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 log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
  63     int poc_type;                      ///< pic_order_cnt_type
  64     int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
  65     int delta_pic_order_always_zero_flag;
  66     int offset_for_non_ref_pic;
  67     int offset_for_top_to_bottom_field;
  68     int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
  69     int ref_frame_count;               ///< num_ref_frames
  70     int gaps_in_frame_num_allowed_flag;
  71     int mb_width;                      ///< frame_width_in_mbs_minus1 + 1
  72     int mb_height;                     ///< frame_height_in_mbs_minus1 + 1
  73     int frame_mbs_only_flag;
  74     int mb_aff;                        ///<mb_adaptive_frame_field_flag
  75     int direct_8x8_inference_flag;
  76     int crop;                   ///< frame_cropping_flag
  77     int crop_left;              ///< frame_cropping_rect_left_offset
  78     int crop_right;             ///< frame_cropping_rect_right_offset
  79     int crop_top;               ///< frame_cropping_rect_top_offset
  80     int crop_bottom;            ///< frame_cropping_rect_bottom_offset
  81     int vui_parameters_present_flag;
  82     AVRational sar;
  83     short offset_for_ref_frame[256]; //FIXME dyn aloc?
  84 }SPS;
  85
  86 /**
  87  * Picture parameter set
  88  */
  89 typedef struct PPS{
  90     int sps_id;
  91     int cabac;                  ///< entropy_coding_mode_flag
  92     int pic_order_present;      ///< pic_order_present_flag
  93     int slice_group_count;      ///< num_slice_groups_minus1 + 1
  94     int mb_slice_group_map_type;
  95     int ref_count[2];           ///< num_ref_idx_l0/1_active_minus1 + 1
  96     int weighted_pred;          ///< weighted_pred_flag
  97     int weighted_bipred_idc;
  98     int init_qp;                ///< pic_init_qp_minus26 + 26
  99     int init_qs;                ///< pic_init_qs_minus26 + 26
 100     int chroma_qp_index_offset;
 101     int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
 102     int constrained_intra_pred; ///< constrained_intra_pred_flag
 103     int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
 104 }PPS;
 105
 106 /**
 107  * Memory management control operation opcode.
 108  */
 109 typedef enum MMCOOpcode{
 110     MMCO_END=0,
 111     MMCO_SHORT2UNUSED,
 112     MMCO_LONG2UNUSED,
 113     MMCO_SHORT2LONG,
 114     MMCO_SET_MAX_LONG,
 115     MMCO_RESET,
 116     MMCO_LONG,
 117 } MMCOOpcode;
 118
 119 /**
 120  * Memory management control operation.
 121  */
 122 typedef struct MMCO{
 123     MMCOOpcode opcode;
 124     int short_frame_num;
 125     int long_index;
 126 } MMCO;
 127
 128 /**
 129  * H264Context
 130  */
 131 typedef struct H264Context{
 132     MpegEncContext s;
 133     int nal_ref_idc;
 134     int nal_unit_type;
 135 #define NAL_SLICE               1
 136 #define NAL_DPA                 2
 137 #define NAL_DPB                 3
 138 #define NAL_DPC                 4
 139 #define NAL_IDR_SLICE           5
 140 #define NAL_SEI                 6
 141 #define NAL_SPS                 7
 142 #define NAL_PPS                 8
 143 #define NAL_PICTURE_DELIMITER   9
 144 #define NAL_FILTER_DATA         10
 145     uint8_t *rbsp_buffer;
 146     int rbsp_buffer_size;
 147
 148     int chroma_qp; //QPc
 149
 150     int prev_mb_skiped; //FIXME remove (IMHO not used)
 151
 152     //prediction stuff
 153     int chroma_pred_mode;
 154     int intra16x16_pred_mode;
 155
 156     int8_t intra4x4_pred_mode_cache[5*8];
 157     int8_t (*intra4x4_pred_mode)[8];
 158     void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
 159     void (*pred8x8  [4+3])(uint8_t *src, int stride);
 160     void (*pred16x16[4+3])(uint8_t *src, int stride);
 161     unsigned int topleft_samples_available;
 162     unsigned int top_samples_available;
 163     unsigned int topright_samples_available;
 164     unsigned int left_samples_available;
 165
 166     /**
 167      * non zero coeff count cache.
