     MpegEncContext * const s = &h->s;
2512     int list, i;
2513
2514     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2515     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2516
2517     for(list=0; list<2; list++){
2518         for(i=0; i<h->ref_count[list]; i++){
2519             int luma_weight_flag, chroma_weight_flag;
2520
2521             luma_weight_flag= get_bits1(&s->gb);
2522             if(luma_weight_flag){
2523                 h->luma_weight[list][i]= get_se_golomb(&s->gb);
2524                 h->luma_offset[list][i]= get_se_golomb(&s->gb);
2525             }
2526
2527             chroma_weight_flag= get_bits1(&s->gb);
2528             if(chroma_weight_flag){
2529                 int j;
2530                 for(j=0; j<2; j++){
2531                     h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2532                     h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2533                 }
2534             }
2535         }
2536         if(h->slice_type != B_TYPE) break;
2537     }
2538     return 0;
2539 }
2540
2541 /**
2542  * instantaneos decoder refresh.
2543  */
2544 static void idr(H264Context *h){
2545     int i;
2546
2547     for(i=0; i<h->long_ref_count; i++){
2548         h->long_ref[i]->reference=0;
2549         h->long_ref[i]= NULL;
2550     }
2551     h->long_ref_count=0;
2552
2553     for(i=0; i<h->short_ref_count; i++){
2554         h->short_ref[i]->reference=0;
2555         h->short_ref[i]= NULL;
2556     }
2557     h->short_ref_count=0;
2558 }
2559
2560 /**
2561  *
2562  * @return the removed picture or NULL if an error occures
2563  */
2564 static Picture * remove_short(H264Context *h, int frame_num){
2565     MpegEncContext * const s = &h->s;
2566     int i;
2567
2568     if(s->avctx->debug&FF_DEBUG_MMCO)
2569         printf("remove short %d count %d\n", frame_num, h->short_ref_count);
2570
2571     for(i=0; i<h->short_ref_count; i++){
2572         Picture *pic= h->short_ref[i];
2573         if(s->avctx->debug&FF_DEBUG_MMCO)
2574             printf("%d %d %p\n", i, pic->frame_num, pic);
2575         if(pic->frame_num == frame_num){
2576             h->short_ref[i]= NULL;
2577             memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2578             h->short_ref_count--;
2579             return pic;
2580         }
2581     }
2582     return NULL;
2583 }
2584
2585 /**
2586  *
2587  * @return the removed picture or NULL if an error occures
2588  */
2589 static Picture * remove_long(H264Context *h, int i){
2590     Picture *pic;
2591
2592     if(i >= h->long_ref_count) return NULL;
2593     pic= h->long_ref[i];
2594     if(pic==NULL) return NULL;
2595
2596     h->long_ref[i]= NULL;
2597     memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2598     h->long_ref_count--;
2599
2600     return pic;
2601 }
2602
2603 /**
2604  * Executes the reference picture marking (memory management control operations).
2605  */
2606 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2607     MpegEncContext * const s = &h->s;
2608     int i;
2609     int current_is_long=0;
2610     Picture *pic;
2611
2612     if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2613         printf("no mmco here\n");
2614
2615     for(i=0; i<mmco_count; i++){
2616         if(s->avctx->debug&FF_DEBUG_MMCO)
2617             printf("mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
2618
2619         switch(mmco[i].opcode){
2620         case MMCO_SHORT2UNUSED:
2621             pic= remove_short(h, mmco[i].short_frame_num);
2622             if(pic==NULL) return -1;
2623             pic->reference= 0;
2624             break;
2625         case MMCO_SHORT2LONG:
2626             pic= remove_long(h, mmco[i].long_index);
2627             if(pic) pic->reference=0;
2628
2629             h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2630             h->long_ref[ mmco[i].long_index ]->long_ref=1;
2631             break;
2632         case MMCO_LONG2UNUSED:
2633             pic= remove_long(h, mmco[i].long_index);
2634             if(pic==NULL) return -1;
2635             pic->reference= 0;
2636             break;
2637         case MMCO_LONG:
2638             pic= remove_long(h, mmco[i].long_index);
2639             if(pic) pic->reference=0;
2640
2641             h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2642             h->long_ref[ mmco[i].long_index ]->long_ref=1;
2643             h->long_ref_count++;
2644
2645             current_is_long=1;
2646             break;
2647         case MMCO_SET_MAX_LONG:
2648             assert(mmco[i].long_index <= 16);
2649             while(mmco[i].long_index < h->long_ref_count){
2650                 pic= remove_long(h, mmco[i].long_index);
2651                 pic->reference=0;
2652             }
2653             while(mmco[i].long_index > h->long_ref_count){
2654                 h->long_ref[ h->long_ref_count++ ]= NULL;
2655             }
2656             break;
2657         case MMCO_RESET:
2658             while(h->short_ref_count){
2659                 pic= remove_short(h, h->short_ref[0]->frame_num);
2660                 pic->reference=0;
2661             }
2662             while(h->long_ref_count){
2663                 pic= remove_long(h, h->long_ref_count-1);
2664                 pic->reference=0;
2665             }
2666             break;
2667         default: assert(0);
2668         }
2669     }
2670
2671     if(!current_is_long){
2672         pic= remove_short(h, s->current_picture_ptr->frame_num);
2673         if(pic){
2674             pic->reference=0;
2675             fprintf(stderr, "illegal short term buffer state detected\n");
2676         }
2677
2678         if(h->short_ref_count)
2679             memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2680
2681         h->short_ref[0]= s->current_picture_ptr;
2682         h->short_ref[0]->long_ref=0;
2683         h->short_ref_count++;
2684     }
2685
2686     return 0;
2687 }
2688
2689 static int decode_ref_pic_marking(H264Context *h){
2690     MpegEncContext * const s = &h->s;
2691     int i;
2692
2693     if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2694         s->broken_link= get_bits1(&s->gb) -1;
2695         h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2696         if(h->mmco[0].long_index == -1)
2697             h->mmco_index= 0;
2698         else{
2699             h->mmco[0].opcode= MMCO_LONG;
2700             h->mmco_index= 1;
2701         }
2702     }else{
2703         if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2704             for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) {
2705                 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2706
2707                 h->mmco[i].opcode= opcode;
2708                 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2709                     h->mmco[i].short_frame_num= (h->frame_num - get_ue_golomb(&s->gb) - 1) & ((1<<h->sps.log2_max_frame_num)-1); //FIXME fields
2710 /*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2711                         fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2712                         return -1;
2713                     }*/
2714                 }
2715                 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2716                     h->mmco[i].long_index= get_ue_golomb(&s->gb);
2717                     if(/*h->mmco[i].long_index >= h->long_ref_count || h->long_ref[ h->mmco[i].long_index ] == NULL*/ h->mmco[i].long_index >= 16){
2718                         fprintf(stderr, "illegal long ref in memory management control operation %d\n", opcode);
2719                         return -1;
2720                     }
2721                 }
2722
2723                 if(opcode > MMCO_LONG){
2724                     fprintf(stderr, "illegal memory management control operation %d\n", opcode);
2725                     return -1;
2726                 }
2727             }
2728             h->mmco_index= i;
2729         }else{
2730             assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2731
2732             if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2733                 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2734                 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2735                 h->mmco_index= 1;
2736             }else
2737                 h->mmco_index= 0;
2738         }
2739     }
2740
2741     return 0;
2742 }
2743
2744 static int init_poc(H264Context *h){
2745     MpegEncContext * const s = &h->s;
2746     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2747     int field_poc[2];
2748
2749     if(h->nal_unit_type == NAL_IDR_SLICE){
2750         h->frame_num_offset= 0;
2751     }else{
2752         if(h->frame_num < h->prev_frame_num)
2753             h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2754         else
2755             h->frame_num_offset= h->prev_frame_num_offset;
2756     }
2757
2758     if(h->sps.poc_type==0){
2759         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2760
2761         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2762             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2763         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2764             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2765         else
2766             h->poc_msb = h->prev_poc_msb;
2767 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2768         field_poc[0] =
2769         field_poc[1] = h->poc_msb + h->poc_lsb;
2770         if(s->picture_structure == PICT_FRAME)
2771             field_poc[1] += h->delta_poc_bottom;
2772     }else if(h->sps.poc_type==1){
2773         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2774         int i;
2775
2776         if(h->sps.poc_cycle_length != 0)
2777             abs_frame_num = h->frame_num_offset + h->frame_num;
2778         else
2779             abs_frame_num = 0;
2780
2781         if(h->nal_ref_idc==0 && abs_frame_num > 0)
2782             abs_frame_num--;
2783
2784         expected_delta_per_poc_cycle = 0;
2785         for(i=0; i < h->sps.poc_cycle_length; i++)
2786             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2787
2788         if(abs_frame_num > 0){
2789             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2790             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2791
2792             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2793             for(i = 0; i <= frame_num_in_poc_cycle; i++)
2794                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2795         } else
2796             expectedpoc = 0;
2797
2798         if(h->nal_ref_idc == 0)
2799             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2800
2801         field_poc[0] = expectedpoc + h->delta_poc[0];
2802         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2803
2804         if(s->picture_structure == PICT_FRAME)
2805             field_poc[1] += h->delta_poc[1];
2806     }else{
2807         int poc;
2808         if(h->nal_unit_type == NAL_IDR_SLICE){
2809             poc= 0;
2810         }else{
2811             if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2812             else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2813         }
2814         field_poc[0]= poc;
2815         field_poc[1]= poc;
2816     }
2817
2818     if(s->picture_structure != PICT_BOTTOM_FIELD)
2819         s->current_picture_ptr->field_poc[0]= field_poc[0];
2820     if(s->picture_structure != PICT_TOP_FIELD)
2821         s->current_picture_ptr->field_poc[1]= field_poc[1];
2822     if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2823         s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2824
2825     return 0;
2826 }
2827
2828 /**
2829  * decodes a slice header.