 168      * is 64 if not available.
 169      */
 170     uint8_t non_zero_count_cache[6*8];
 171     uint8_t (*non_zero_count)[16];
 172
 173     /**
 174      * Motion vector cache.
 175      */
 176     int16_t mv_cache[2][5*8][2];
 177     int8_t ref_cache[2][5*8];
 178 #define LIST_NOT_USED -1 //FIXME rename?
 179 #define PART_NOT_AVAILABLE -2
 180
 181     /**
 182      * is 1 if the specific list MV&references are set to 0,0,-2.
 183      */
 184     int mv_cache_clean[2];
 185
 186     int block_offset[16+8];
 187     int chroma_subblock_offset[16]; //FIXME remove
 188
 189     uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
 190     uint16_t *mb2b8_xy;
 191     int b_stride;
 192     int b8_stride;
 193
 194     int halfpel_flag;
 195     int thirdpel_flag;
 196
 197     int unknown_svq3_flag;
 198     int next_slice_index;
 199
 200     SPS sps_buffer[MAX_SPS_COUNT];
 201     SPS sps; ///< current sps
 202
 203     PPS pps_buffer[MAX_PPS_COUNT];
 204     /**
 205      * current pps
 206      */
 207     PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
 208
 209     int slice_num;
 210     uint8_t *slice_table_base;
 211     uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
 212     int slice_type;
 213     int slice_type_fixed;
 214
 215     //interlacing specific flags
 216     int mb_field_decoding_flag;
 217
 218     int sub_mb_type[4];
 219
 220     //POC stuff
 221     int poc_lsb;
 222     int poc_msb;
 223     int delta_poc_bottom;
 224     int delta_poc[2];
 225     int frame_num;
 226     int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
 227     int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
 228     int frame_num_offset;         ///< for POC type 2
 229     int prev_frame_num_offset;    ///< for POC type 2
 230     int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
 231
 232     /**
 233      * frame_num for frames or 2*frame_num for field pics.
 234      */
 235     int curr_pic_num;
 236
 237     /**
 238      * max_frame_num or 2*max_frame_num for field pics.
 239      */
 240     int max_pic_num;
 241
 242     //Weighted pred stuff
 243     int luma_log2_weight_denom;
 244     int chroma_log2_weight_denom;
 245     int luma_weight[2][16];
 246     int luma_offset[2][16];
 247     int chroma_weight[2][16][2];
 248     int chroma_offset[2][16][2];
 249
 250     //deblock
 251     int disable_deblocking_filter_idc;
 252     int slice_alpha_c0_offset_div2;
 253     int slice_beta_offset_div2;
 254
 255     int redundant_pic_count;
 256
 257     int direct_spatial_mv_pred;
 258
 259     /**
 260      * num_ref_idx_l0/1_active_minus1 + 1
 261      */
 262     int ref_count[2];// FIXME split for AFF
 263     Picture *short_ref[16];
 264     Picture *long_ref[16];
 265     Picture default_ref_list[2][32];
 266     Picture ref_list[2][32]; //FIXME size?
 267     Picture field_ref_list[2][32]; //FIXME size?
 268
 269     /**
 270      * memory management control operations buffer.
 271      */
 272     MMCO mmco[MAX_MMCO_COUNT];
 273     int mmco_index;
 274
 275     int long_ref_count;  ///< number of actual long term references
 276     int short_ref_count; ///< number of actual short term references
 277
 278     //data partitioning
 279     GetBitContext intra_gb;
 280     GetBitContext inter_gb;
 281     GetBitContext *intra_gb_ptr;
 282     GetBitContext *inter_gb_ptr;
 283
 284     DCTELEM mb[16*24] __align8;
 285 }H264Context;
 286
 287 static VLC coeff_token_vlc[4];
 288 static VLC chroma_dc_coeff_token_vlc;
 289
 290 static VLC total_zeros_vlc[15];
 291 static VLC chroma_dc_total_zeros_vlc[3];
 292
 293 static VLC run_vlc[6];
 294 static VLC run7_vlc;
 295
 296 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
 297 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
 298
 299 static inline uint32_t pack16to32(int a, int b){
 300 #ifdef WORDS_BIGENDIAN
 301    return (b&0xFFFF) + (a<<16);
 302 #else
 303    return (a&0xFFFF) + (b<<16);
 304 #endif
 305 }
 306
 307 /**
 308  * fill a rectangle.
 309  * @param h height of the recatangle, should be a constant
 310  * @param w width of the recatangle, should be a constant
 311  * @param size the size of val (1 or 4), should be a constant
 312  */
 313 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
 314     uint8_t *p= (uint8_t*)vp;
 315     assert(size==1 || size==4);
 316
 317     w      *= size;
 318     stride *= size;
 319
 320 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
 321     if(w==2 && h==2){
 322         *(uint16_t*)(p + 0)=
 323         *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
 324     }else if(w==2 && h==4){
 325         *(uint16_t*)(p + 0*stride)=
 326         *(uint16_t*)(p + 1*stride)=
 327         *(uint16_t*)(p + 2*stride)=
 328         *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
 329     }else if(w==4 && h==1){
 330         *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
 331     }else if(w==4 && h==2){
 332         *(uint32_t*)(p + 0*stride)=
 333         *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
 334     }else if(w==4 && h==4){
 335         *(uint32_t*)(p + 0*stride)=
 336         *(uint32_t*)(p + 1*stride)=
 337         *(uint32_t*)(p + 2*stride)=
 338         *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
 339     }else if(w==8 && h==1){
 340         *(uint32_t*)(p + 0)=
 341         *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
 342     }else if(w==8 && h==2){
 343         *(uint32_t*)(p + 0 + 0*stride)=
 344         *(uint32_t*)(p + 4 + 0*stride)=
 345         *(uint32_t*)(p + 0 + 1*stride)=
 346         *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
 347     }else if(w==8 && h==4){
 348         *(uint64_t*)(p + 0*stride)=
 349         *(uint64_t*)(p + 1*stride)=
 350         *(uint64_t*)(p + 2*stride)=
 351         *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
 352     }else if(w==16 && h==2){
 353         *(uint64_t*)(p + 0+0*stride)=
 354         *(uint64_t*)(p + 8+0*stride)=
 355         *(uint64_t*)(p + 0+1*stride)=
 356         *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
 357     }else if(w==16 && h==4){
 358         *(uint64_t*)(p + 0+0*stride)=
 359         *(uint64_t*)(p + 8+0*stride)=
 360         *(uint64_t*)(p + 0+1*stride)=
 361         *(uint64_t*)(p + 8+1*stride)=
 362         *(uint64_t*)(p + 0+2*stride)=
 363         *(uint64_t*)(p + 8+2*stride)=
 364         *(uint64_t*)(p + 0+3*stride)=
 365         *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
 366     }else
 367         assert(0);
 368 }
 369
 370 static inline void fill_caches(H264Context *h, int mb_type){
 371     MpegEncContext * const s = &h->s;
 372     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
 373     int topleft_xy, top_xy, topright_xy, left_xy[2];
 374     int topleft_type, top_type, topright_type, left_type[2];
 375     int left_block[4];
 376     int i;
 377
 378     //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
 379
 380     if(h->sps.mb_aff){
 381     //FIXME
 382         topleft_xy = 0; /* avoid warning */
 383         top_xy = 0; /* avoid warning */
 384         topright_xy = 0; /* avoid warning */
 385     }else{
 386         topleft_xy = mb_xy-1 - s->mb_stride;
 387         top_xy     = mb_xy   - s->mb_stride;
 388         topright_xy= mb_xy+1 - s->mb_stride;
 389         left_xy[0]   = mb_xy-1;
 390         left_xy[1]   = mb_xy-1;
 391         left_block[0]= 0;
 392         left_block[1]= 1;
 393         left_block[2]= 2;
 394         left_block[3]= 3;
 395     }
 396
 397     topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
 398     top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
 399     topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
 400     left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
 401     left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
 402
 403     if(IS_INTRA(mb_type)){
 404         h->topleft_samples_available=
 405         h->top_samples_available=
 406         h->left_samples_available= 0xFFFF;
 407         h->topright_samples_available= 0xEEEA;
 408
 409         if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
 410             h->topleft_samples_available= 0xB3FF;
 411             h->top_samples_available= 0x33FF;
 412             h->topright_samples_available= 0x26EA;
 413         }
 414         for(i=0; i<2; i++){
 415             if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
 416                 h->topleft_samples_available&= 0xDF5F;
 417                 h->left_samples_available&= 0x5F5F;
 418             }
 419         }
 420