2830  * this will allso call MPV_common_init() and frame_start() as needed
2831  */
2832 static int decode_slice_header(H264Context *h){
2833     MpegEncContext * const s = &h->s;
2834     int first_mb_in_slice, pps_id;
2835     int num_ref_idx_active_override_flag;
2836     static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2837     float new_aspect;
2838
2839     s->current_picture.reference= h->nal_ref_idc != 0;
2840
2841     first_mb_in_slice= get_ue_golomb(&s->gb);
2842
2843     h->slice_type= get_ue_golomb(&s->gb);
2844     if(h->slice_type > 9){
2845         fprintf(stderr, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
2846     }
2847     if(h->slice_type > 4){
2848         h->slice_type -= 5;
2849         h->slice_type_fixed=1;
2850     }else
2851         h->slice_type_fixed=0;
2852
2853     h->slice_type= slice_type_map[ h->slice_type ];
2854
2855     s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
2856
2857     pps_id= get_ue_golomb(&s->gb);
2858     if(pps_id>255){
2859         fprintf(stderr, "pps_id out of range\n");
2860         return -1;
2861     }
2862     h->pps= h->pps_buffer[pps_id];
2863     if(h->pps.slice_group_count == 0){
2864         fprintf(stderr, "non existing PPS referenced\n");
2865         return -1;
2866     }
2867
2868     h->sps= h->sps_buffer[ h->pps.sps_id ];
2869     if(h->sps.log2_max_frame_num == 0){
2870         fprintf(stderr, "non existing SPS referenced\n");
2871         return -1;
2872     }
2873
2874     s->mb_width= h->sps.mb_width;
2875     s->mb_height= h->sps.mb_height;
2876
2877     h->b_stride=  s->mb_width*4;
2878     h->b8_stride= s->mb_width*2;
2879
2880     s->mb_x = first_mb_in_slice % s->mb_width;
2881     s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
2882
2883     s->width = 16*s->mb_width - 2*(h->pps.crop_left + h->pps.crop_right );
2884     if(h->sps.frame_mbs_only_flag)
2885         s->height= 16*s->mb_height - 2*(h->pps.crop_top  + h->pps.crop_bottom);
2886     else
2887         s->height= 16*s->mb_height - 4*(h->pps.crop_top  + h->pps.crop_bottom); //FIXME recheck
2888
2889     if(h->pps.crop_left || h->pps.crop_top){
2890         fprintf(stderr, "insane croping not completly supported, this could look slightly wrong ...\n");
2891     }
2892
2893     if(s->aspected_height) //FIXME emms at end of slice ?
2894         new_aspect= h->sps.sar_width*s->width / (float)(s->height*h->sps.sar_height);
2895     else
2896         new_aspect=0;
2897
2898     if (s->context_initialized
2899         && (   s->width != s->avctx->width || s->height != s->avctx->height
2900             || ABS(new_aspect - s->avctx->aspect_ratio) > 0.001)) {
2901         free_tables(h);
2902         MPV_common_end(s);
2903     }
2904     if (!s->context_initialized) {
2905         if (MPV_common_init(s) < 0)
2906             return -1;
2907
2908         alloc_tables(h);
2909
2910         s->avctx->width = s->width;
2911         s->avctx->height = s->height;
2912         s->avctx->aspect_ratio= new_aspect;
2913     }
2914
2915     if(first_mb_in_slice == 0){
2916         frame_start(h);
2917     }
2918
2919     s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
2920     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2921
2922     if(h->sps.frame_mbs_only_flag){
2923         s->picture_structure= PICT_FRAME;
2924     }else{
2925         if(get_bits1(&s->gb)) //field_pic_flag
2926             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2927         else
2928             s->picture_structure= PICT_FRAME;
2929     }
2930
2931     if(s->picture_structure==PICT_FRAME){
2932         h->curr_pic_num=   h->frame_num;
2933         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2934     }else{
2935         h->curr_pic_num= 2*h->frame_num;
2936         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2937     }
2938
2939     if(h->nal_unit_type == NAL_IDR_SLICE){
2940         int idr_pic_id= get_ue_golomb(&s->gb);
2941     }
2942
2943     if(h->sps.poc_type==0){
2944         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2945
2946         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2947             h->delta_poc_bottom= get_se_golomb(&s->gb);
2948         }
2949     }
2950
2951     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2952         h->delta_poc[0]= get_se_golomb(&s->gb);
2953
2954         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2955             h->delta_poc[1]= get_se_golomb(&s->gb);
2956     }
2957
2958     init_poc(h);
2959
2960     if(h->pps.redundant_pic_cnt_present){
2961         h->redundant_pic_count= get_ue_golomb(&s->gb);
2962     }
2963
2964     //set defaults, might be overriden a few line later
2965     h->ref_count[0]= h->pps.ref_count[0];
2966     h->ref_count[1]= h->pps.ref_count[1];
2967
2968     if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
2969         if(h->slice_type == B_TYPE){
2970             h->direct_spatial_mv_pred= get_bits1(&s->gb);
2971         }
2972         num_ref_idx_active_override_flag= get_bits1(&s->gb);
2973
2974         if(num_ref_idx_active_override_flag){
2975             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2976             if(h->slice_type==B_TYPE)
2977                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2978
2979             if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
2980                 fprintf(stderr, "reference overflow\n");
2981                 return -1;
2982             }
2983         }
2984     }
2985
2986     if(first_mb_in_slice == 0){
2987         fill_default_ref_list(h);
2988     }
2989
2990     decode_ref_pic_list_reordering(h);
2991
2992     if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
2993        || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
2994         pred_weight_table(h);
2995
2996     if(s->current_picture.reference)
2997         decode_ref_pic_marking(h);
2998     //FIXME CABAC stuff
2999
3000     s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta
3001     //FIXME qscale / qp ... stuff
3002     if(h->slice_type == SP_TYPE){
3003         int sp_for_switch_flag= get_bits1(&s->gb);
3004     }
3005     if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3006         int slice_qs_delta= get_se_golomb(&s->gb);
3007     }
3008
3009     if( h->pps.deblocking_filter_parameters_present ) {
3010         h->disable_deblocking_filter_idc= get_ue_golomb(&s->gb);
3011         if( h->disable_deblocking_filter_idc  !=  1 ) {
3012             h->slice_alpha_c0_offset_div2= get_se_golomb(&s->gb);