 421         if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
 422             h->topleft_samples_available&= 0x7FFF;
 423
 424         if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
 425             h->topright_samples_available&= 0xFBFF;
 426
 427         if(IS_INTRA4x4(mb_type)){
 428             if(IS_INTRA4x4(top_type)){
 429                 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
 430                 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
 431                 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
 432                 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
 433             }else{
 434                 int pred;
 435                 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
 436                     pred= 2;
 437                 else{
 438                     pred= -1;
 439                 }
 440                 h->intra4x4_pred_mode_cache[4+8*0]=
 441                 h->intra4x4_pred_mode_cache[5+8*0]=
 442                 h->intra4x4_pred_mode_cache[6+8*0]=
 443                 h->intra4x4_pred_mode_cache[7+8*0]= pred;
 444             }
 445             for(i=0; i<2; i++){
 446                 if(IS_INTRA4x4(left_type[i])){
 447                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
 448                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
 449                 }else{
 450                     int pred;
 451                     if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
 452                         pred= 2;
 453                     else{
 454                         pred= -1;
 455                     }
 456                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
 457                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
 458                 }
 459             }
 460         }
 461     }
 462
 463
 464 /*
 465 0 . T T. T T T T
 466 1 L . .L . . . .
 467 2 L . .L . . . .
 468 3 . T TL . . . .
 469 4 L . .L . . . .
 470 5 L . .. . . . .
 471 */
 472 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
 473     if(top_type){
 474         h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
 475         h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
 476         h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
 477         h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
 478
 479         h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
 480         h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
 481
 482         h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
 483         h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
 484     }else{
 485         h->non_zero_count_cache[4+8*0]=
 486         h->non_zero_count_cache[5+8*0]=
 487         h->non_zero_count_cache[6+8*0]=
 488         h->non_zero_count_cache[7+8*0]=
 489
 490         h->non_zero_count_cache[1+8*0]=
 491         h->non_zero_count_cache[2+8*0]=
 492
 493         h->non_zero_count_cache[1+8*3]=
 494         h->non_zero_count_cache[2+8*3]= 64;
 495     }
 496
 497     if(left_type[0]){
 498         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
 499         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
 500         h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
 501         h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
 502     }else{
 503         h->non_zero_count_cache[3+8*1]=
 504         h->non_zero_count_cache[3+8*2]=
 505         h->non_zero_count_cache[0+8*1]=
 506         h->non_zero_count_cache[0+8*4]= 64;
 507     }
 508
 509     if(left_type[1]){
 510         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
 511         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
 512         h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
 513         h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
 514     }else{
 515         h->non_zero_count_cache[3+8*3]=
 516         h->non_zero_count_cache[3+8*4]=
 517         h->non_zero_count_cache[0+8*2]=
 518         h->non_zero_count_cache[0+8*5]= 64;
 519     }
 520
 521 #if 1
 522     if(IS_INTER(mb_type)){
 523         int list;
 524         for(list=0; list<2; list++){
 525             if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
 526                 /*if(!h->mv_cache_clean[list]){
 527                     memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
 528                     memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
 529                     h->mv_cache_clean[list]= 1;
 530                 }*/
 531                 continue; //FIXME direct mode ...
 532             }
 533             h->mv_cache_clean[list]= 0;
 534
 535             if(IS_INTER(topleft_type)){
 536                 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
 537                 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
 538                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
 539                 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
 540             }else{
 541                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
 542                 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
 543             }
 544
 545             if(IS_INTER(top_type)){
 546                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
 547                 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
 548                 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
 549                 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
 550                 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
 551                 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
 552                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
 553                 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
 554                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
 555                 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
 556             }else{
 557                 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
 558                 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
 559                 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
 560                 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
 561                 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
 562             }
 563
 564             if(IS_INTER(topright_type)){
 565                 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
 566                 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
 567                 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
 568                 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
 569             }else{
 570                 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
 571                 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
 572             }
 573
 574             //FIXME unify cleanup or sth
 575             if(IS_INTER(left_type[0])){
 576                 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
 577                 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
 578                 *(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]];
 579                 *(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]];
 580                 h->ref_cache[list][scan8[0] - 1 + 0*8]=
 581                 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
 582             }else{
 583                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
 584                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
 585                 h->ref_cache[list][scan8[0] - 1 + 0*8]=
 586                 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
 587             }
 588
 589             if(IS_INTER(left_type[1])){
 590                 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
 591                 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
 592                 *(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]];
 593                 *(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]];
 594                 h->ref_cache[list][scan8[0] - 1 + 2*8]=
 595                 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
 596             }else{
 597                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
 598                 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
 599                 h->ref_cache[list][scan8[0] - 1 + 2*8]=
 600                 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
 601             }
 602
 603             h->ref_cache[list][scan8[5 ]+1] =
 604             h->ref_cache[list][scan8[7 ]+1] =
 605             h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewher else)
 606             h->ref_cache[list][scan8[4 ]] =
 607             h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
 608             *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
 609             *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
 610             *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
 611             *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
 612             *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
 613         }
 614 //FIXME
 615
 616     }
 617 #endif
 618 }
 619
 620 static inline void write_back_intra_pred_mode(H264Context *h){
 621     MpegEncContext * const s = &h->s;
 622     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
 623
 624     h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
 625     h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
 626     h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
 627     h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
 628     h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
 629     h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
 630     h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
 631 }
 632
 633 /**
 634  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 635  */
 636 static inline int check_intra4x4_pred_mode(H264Context *h){
 637     MpegEncContext * const s = &h->s;
 638     static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
 639     static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
 640     int i;
 641
 642     if(!(h->top_samples_available&0x8000)){
 643         for(i=0; i<4; i++){
 644             int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
 645             if(status<0){
 646                 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
 647                 return -1;
 648             } else if(status){
 649                 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
 650             }
 651         }
 652     }
 653
 654     if(!(h->left_samples_available&0x8000)){
 655         for(i=0; i<4; i++){
 656             int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
 657             if(status<0){
 658                 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
 659                 return -1;
 660             } else if(status){
 661                 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
 662             }
 663         }
 664     }
 665
 666     return 0;
 667 } //FIXME cleanup like next
 668
 669 /**
 670  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
 671  */
 672 static inline int check_intra_pred_mode(H264Context *h, int mode){
 673     MpegEncContext * const s = &h->s;
 674     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
 675     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
 676
 677     if(!(h->top_samples_available&0x8000)){
 678         mode= top[ mode ];
 679         if(mode<0){
 680             av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 681             return -1;
 682         }
 683     }
 684
 685     if(!(h->left_samples_available&0x8000)){
 686         mode= left[ mode ];
 687         if(mode<0){
 688             av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
 689             return -1;
 690         }
 691     }
 692
 693     return mode;
 694 }
 695
 696 /**
 697  * gets the predicted intra4x4 prediction mode.