3013             h->slice_beta_offset_div2= get_se_golomb(&s->gb);
3014         }
3015     }else
3016         h->disable_deblocking_filter_idc= 0;
3017
3018 #if 0 //FMO
3019     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3020         slice_group_change_cycle= get_bits(&s->gb, ?);
3021 #endif
3022
3023     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3024         printf("mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n",
3025                first_mb_in_slice,
3026                av_get_pict_type_char(h->slice_type),
3027                pps_id, h->frame_num,
3028                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3029                h->ref_count[0], h->ref_count[1],
3030                s->qscale,
3031                h->disable_deblocking_filter_idc
3032                );
3033     }
3034
3035     return 0;
3036 }
3037
3038 /**
3039  *
3040  */
3041 static inline int get_level_prefix(GetBitContext *gb){
3042     unsigned int buf;
3043     int log;
3044
3045     OPEN_READER(re, gb);
3046     UPDATE_CACHE(re, gb);
3047     buf=GET_CACHE(re, gb);
3048
3049     log= 32 - av_log2(buf);
3050 #ifdef TRACE
3051     print_bin(buf>>(32-log), log);
3052     printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3053 #endif
3054
3055     LAST_SKIP_BITS(re, gb, log);
3056     CLOSE_READER(re, gb);
3057
3058     return log-1;
3059 }
3060
3061 /**
3062  * decodes a residual block.
3063  * @param n block index
3064  * @param scantable scantable
3065  * @param max_coeff number of coefficients in the block
3066  * @return <0 if an error occured
3067  */
3068 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3069     MpegEncContext * const s = &h->s;
3070     const uint16_t *qmul= dequant_coeff[qp];
3071     static const int coeff_token_table_index[17]= {0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3};
3072     int level[16], run[16];
3073     int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3074
3075     //FIXME put trailing_onex into the context
3076
3077     if(n == CHROMA_DC_BLOCK_INDEX){
3078         coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3079         total_coeff= coeff_token>>2;
3080     }else{
3081         if(n == LUMA_DC_BLOCK_INDEX){
3082             total_coeff= pred_non_zero_count(h, 0);
3083             coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3084             total_coeff= coeff_token>>2;
3085         }else{
3086             total_coeff= pred_non_zero_count(h, n);
3087             coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3088             total_coeff= coeff_token>>2;
3089             h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3090         }
3091     }
3092
3093     //FIXME set last_non_zero?
3094
3095     if(total_coeff==0)
3096         return 0;
3097
3098     trailing_ones= coeff_token&3;
3099     tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3100     assert(total_coeff<=16);
3101
3102     for(i=0; i<trailing_ones; i++){
3103         level[i]= 1 - 2*get_bits1(gb);
3104     }
3105
3106     suffix_length= total_coeff > 10 && trailing_ones < 3;
3107
3108     for(; i<total_coeff; i++){
3109         const int prefix= get_level_prefix(gb);
3110         int level_code, mask;
3111
3112         if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3113             if(suffix_length)
3114                 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3115             else
3116                 level_code= (prefix<<suffix_length); //part
3117         }else if(prefix==14){
3118             if(suffix_length)
3119                 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3120             else
3121                 level_code= prefix + get_bits(gb, 4); //part
3122         }else if(prefix==15){
3123             level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3124             if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3125         }else{
3126             fprintf(stderr, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3127             return -1;
3128         }
3129
3130         if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3131
3132         mask= -(level_code&1);
3133         level[i]= (((2+level_code)>>1) ^ mask) - mask;
3134
3135         if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3136
3137 #if 1
3138         if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3139 #else
3140         if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3141         ? == prefix > 2 or sth
3142 #endif
3143         tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3144     }
3145
3146     if(total_coeff == max_coeff)
3147         zeros_left=0;
3148     else{
3149         if(n == CHROMA_DC_BLOCK_INDEX)
3150             zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3151         else
3152             zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3153     }
3154
3155     for(i=0; i<total_coeff-1; i++){
3156         if(zeros_left <=0)
3157             break;
3158         else if(zeros_left < 7){
3159             run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3160         }else{
3161             run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3162         }
3163         zeros_left -= run[i];
3164     }
3165
3166     if(zeros_left<0){
3167         fprintf(stderr, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3168         return -1;
3169     }
3170
3171     for(; i<total_coeff-1; i++){
3172         run[i]= 0;
3173     }
3174
3175     run[i]= zeros_left;
3176
3177     coeff_num=-1;
3178     if(n > 24){
3179         for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3180             int j;
3181
3182             coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3183             j= scantable[ coeff_num ];
3184
3185             block[j]= level[i];
3186         }
3187     }else{
3188         for(i=total_coeff-1; i>=0; i--){ //FIXME merge into  rundecode?
3189             int j;
3190
3191             coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3192             j= scantable[ coeff_num ];
3193
3194             block[j]= level[i] * qmul[j];
3195 //            printf("%d %d  ", block[j], qmul[j]);
3196         }
3197     }
3198     return 0;
3199 }
3200
3201 /**
3202  * decodes a macroblock
3203  * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3204  */
3205 static int decode_mb(H264Context *h){
3206     MpegEncContext * const s = &h->s;
3207     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3208     int mb_type, partition_count, cbp;
3209
3210     s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?