 698  */
 699 static inline int pred_intra_mode(H264Context *h, int n){
 700     const int index8= scan8[n];
 701     const int left= h->intra4x4_pred_mode_cache[index8 - 1];
 702     const int top = h->intra4x4_pred_mode_cache[index8 - 8];
 703     const int min= FFMIN(left, top);
 704
 705     tprintf("mode:%d %d min:%d\n", left ,top, min);
 706
 707     if(min<0) return DC_PRED;
 708     else      return min;
 709 }
 710
 711 static inline void write_back_non_zero_count(H264Context *h){
 712     MpegEncContext * const s = &h->s;
 713     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
 714
 715     h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
 716     h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
 717     h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
 718     h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
 719     h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
 720     h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
 721     h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
 722
 723     h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
 724     h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
 725     h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
 726
 727     h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
 728     h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
 729     h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
 730 }
 731
 732 /**
 733  * gets the predicted number of non zero coefficients.
 734  * @param n block index
 735  */
 736 static inline int pred_non_zero_count(H264Context *h, int n){
 737     const int index8= scan8[n];
 738     const int left= h->non_zero_count_cache[index8 - 1];
 739     const int top = h->non_zero_count_cache[index8 - 8];
 740     int i= left + top;
 741
 742     if(i<64) i= (i+1)>>1;
 743
 744     tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
 745
 746     return i&31;
 747 }
 748
 749 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
 750     const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
 751
 752     if(topright_ref != PART_NOT_AVAILABLE){
 753         *C= h->mv_cache[list][ i - 8 + part_width ];
 754         return topright_ref;
 755     }else{
 756         tprintf("topright MV not available\n");
 757
 758         *C= h->mv_cache[list][ i - 8 - 1 ];
 759         return h->ref_cache[list][ i - 8 - 1 ];
 760     }
 761 }
 762
 763 /**
 764  * gets the predicted MV.
 765  * @param n the block index
 766  * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
 767  * @param mx the x component of the predicted motion vector
 768  * @param my the y component of the predicted motion vector
 769  */
 770 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
 771     const int index8= scan8[n];
 772     const int top_ref=      h->ref_cache[list][ index8 - 8 ];
 773     const int left_ref=     h->ref_cache[list][ index8 - 1 ];
 774     const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
 775     const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
 776     const int16_t * C;
 777     int diagonal_ref, match_count;
 778
 779     assert(part_width==1 || part_width==2 || part_width==4);
 780
 781 /* mv_cache
 782   B . . A T T T T
 783   U . . L . . , .
 784   U . . L . . . .
 785   U . . L . . , .
 786   . . . L . . . .
 787 */
 788
 789     diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
 790     match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
 791     if(match_count > 1){ //most common
 792         *mx= mid_pred(A[0], B[0], C[0]);
 793         *my= mid_pred(A[1], B[1], C[1]);
 794     }else if(match_count==1){
 795         if(left_ref==ref){
 796             *mx= A[0];
 797             *my= A[1];
 798         }else if(top_ref==ref){
 799             *mx= B[0];
 800             *my= B[1];
 801         }else{
 802             *mx= C[0];
 803             *my= C[1];
 804         }
 805     }else{
 806         if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
 807             *mx= A[0];
 808             *my= A[1];
 809         }else{
 810             *mx= mid_pred(A[0], B[0], C[0]);
 811             *my= mid_pred(A[1], B[1], C[1]);
 812         }
 813     }
 814
 815     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);
 816 }
 817
 818 /**
 819  * gets the directionally predicted 16x8 MV.
 820  * @param n the block index
 821  * @param mx the x component of the predicted motion vector
 822  * @param my the y component of the predicted motion vector
 823  */
 824 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
 825     if(n==0){
 826         const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
 827         const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
 828
 829         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);
 830
 831         if(top_ref == ref){
 832             *mx= B[0];
 833             *my= B[1];
 834             return;
 835         }
 836     }else{
 837         const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
 838         const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
 839
 840         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);
 841
 842         if(left_ref == ref){
 843             *mx= A[0];
 844             *my= A[1];
 845             return;
 846         }
 847     }
 848
 849     //RARE
 850     pred_motion(h, n, 4, list, ref, mx, my);
 851 }
 852
 853 /**
 854  * gets the directionally predicted 8x16 MV.