3211
3212     tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3213
3214     if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3215         if(s->mb_skip_run==-1)
3216             s->mb_skip_run= get_ue_golomb(&s->gb);
3217
3218         if (s->mb_skip_run--) {
3219             int mx, my;
3220             /* skip mb */
3221 //FIXME b frame
3222             mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3223
3224             memset(h->non_zero_count[mb_xy], 0, 16);
3225             memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3226
3227             if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3228                 h->mb_field_decoding_flag= get_bits1(&s->gb);
3229             }
3230
3231             if(h->mb_field_decoding_flag)
3232                 mb_type|= MB_TYPE_INTERLACED;
3233
3234             fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3235             pred_pskip_motion(h, &mx, &my);
3236             fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3237             fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3238             write_back_motion(h, mb_type);
3239
3240             s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3241             h->slice_table[ mb_xy ]= h->slice_num;
3242
3243             h->prev_mb_skiped= 1;
3244             return 0;
3245         }
3246     }
3247     if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3248         if((s->mb_y&1)==0)
3249             h->mb_field_decoding_flag = get_bits1(&s->gb);
3250     }else
3251         h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3252
3253     h->prev_mb_skiped= 0;
3254
3255     mb_type= get_ue_golomb(&s->gb);
3256     if(h->slice_type == B_TYPE){
3257         if(mb_type < 23){
3258             partition_count= b_mb_type_info[mb_type].partition_count;
3259             mb_type=         b_mb_type_info[mb_type].type;
3260         }else{
3261             mb_type -= 23;
3262             goto decode_intra_mb;
3263         }
3264     }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3265         if(mb_type < 5){
3266             partition_count= p_mb_type_info[mb_type].partition_count;
3267             mb_type=         p_mb_type_info[mb_type].type;
3268         }else{
3269             mb_type -= 5;
3270             goto decode_intra_mb;
3271         }
3272     }else{
3273        assert(h->slice_type == I_TYPE);
3274 decode_intra_mb:
3275         if(mb_type > 25){
3276             fprintf(stderr, "mb_type %d in %c slice to large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);
3277             return -1;
3278         }
3279         partition_count=0;
3280         cbp= i_mb_type_info[mb_type].cbp;
3281         h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3282         mb_type= i_mb_type_info[mb_type].type;
3283     }
3284
3285     if(h->mb_field_decoding_flag)
3286         mb_type |= MB_TYPE_INTERLACED;
3287
3288     s->current_picture.mb_type[mb_xy]= mb_type;
3289     h->slice_table[ mb_xy ]= h->slice_num;
3290
3291     if(IS_INTRA_PCM(mb_type)){
3292         const uint8_t *ptr;
3293         int x, y;
3294
3295         // we assume these blocks are very rare so we dont optimize it
3296         align_get_bits(&s->gb);
3297
3298         ptr= s->gb.buffer + get_bits_count(&s->gb);
3299
3300         for(y=0; y<16; y++){
3301             const int index= 4*(y&3) + 64*(y>>2);
3302             for(x=0; x<16; x++){
3303                 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3304             }
3305         }
3306         for(y=0; y<8; y++){
3307             const int index= 256 + 4*(y&3) + 32*(y>>2);
3308             for(x=0; x<8; x++){
3309                 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3310             }
3311         }
3312         for(y=0; y<8; y++){
3313             const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3314             for(x=0; x<8; x++){
3315                 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3316             }
3317         }
3318
3319         skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3320
3321         memset(h->non_zero_count[mb_xy], 16, 16);
3322
3323         return 0;
3324     }
3325
3326     fill_caches(h, mb_type);
3327
3328     //mb_pred
3329     if(IS_INTRA(mb_type)){
3330 //            init_top_left_availability(h);
3331             if(IS_INTRA4x4(mb_type)){
3332                 int i;
3333
3334 //                fill_intra4x4_pred_table(h);
3335                 for(i=0; i<16; i++){
3336                     const int mode_coded= !get_bits1(&s->gb);
3337                     const int predicted_mode=  pred_intra_mode(h, i);
3338                     int mode;
3339
3340                     if(mode_coded){
3341                         const int rem_mode= get_bits(&s->gb, 3);
3342                         if(rem_mode<predicted_mode)
3343                             mode= rem_mode;
3344                         else
3345                             mode= rem_mode + 1;
3346                     }else{
3347                         mode= predicted_mode;
3348                     }
3349
3350                     h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3351                 }
3352                 write_back_intra_pred_mode(h);
3353                 if( check_intra4x4_pred_mode(h) < 0)
3354                     return -1;
3355             }else{
3356                 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3357                 if(h->intra16x16_pred_mode < 0)
3358                     return -1;
3359             }
3360             h->chroma_pred_mode= get_ue_golomb(&s->gb);
3361
3362             h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3363             if(h->chroma_pred_mode < 0)
3364                 return -1;
3365     }else if(partition_count==4){
3366         int i, j, sub_partition_count[4], list, ref[2][4];
3367
3368         if(h->slice_type == B_TYPE){
3369             for(i=0; i<4; i++){
3370                 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3371                 if(h->sub_mb_type[i] >=13){
3372                     fprintf(stderr, "B sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
3373                     return -1;
3374                 }
3375                 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3376                 h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3377             }
3378         }else{
3379             assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3380             for(i=0; i<4; i++){
3381                 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3382                 if(h->sub_mb_type[i] >=4){
3383                     fprintf(stderr, "P sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
3384                     return -1;
3385                 }
3386                 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3387                 h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3388             }
3389         }
3390
3391         for(list=0; list<2; list++){
3392             const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3393             if(ref_count == 0) continue;
3394             for(i=0; i<4; i++){
3395                 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3396                     ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3397                 }else{
3398                  //FIXME
3399                     ref[list][i] = -1;
3400                 }
3401             }
3402         }
3403
3404         for(list=0; list<2; list++){
3405             const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3406             if(ref_count == 0) continue;
3407
3408             for(i=0; i<4; i++){
3409                 h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3410                 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3411
3412                 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3413                     const int sub_mb_type= h->sub_mb_type[i];
3414                     const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3415                     for(j=0; j<sub_partition_count[i]; j++){
3416                         int mx, my;
3417                         const int index= 4*i + block_width*j;
3418                         int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3419                         pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3420                         mx += get_se_golomb(&s->gb);
3421                         my += get_se_golomb(&s->gb);
3422                         tprintf("final mv:%d %d\n", mx, my);
3423
3424                         if(IS_SUB_8X8(sub_mb_type)){
3425                             mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
3426                             mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3427                             mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
3428                             mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3429                         }else if(IS_SUB_8X4(sub_mb_type)){
3430                             mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3431                             mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3432                         }else if(IS_SUB_4X8(sub_mb_type)){
3433                             mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3434                             mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3435                         }else{
3436                             assert(IS_SUB_4X4(sub_mb_type));
3437                             mv_cache[ 0 ][0]= mx;
3438                             mv_cache[ 0 ][1]= my;
3439                         }
3440                     }
3441                 }else{
3442                     uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3443                     p[0] = p[1]=
3444                     p[8] = p[9]= 0;
3445                 }
3446             }
3447         }
3448     }else if(!IS_DIRECT(mb_type)){
3449         int list, mx, my, i;
3450          //FIXME we should set ref_idx_l? to 0 if we use that later ...
3451         if(IS_16X16(mb_type)){
3452             for(list=0; list<2; list++){
3453                 if(h->ref_count[0]>0){
3454                     if(IS_DIR(mb_type, 0, list)){
3455                         const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3456                         fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3457                     }
3458                 }
3459             }
3460             for(list=0; list<2; list++){
3461                 if(IS_DIR(mb_type, 0, list)){
3462                     pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3463                     mx += get_se_golomb(&s->gb);
3464                     my += get_se_golomb(&s->gb);
3465                     tprintf("final mv:%d %d\n", mx, my);
3466
3467                     fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
3468                 }
3469             }
3470         }
3471         else if(IS_16X8(mb_type)){
3472             for(list=0; list<2; list++){
3473                 if(h->ref_count[list]>0){
3474                     for(i=0; i<2; i++){
3475                         if(IS_DIR(mb_type, i, list)){
3476                             const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3477                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3478                         }
3479                     }
3480                 }
3481             }
3482             for(list=0; list<2; list++){
3483                 for(i=0; i<2; i++){
3484                     if(IS_DIR(mb_type, i, list)){
3485                         pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3486                         mx += get_se_golomb(&s->gb);
3487                         my += get_se_golomb(&s->gb);
3488                         tprintf("final mv:%d %d\n", mx, my);
3489
3490                         fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
3491                     }
3492                 }
3493             }
3494         }else{
3495             assert(IS_8X16(mb_type));
3496             for(list=0; list<2; list++){
3497                 if(h->ref_count[list]>0){
3498                     for(i=0; i<2; i++){
3499                         if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3500                             const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3501                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3502                         }
3503                     }
3504                 }
3505             }
3506             for(list=0; list<2; list++){
3507                 for(i=0; i<2; i++){
3508                     if(IS_DIR(mb_type, i, list)){
3509                         pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3510                         mx += get_se_golomb(&s->gb);
3511                         my += get_se_golomb(&s->gb);
3512                         tprintf("final mv:%d %d\n", mx, my);
3513
3514                         fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
3515                     }
3516                 }
3517             }
3518         }
3519     }
3520
3521     if(IS_INTER(mb_type))
3522         write_back_motion(h, mb_type);
3523
3524     if(!IS_INTRA16x16(mb_type)){
3525         cbp= get_ue_golomb(&s->gb);
3526         if(cbp > 47){
3527             fprintf(stderr, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3528             return -1;
3529         }
3530
3531         if(IS_INTRA4x4(mb_type))
3532             cbp= golomb_to_intra4x4_cbp[cbp];
3533         else
3534             cbp= golomb_to_inter_cbp[cbp];
3535     }
3536
3537     if(cbp || IS_INTRA16x16(mb_type)){
3538         int i8x8, i4x4, chroma_idx;
3539         int chroma_qp, dquant;
3540         GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3541         const uint8_t *scan, *dc_scan;
3542
3543 //        fill_non_zero_count_cache(h);
3544
3545         if(IS_INTERLACED(mb_type)){
3546             scan= field_scan;
3547             dc_scan= luma_dc_field_scan;
3548         }else{
3549             scan= zigzag_scan;
3550             dc_scan= luma_dc_zigzag_scan;
3551         }
3552
3553         dquant= get_se_golomb(&s->gb);
3554
3555         if( dquant > 25 || dquant < -26 ){
3556             fprintf(stderr, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3557             return -1;
3558         }
3559
3560         s->qscale += dquant;
3561         if(((unsigned)s->qscale) > 51){
3562             if(s->qscale<0) s->qscale+= 52;
3563             else            s->qscale-= 52;
3564         }
3565
3566         h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3567         if(IS_INTRA16x16(mb_type)){
3568             if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3569                 return -1; //FIXME continue if partotioned and other retirn -1 too
3570             }
3571
3572             assert((cbp&15) == 0 || (cbp&15) == 15);
3573
3574             if(cbp&15){
3575                 for(i8x8=0; i8x8<4; i8x8++){
3576                     for(i4x4=0; i4x4<4; i4x4++){
3577                         const int index= i4x4 + 4*i8x8;
3578                         if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3579                             return -1;
3580                         }
3581                     }
3582                 }
3583             }else{
3584                 memset(&h->non_zero_count_cache[8], 0, 8*4); //FIXME stupid & slow
3585             }
3586         }else{
3587             for(i8x8=0; i8x8<4; i8x8++){
3588                 if(cbp & (1<<i8x8)){
3589                     for(i4x4=0; i4x4<4; i4x4++){
3590                         const int index= i4x4 + 4*i8x8;
3591
3592                         if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3593                             return -1;
3594                         }
3595                     }
3596                 }else{
3597                     uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3598                     nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3599                 }
3600             }
3601         }
3602
3603         if(cbp&0x30){
3604             for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3605                 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3606                     return -1;
3607                 }
3608         }
3609
3610         if(cbp&0x20){
3611             for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3612                 for(i4x4=0; i4x4<4; i4x4++){
3613                     const int index= 16 + 4*chroma_idx + i4x4;
3614                     if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3615                         return -1;
3616                     }
3617                 }
3618             }
3619         }else{
3620             uint8_t * const nnz= &h->non_zero_count_cache[0];
3621             nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3622             nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3623         }
3624     }else{
3625         memset(&h->non_zero_count_cache[8], 0, 8*5);
3626     }
3627     write_back_non_zero_count(h);
3628
3629     return 0;
3630 }
3631
3632 static int decode_slice(H264Context *h){
3633     MpegEncContext * const s = &h->s;
3634     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3635
3636     s->mb_skip_run= -1;
3637
3638 #if 1
3639     for(;;){
3640         int ret= decode_mb(h);
3641
3642         hl_decode_mb(h);
3643
3644         if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
3645             s->mb_y++;
3646             ret= decode_mb(h);
3647
3648             hl_decode_mb(h);
3649             s->mb_y--;
3650         }
3651
3652         if(ret<0){
3653             fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3654             ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3655
3656             return -1;
3657         }
3658
3659         if(++s->mb_x >= s->mb_width){
3660             s->mb_x=0;
3661             ff_draw_horiz_band(s, 16*s->mb_y, 16);
3662             if(++s->mb_y >= s->mb_height){
3663                 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3664
3665                 if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3666                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3667
3668                     return 0;
3669                 }else{
3670                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3671
3672                     return -1;
3673                 }
3674             }
3675         }
3676
3677         if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3678             if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3679                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3680
3681                 return 0;
3682             }else{
3683                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3684
3685                 return -1;
3686             }
3687         }
3688     }
3689 #endif
3690 #if 0
3691     for(;s->mb_y < s->mb_height; s->mb_y++){
3692         for(;s->mb_x < s->mb_width; s->mb_x++){
3693             int ret= decode_mb(h);
3694
3695             hl_decode_mb(h);
3696
3697             if(ret<0){
3698                 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3699                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3700
3701                 return -1;
3702             }
3703
3704             if(++s->mb_x >= s->mb_width){
3705                 s->mb_x=0;
3706                 if(++s->mb_y >= s->mb_height){
3707                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
3708                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3709
3710                         return 0;
3711                     }else{
3712                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3713
3714                         return -1;
3715                     }
3716                 }
3717             }
3718
3719             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
3720                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
3721                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3722
3723                     return 0;
3724                 }else{
3725                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3726
3727                     return -1;
3728                 }
3729             }
3730         }
3731         s->mb_x=0;
3732         ff_draw_horiz_band(s, 16*s->mb_y, 16);
3733     }
3734 #endif
3735     return -1; //not reached
3736 }
3737
3738 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
3739     MpegEncContext * const s = &h->s;
3740     int aspect_ratio_info_present_flag, aspect_ratio_idc;
3741
3742     aspect_ratio_info_present_flag= get_bits1(&s->gb);
3743
3744     if( aspect_ratio_info_present_flag ) {
3745         aspect_ratio_idc= get_bits(&s->gb, 8);