 855  * @param n the block index
 856  * @param mx the x component of the predicted motion vector
 857  * @param my the y component of the predicted motion vector
 858  */
 859 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
 860     if(n==0){
 861         const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
 862         const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
 863
 864         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);
 865
 866         if(left_ref == ref){
 867             *mx= A[0];
 868             *my= A[1];
 869             return;
 870         }
 871     }else{
 872         const int16_t * C;
 873         int diagonal_ref;
 874
 875         diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
 876
 877         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);
 878
 879         if(diagonal_ref == ref){
 880             *mx= C[0];
 881             *my= C[1];
 882             return;
 883         }
 884     }
 885
 886     //RARE
 887     pred_motion(h, n, 2, list, ref, mx, my);
 888 }
 889
 890 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
 891     const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
 892     const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
 893
 894     tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
 895
 896     if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
 897        || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
 898        || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
 899
 900         *mx = *my = 0;
 901         return;
 902     }
 903
 904     pred_motion(h, 0, 4, 0, 0, mx, my);
 905
 906     return;
 907 }
 908
 909 static inline void write_back_motion(H264Context *h, int mb_type){
 910     MpegEncContext * const s = &h->s;
 911     const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
 912     const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
 913     int list;
 914
 915     for(list=0; list<2; list++){
 916         int y;
 917         if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
 918             if(1){ //FIXME skip or never read if mb_type doesnt use it
 919                 for(y=0; y<4; y++){
 920                     *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
 921                     *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
 922                 }
 923                 for(y=0; y<2; y++){
 924                     *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
 925                 }
 926             }
 927             continue; //FIXME direct mode ...
 928         }
 929
 930         for(y=0; y<4; y++){
 931             *(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];
 932             *(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];
 933         }
 934         for(y=0; y<2; y++){
 935             s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
 936             s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
 937         }
 938     }
 939 }
 940
 941 /**
 942  * Decodes a network abstraction layer unit.
 943  * @param consumed is the number of bytes used as input
 944  * @param length is the length of the array
 945  * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
 946  * @returns decoded bytes, might be src+1 if no escapes
 947  */
 948 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
 949     int i, si, di;
 950     uint8_t *dst;
 951
 952 //    src[0]&0x80;              //forbidden bit
 953     h->nal_ref_idc= src[0]>>5;
 954     h->nal_unit_type= src[0]&0x1F;
 955
 956     src++; length--;
 957 #if 0
 958     for(i=0; i<length; i++)
 959         printf("%2X ", src[i]);
 960 #endif
 961     for(i=0; i+1<length; i+=2){
 962         if(src[i]) continue;
 963         if(i>0 && src[i-1]==0) i--;
 964         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
 965             if(src[i+2]!=3){
 966                 /* startcode, so we must be past the end */
 967                 length=i;
 968             }
 969             break;
 970         }
 971     }
 972
 973     if(i>=length-1){ //no escaped 0
 974         *dst_length= length;
 975         *consumed= length+1; //+1 for the header
 976         return src;
 977     }
 978
 979     h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
 980     dst= h->rbsp_buffer;
 981
 982 //printf("deoding esc\n");
 983     si=di=0;
 984     while(si<length){
 985         //remove escapes (very rare 1:2^22)
 986         if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
 987             if(src[si+2]==3){ //escape
 988                 dst[di++]= 0;
 989                 dst[di++]= 0;
 990                 si+=3;
 991             }else //next start code
 992                 break;
 993         }
 994
 995         dst[di++]= src[si++];
 996     }
 997
 998     *dst_length= di;
 999     *consumed= si + 1;//+1 for the header
1000 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1001     return dst;