3746         if( aspect_ratio_idc == EXTENDED_SAR ) {
3747             sps->sar_width= get_bits(&s->gb, 16);
3748             sps->sar_height= get_bits(&s->gb, 16);
3749         }else if(aspect_ratio_idc < 16){
3750             sps->sar_width=  pixel_aspect[aspect_ratio_idc][0];
3751             sps->sar_height= pixel_aspect[aspect_ratio_idc][1];
3752         }else{
3753             fprintf(stderr, "illegal aspect ratio\n");
3754             return -1;
3755         }
3756     }else{
3757         sps->sar_width=
3758         sps->sar_height= 0;
3759     }
3760 //            s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
3761 #if 0
3762 | overscan_info_present_flag                        |0  |u(1)    |
3763 | if( overscan_info_present_flag )                  |   |        |
3764 |  overscan_appropriate_flag                        |0  |u(1)    |
3765 | video_signal_type_present_flag                    |0  |u(1)    |
3766 | if( video_signal_type_present_flag ) {            |   |        |
3767 |  video_format                                     |0  |u(3)    |
3768 |  video_full_range_flag                            |0  |u(1)    |
3769 |  colour_description_present_flag                  |0  |u(1)    |
3770 |  if( colour_description_present_flag ) {          |   |        |
3771 |   colour_primaries                                |0  |u(8)    |
3772 |   transfer_characteristics                        |0  |u(8)    |
3773 |   matrix_coefficients                             |0  |u(8)    |
3774 |  }                                                |   |        |
3775 | }                                                 |   |        |
3776 | chroma_location_info_present_flag                 |0  |u(1)    |
3777 | if ( chroma_location_info_present_flag ) {        |   |        |
3778 |  chroma_sample_location_type_top_field            |0  |ue(v)   |
3779 |  chroma_sample_location_type_bottom_field         |0  |ue(v)   |
3780 | }                                                 |   |        |
3781 | timing_info_present_flag                          |0  |u(1)    |
3782 | if( timing_info_present_flag ) {                  |   |        |
3783 |  num_units_in_tick                                |0  |u(32)   |
3784 |  time_scale                                       |0  |u(32)   |
3785 |  fixed_frame_rate_flag                            |0  |u(1)    |
3786 | }                                                 |   |        |
3787 | nal_hrd_parameters_present_flag                   |0  |u(1)    |
3788 | if( nal_hrd_parameters_present_flag  = =  1)      |   |        |
3789 |  hrd_parameters( )                                |   |        |
3790 | vcl_hrd_parameters_present_flag                   |0  |u(1)    |
3791 | if( vcl_hrd_parameters_present_flag  = =  1)      |   |        |
3792 |  hrd_parameters( )                                |   |        |
3793 | if( ( nal_hrd_parameters_present_flag  = =  1  | ||   |        |
3794 |                                                   |   |        |
3795 |( vcl_hrd_parameters_present_flag  = =  1 ) )      |   |        |
3796 |  low_delay_hrd_flag                               |0  |u(1)    |
3797 | bitstream_restriction_flag                        |0  |u(1)    |
3798 | if( bitstream_restriction_flag ) {                |0  |u(1)    |
3799 |  motion_vectors_over_pic_boundaries_flag          |0  |u(1)    |
3800 |  max_bytes_per_pic_denom                          |0  |ue(v)   |
3801 |  max_bits_per_mb_denom                            |0  |ue(v)   |
3802 |  log2_max_mv_length_horizontal                    |0  |ue(v)   |
3803 |  log2_max_mv_length_vertical                      |0  |ue(v)   |
3804 |  num_reorder_frames                               |0  |ue(v)   |
3805 |  max_dec_frame_buffering                          |0  |ue(v)   |
3806 | }                                                 |   |        |
3807 |}                                                  |   |        |
3808 #endif
3809     return 0;
3810 }
3811
3812 static inline int decode_seq_parameter_set(H264Context *h){
3813     MpegEncContext * const s = &h->s;
3814     int profile_idc, level_idc, multiple_slice_groups, arbitrary_slice_order, redundant_slices;
3815     int sps_id, i;
3816     SPS *sps;
3817
3818     profile_idc= get_bits(&s->gb, 8);
3819     level_idc= get_bits(&s->gb, 8);
3820     multiple_slice_groups= get_bits1(&s->gb);
3821     arbitrary_slice_order= get_bits1(&s->gb);
3822     redundant_slices= get_bits1(&s->gb);
3823
3824     sps_id= get_ue_golomb(&s->gb);
3825
3826     sps= &h->sps_buffer[ sps_id ];
3827
3828     sps->profile_idc= profile_idc;
3829     sps->level_idc= level_idc;
3830     sps->multiple_slice_groups= multiple_slice_groups;
3831     sps->arbitrary_slice_order= arbitrary_slice_order;
3832     sps->redundant_slices= redundant_slices;
3833
3834     sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
3835
3836     sps->poc_type= get_ue_golomb(&s->gb);
3837
3838     if(sps->poc_type == 0){ //FIXME #define
3839         sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
3840     } else if(sps->poc_type == 1){//FIXME #define
3841         sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
3842         sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
3843         sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
3844         sps->poc_cycle_length= get_ue_golomb(&s->gb);
3845
3846         for(i=0; i<sps->poc_cycle_length; i++)
3847             sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
3848     }
3849     if(sps->poc_type > 2){
3850         fprintf(stderr, "illegal POC type %d\n", sps->poc_type);
3851         return -1;
3852     }
3853
3854     sps->ref_frame_count= get_ue_golomb(&s->gb);
3855     sps->required_frame_num_update_behaviour_flag= get_bits1(&s->gb);
3856     sps->mb_width= get_ue_golomb(&s->gb) + 1;
3857     sps->mb_height= get_ue_golomb(&s->gb) + 1;
3858     sps->frame_mbs_only_flag= get_bits1(&s->gb);
3859     if(!sps->frame_mbs_only_flag)
3860         sps->mb_aff= get_bits1(&s->gb);
3861     else
3862         sps->mb_aff= 0;
3863
3864     sps->direct_8x8_inference_flag= get_bits1(&s->gb);
3865
3866     sps->vui_parameters_present_flag= get_bits1(&s->gb);
3867     if( sps->vui_parameters_present_flag )
3868         decode_vui_parameters(h, sps);
3869
3870     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3871         printf("sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s %s\n",
3872                sps_id, sps->profile_idc, sps->level_idc,
3873                sps->poc_type,
3874                sps->ref_frame_count,
3875                sps->mb_width, sps->mb_height,
3876                sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
3877                sps->direct_8x8_inference_flag ? "8B8" : "",
3878                sps->vui_parameters_present_flag ? "VUI" : ""
3879                );
3880     }
3881     return 0;
3882 }
3883
3884 static inline int decode_picture_parameter_set(H264Context *h){
3885     MpegEncContext * const s = &h->s;
3886     int pps_id= get_ue_golomb(&s->gb);
3887     PPS *pps= &h->pps_buffer[pps_id];
3888
3889     pps->sps_id= get_ue_golomb(&s->gb);
3890     pps->cabac= get_bits1(&s->gb);
3891     pps->pic_order_present= get_bits1(&s->gb);
3892     pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
3893     if(pps->slice_group_count > 1 ){
3894         pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
3895 fprintf(stderr, "FMO not supported\n");
3896         switch(pps->mb_slice_group_map_type){
3897         case 0:
3898 #if 0
3899 |   for( i = 0; i <= num_slice_groups_minus1; i++ ) |   |        |
3900 |    run_length[ i ]                                |1  |ue(v)   |
3901 #endif
3902             break;
3903         case 2:
3904 #if 0
3905 |   for( i = 0; i < num_slice_groups_minus1; i++ )  |   |        |
3906 |{                                                  |   |        |
3907 |    top_left_mb[ i ]                               |1  |ue(v)   |
3908 |    bottom_right_mb[ i ]                           |1  |ue(v)   |
3909 |   }                                               |   |        |
3910 #endif
3911             break;
3912         case 3:
3913         case 4:
3914         case 5:
3915 #if 0
3916 |   slice_group_change_direction_flag               |1  |u(1)    |
3917 |   slice_group_change_rate_minus1                  |1  |ue(v)   |
3918 #endif
3919             break;
3920         case 6:
3921 #if 0
3922 |   slice_group_id_cnt_minus1                       |1  |ue(v)   |
3923 |   for( i = 0; i <= slice_group_id_cnt_minus1; i++ |   |        |
3924 |)                                                  |   |        |
3925 |    slice_group_id[ i ]                            |1  |u(v)    |
3926 #endif
3927             break;
3928         }
3929     }
3930     pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3931     pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3932     if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
3933         fprintf(stderr, "reference overflow (pps)\n");
3934         return -1;
3935     }
3936
3937     pps->weighted_pred= get_bits1(&s->gb);
3938     pps->weighted_bipred_idc= get_bits(&s->gb, 2);
3939     pps->init_qp= get_se_golomb(&s->gb) + 26;
3940     pps->init_qs= get_se_golomb(&s->gb) + 26;
3941     pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
3942     pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
3943     pps->constrained_intra_pred= get_bits1(&s->gb);
3944     pps->redundant_pic_cnt_present = get_bits1(&s->gb);
3945     pps->crop= get_bits1(&s->gb);
3946     if(pps->crop){
3947         pps->crop_left  = get_ue_golomb(&s->gb);
3948         pps->crop_right = get_ue_golomb(&s->gb);
3949         pps->crop_top   = get_ue_golomb(&s->gb);
3950         pps->crop_bottom= get_ue_golomb(&s->gb);
3951     }else{
3952         pps->crop_left  =
3953         pps->crop_right =
3954         pps->crop_top   =
3955         pps->crop_bottom= 0;
3956     }
3957
3958     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3959         printf("pps:%d sps:%d %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s crop:%d/%d/%d/%d\n",
3960                pps_id, pps->sps_id,
3961                pps->cabac ? "CABAC" : "CAVLC",
3962                pps->slice_group_count,
3963                pps->ref_count[0], pps->ref_count[1],
3964                pps->weighted_pred ? "weighted" : "",
3965                pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
3966                pps->deblocking_filter_parameters_present ? "LPAR" : "",
3967                pps->constrained_intra_pred ? "CONSTR" : "",
3968                pps->redundant_pic_cnt_present ? "REDU" : "",
3969                pps->crop_left, pps->crop_right,
3970                pps->crop_top, pps->crop_bottom
3971                );
3972     }
3973
3974     return 0;
3975 }
3976
3977 /**
3978  * finds the end of the current frame in the bitstream.
3979  * @return the position of the first byte of the next frame, or -1
3980  */
3981 static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){
3982     ParseContext *pc= &s->parse_context;
3983     int i;
3984     uint32_t state;
3985 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
3986 //    mb_addr= pc->mb_addr - 1;
3987     state= pc->state;
3988     //FIXME this will fail with slices
3989     for(i=0; i<buf_size; i++){
3990         state= (state<<8) | buf[i];
3991         if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
3992             if(pc->frame_start_found){
3993                 pc->state=-1;
3994                 pc->frame_start_found= 0;
3995                 return i-3;
3996             }
3997             pc->frame_start_found= 1;
3998         }
3999     }
4000
4001     pc->state= state;
4002     return END_NOT_FOUND;
4003 }
4004
4005 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
4006     MpegEncContext * const s = &h->s;
4007     AVCodecContext * const avctx= s->avctx;
4008     int buf_index=0;
4009     int i;
4010 #if 0
4011     for(i=0; i<32; i++){
4012         printf("%X ", buf[i]);
4013     }
4014 #endif
4015     for(;;){
4016         int consumed;
4017         int dst_length;
4018         int bit_length;
4019         uint8_t *ptr;
4020
4021         // start code prefix search
4022         for(; buf_index + 3 < buf_size; buf_index++){
4023             // this should allways succeed in the first iteration
4024             if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
4025                 break;
4026         }
4027
4028         if(buf_index+3 >= buf_size) break;
4029
4030         buf_index+=3;
4031
4032         ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index);
4033         if(ptr[dst_length - 1] == 0) dst_length--;
4034         bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
4035
4036         if(s->avctx->debug&FF_DEBUG_STARTCODE){
4037             printf("NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
4038         }
4039
4040         buf_index += consumed;
4041
4042         if(h->nal_ref_idc < s->hurry_up)
4043             continue;
4044
4045         switch(h->nal_unit_type){
4046         case NAL_IDR_SLICE:
4047             idr(h); //FIXME ensure we dont loose some frames if there is reordering
4048         case NAL_SLICE:
4049             init_get_bits(&s->gb, ptr, bit_length);
4050             h->intra_gb_ptr=
4051             h->inter_gb_ptr= &s->gb;
4052             s->data_partitioning = 0;
4053
4054             if(decode_slice_header(h) < 0) return -1;
4055             if(h->redundant_pic_count==0)
4056                 decode_slice(h);
4057             break;
4058         case NAL_DPA:
4059             init_get_bits(&s->gb, ptr, bit_length);
4060             h->intra_gb_ptr=
4061             h->inter_gb_ptr= NULL;
4062             s->data_partitioning = 1;
4063
4064             if(decode_slice_header(h) < 0) return -1;
4065             break;
4066         case NAL_DPB:
4067             init_get_bits(&h->intra_gb, ptr, bit_length);
4068             h->intra_gb_ptr= &h->intra_gb;
4069             break;
4070         case NAL_DPC:
4071             init_get_bits(&h->inter_gb, ptr, bit_length);
4072             h->inter_gb_ptr= &h->inter_gb;
4073
4074             if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning)
4075                 decode_slice(h);
4076             break;
4077         case NAL_SEI:
4078             break;
4079         case NAL_SPS:
4080             init_get_bits(&s->gb, ptr, bit_length);
4081             decode_seq_parameter_set(h);
4082
4083             if(s->flags& CODEC_FLAG_LOW_DELAY)
4084                 s->low_delay=1;
4085
4086             avctx->has_b_frames= !s->low_delay;
4087             break;
4088         case NAL_PPS:
4089             init_get_bits(&s->gb, ptr, bit_length);
4090
4091             decode_picture_parameter_set(h);
4092
4093             break;
4094         case NAL_PICTURE_DELIMITER:
4095             break;
4096         case NAL_FILTER_DATA:
4097             break;
4098         }
4099
4100         //FIXME move after where irt is set
4101         s->current_picture.pict_type= s->pict_type;
4102         s->current_picture.key_frame= s->pict_type == I_TYPE;
4103     }
4104
4105     if(!s->current_picture_ptr) return buf_index; //no frame
4106
4107     h->prev_frame_num_offset= h->frame_num_offset;
4108     h->prev_frame_num= h->frame_num;
4109     if(s->current_picture_ptr->reference){
4110         h->prev_poc_msb= h->poc_msb;
4111         h->prev_poc_lsb= h->poc_lsb;
4112     }
4113     if(s->current_picture_ptr->reference)
4114         execute_ref_pic_marking(h, h->mmco, h->mmco_index);
4115     else
4116         assert(h->mmco_index==0);
4117
4118     ff_er_frame_end(s);
4119     MPV_frame_end(s);
4120
4121     return buf_index;
4122 }
4123
4124 /**
4125  * retunrs the number of bytes consumed for building the current frame
4126  */
4127 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
4128     if(s->flags&CODEC_FLAG_TRUNCATED){
4129         pos -= s->parse_context.last_index;
4130         if(pos<0) pos=0; // FIXME remove (uneeded?)
4131
4132         return pos;
4133     }else{
4134         if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
4135         if(pos+10>buf_size) pos=buf_size; // oops ;)
4136
4137         return pos;
4138     }
4139 }
4140
4141 static int decode_frame(AVCodecContext *avctx,
4142                              void *data, int *data_size,
4143                              uint8_t *buf, int buf_size)
4144 {
4145     H264Context *h = avctx->priv_data;
4146     MpegEncContext *s = &h->s;
4147     AVFrame *pict = data;
4148     int buf_index;
4149
4150     s->flags= avctx->flags;
4151
4152     *data_size = 0;
4153
4154    /* no supplementary picture */
4155     if (buf_size == 0) {
4156         return 0;
4157     }
4158
4159     if(s->flags&CODEC_FLAG_TRUNCATED){
4160         int next= find_frame_end(s, buf, buf_size);
4161
4162         if( ff_combine_frame(s, next, &buf, &buf_size) < 0 )
4163             return buf_size;
4164 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
4165     }
4166
4167     if(s->avctx->extradata_size && s->picture_number==0){
4168         if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
4169             return -1;
4170     }
4171
4172     buf_index=decode_nal_units(h, buf, buf_size);
4173     if(buf_index < 0)
4174         return -1;
4175
4176     //FIXME do something with unavailable reference frames
4177
4178 //    if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
4179 #if 0
4180     if(s->pict_type==B_TYPE || s->low_delay){
4181         *pict= *(AVFrame*)&s->current_picture;
4182     } else {
4183         *pict= *(AVFrame*)&s->last_picture;
4184     }
4185 #endif
4186     if(!s->current_picture_ptr){
4187         fprintf(stderr, "error, NO frame\n");
4188         return -1;
4189     }
4190
4191     *pict= *(AVFrame*)&s->current_picture; //FIXME
4192     ff_print_debug_info(s, s->current_picture_ptr);
4193     assert(pict->data[0]);
4194 //printf("out %d\n", (int)pict->data[0]);
4195 #if 0 //?
4196
4197     /* Return the Picture timestamp as the frame number */
4198     /* we substract 1 because it is added on utils.c    */
4199     avctx->frame_number = s->picture_number - 1;
4200 #endif
4201 #if 0
4202     /* dont output the last pic after seeking */
4203     if(s->last_picture_ptr || s->low_delay)
4204     //Note this isnt a issue as a IDR pic should flush teh buffers
4205 #endif
4206         *data_size = sizeof(AVFrame);
4207     return get_consumed_bytes(s, buf_index, buf_size);
4208 }
4209 #if 0
4210 static inline void fill_mb_avail(H264Context *h){
4211     MpegEncContext * const s = &h->s;
4212     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4213
4214     if(s->mb_y){
4215         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
4216         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
4217         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
4218     }else{
4219         h->mb_avail[0]=
4220         h->mb_avail[1]=
4221         h->mb_avail[2]= 0;
4222     }
4223     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
4224     h->mb_avail[4]= 1; //FIXME move out
4225     h->mb_avail[5]= 0; //FIXME move out
4226 }
4227 #endif
4228
4229 #if 0 //selftest
4230 #define COUNT 8000
4231 #define SIZE (COUNT*40)
4232 int main(){
4233     int i;
4234     uint8_t temp[SIZE];
4235     PutBitContext pb;
4236     GetBitContext gb;
4237 //    int int_temp[10000];
4238     DSPContext dsp;
4239     AVCodecContext avctx;
4240
4241     dsputil_init(&dsp, &avctx);
4242
4243     init_put_bits(&pb, temp, SIZE, NULL, NULL);
4244     printf("testing unsigned exp golomb\n");
4245     for(i=0; i<COUNT; i++){
4246         START_TIMER
4247         set_ue_golomb(&pb, i);
4248         STOP_TIMER("set_ue_golomb");
4249     }
4250     flush_put_bits(&pb);
4251
4252     init_get_bits(&gb, temp, 8*SIZE);
4253     for(i=0; i<COUNT; i++){
4254         int j, s;
4255
4256         s= show_bits(&gb, 24);
4257
4258         START_TIMER
4259         j= get_ue_golomb(&gb);
4260         if(j != i){
4261             printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4262 //            return -1;
4263         }
4264         STOP_TIMER("get_ue_golomb");
4265     }
4266
4267
4268     init_put_bits(&pb, temp, SIZE, NULL, NULL);
4269     printf("testing signed exp golomb\n");
4270     for(i=0; i<COUNT; i++){
4271         START_TIMER
4272         set_se_golomb(&pb, i - COUNT/2);
4273         STOP_TIMER("set_se_golomb");
4274     }
4275     flush_put_bits(&pb);
4276
4277     init_get_bits(&gb, temp, 8*SIZE);
4278     for(i=0; i<COUNT; i++){
4279         int j, s;
4280
4281         s= show_bits(&gb, 24);
4282
4283         START_TIMER
4284         j= get_se_golomb(&gb);
4285         if(j != i - COUNT/2){
4286             printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4287 //            return -1;
4288         }
4289         STOP_TIMER("get_se_golomb");
4290     }
4291
4292     printf("testing 4x4 (I)DCT\n");
4293
4294     DCTELEM block[16];
4295     uint8_t src[16], ref[16];
4296     uint64_t error= 0, max_error=0;
4297
4298     for(i=0; i<COUNT; i++){
4299         int j;
4300 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
4301         for(j=0; j<16; j++){
4302             ref[j]= random()%255;
4303             src[j]= random()%255;
4304         }
4305
4306         h264_diff_dct_c(block, src, ref, 4);
4307
4308         //normalize
4309         for(j=0; j<16; j++){
4310 //            printf("%d ", block[j]);
4311             block[j]= block[j]*4;
4312             if(j&1) block[j]= (block[j]*4 + 2)/5;
4313             if(j&4) block[j]= (block[j]*4 + 2)/5;
4314         }
4315 //        printf("\n");
4316
4317         h264_add_idct_c(ref, block, 4);
4318 /*        for(j=0; j<16; j++){
4319             printf("%d ", ref[j]);
4320         }
4321         printf("\n");*/
4322
4323         for(j=0; j<16; j++){
4324             int diff= ABS(src[j] - ref[j]);
4325
4326             error+= diff*diff;
4327             max_error= FFMAX(max_error, diff);
4328         }
4329     }
4330     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
4331 #if 0
4332     printf("testing quantizer\n");
4333     for(qp=0; qp<52; qp++){
4334         for(i=0; i<16; i++)
4335             src1_block[i]= src2_block[i]= random()%255;
4336
4337     }
4338 #endif
4339     printf("Testing NAL layer\n");
4340
4341     uint8_t bitstream[COUNT];
4342     uint8_t nal[COUNT*2];
4343     H264Context h;
4344     memset(&h, 0, sizeof(H264Context));
4345
4346     for(i=0; i<COUNT; i++){
4347         int zeros= i;
4348         int nal_length;
4349         int consumed;
4350         int out_length;
4351         uint8_t *out;
4352         int j;
4353
4354         for(j=0; j<COUNT; j++){
4355             bitstream[j]= (random() % 255) + 1;
4356         }
4357
4358         for(j=0; j<zeros; j++){
4359             int pos= random() % COUNT;
4360             while(bitstream[pos] == 0){
4361                 pos++;
4362                 pos %= COUNT;
4363             }
4364             bitstream[pos]=0;
4365         }
4366
4367         START_TIMER
4368
4369         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
4370         if(nal_length<0){
4371             printf("encoding failed\n");
4372             return -1;
4373         }
4374
4375         out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
4376
4377         STOP_TIMER("NAL")
4378
4379         if(out_length != COUNT){
4380             printf("incorrect length %d %d\n", out_length, COUNT);
4381             return -1;
4382         }
4383
4384         if(consumed != nal_length){
4385             printf("incorrect consumed length %d %d\n", nal_length, consumed);
4386             return -1;
4387         }
4388
4389         if(memcmp(bitstream, out, COUNT)){
4390             printf("missmatch\n");
4391             return -1;
4392         }
4393     }
4394
4395     printf("Testing RBSP\n");
4396
4397
4398     return 0;
4399 }
4400 #endif
4401
4402
4403 static int decode_end(AVCodecContext *avctx)
4404 {
4405     H264Context *h = avctx->priv_data;
4406     MpegEncContext *s = &h->s;
4407
4408     free_tables(h); //FIXME cleanup init stuff perhaps
4409     MPV_common_end(s);
4410
4411 //    memset(h, 0, sizeof(H264Context));
4412
4413     return 0;
4414 }
4415
4416
4417 AVCodec h264_decoder = {
4418     "h264",
4419     CODEC_TYPE_VIDEO,
4420     CODEC_ID_H264,
4421     sizeof(H264Context),
4422     decode_init,
4423     NULL,
4424     decode_end,
4425     decode_frame,
4426     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,
4427 };
4428
4429 #include "svq3.c"