1002 }
1003
1004 /**
1005  * @param src the data which should be escaped
1006  * @param dst the target buffer, dst+1 == src is allowed as a special case
1007  * @param length the length of the src data
1008  * @param dst_length the length of the dst array
1009  * @returns length of escaped data in bytes or -1 if an error occured
1010  */
1011 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1012     int i, escape_count, si, di;
1013     uint8_t *temp;
1014
1015     assert(length>=0);
1016     assert(dst_length>0);
1017
1018     dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1019
1020     if(length==0) return 1;
1021
1022     escape_count= 0;
1023     for(i=0; i<length; i+=2){
1024         if(src[i]) continue;
1025         if(i>0 && src[i-1]==0)
1026             i--;
1027         if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1028             escape_count++;
1029             i+=2;
1030         }
1031     }
1032
1033     if(escape_count==0){
1034         if(dst+1 != src)
1035             memcpy(dst+1, src, length);
1036         return length + 1;
1037     }
1038
1039     if(length + escape_count + 1> dst_length)
1040         return -1;
1041
1042     //this should be damn rare (hopefully)
1043
1044     h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1045     temp= h->rbsp_buffer;
1046 //printf("encoding esc\n");
1047
1048     si= 0;
1049     di= 0;
1050     while(si < length){
1051         if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1052             temp[di++]= 0; si++;
1053             temp[di++]= 0; si++;
1054             temp[di++]= 3;
1055             temp[di++]= src[si++];
1056         }
1057         else
1058             temp[di++]= src[si++];
1059     }
1060     memcpy(dst+1, temp, length+escape_count);
1061
1062     assert(di == length+escape_count);
1063
1064     return di + 1;
1065 }
1066
1067 /**
1068  * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1069  */
1070 static void encode_rbsp_trailing(PutBitContext *pb){
1071     int length;
1072     put_bits(pb, 1, 1);
1073     length= (-put_bits_count(pb))&7;
1074     if(length) put_bits(pb, length, 0);
1075 }
1076
1077 /**
1078  * identifies the exact end of the bitstream
1079  * @return the length of the trailing, or 0 if damaged
1080  */
1081 static int decode_rbsp_trailing(uint8_t *src){
1082     int v= *src;
1083     int r;
1084
1085     tprintf("rbsp trailing %X\n", v);
1086
1087     for(r=1; r<9; r++){
1088         if(v&1) return r;
1089         v>>=1;
1090     }
1091     return 0;
1092 }
1093
1094 /**
1095  * idct tranforms the 16 dc values and dequantize them.
1096  * @param qp quantization parameter
1097  */
1098 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1099     const int qmul= dequant_coeff[qp][0];
1100 #define stride 16
1101     int i;
1102     int temp[16]; //FIXME check if this is a good idea
1103     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1104     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1105
1106 //memset(block, 64, 2*256);
1107 //return;
1108     for(i=0; i<4; i++){
1109         const int offset= y_offset[i];
1110         const int z0= block[offset+stride*0] + block[offset+stride*4];
1111         const int z1= block[offset+stride*0] - block[offset+stride*4];
1112         const int z2= block[offset+stride*1] - block[offset+stride*5];
1113         const int z3= block[offset+stride*1] + block[offset+stride*5];
1114
1115         temp[4*i+0]= z0+z3;
1116         temp[4*i+1]= z1+z2;
1117         temp[4*i+2]= z1-z2;
1118         temp[4*i+3]= z0-z3;
1119     }
1120
1121     for(i=0; i<4; i++){
1122         const int offset= x_offset[i];
1123         const int z0= temp[4*0+i] + temp[4*2+i];
1124         const int z1= temp[4*0+i] - temp[4*2+i];
1125         const int z2= temp[4*1+i] - temp[4*3+i];
1126         const int z3= temp[4*1+i] + temp[4*3+i];
1127
1128         block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1129         block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1130         block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1131         block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1132     }
1133 }
1134
1135 /**
1136  * dct tranforms the 16 dc values.
1137  * @param qp quantization parameter ??? FIXME
1138  */
1139 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1140 //    const int qmul= dequant_coeff[qp][0];
1141     int i;
1142     int temp[16]; //FIXME check if this is a good idea
1143     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1144     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1145
1146     for(i=0; i<4; i++){
1147         const int offset= y_offset[i];
1148         const int z0= block[offset+stride*0] + block[offset+stride*4];
1149         const int z1= block[offset+stride*0] - block[offset+stride*4];
1150         const int z2= block[offset+stride*1] - block[offset+stride*5];
1151         const int z3= block[offset+stride*1] + block[offset+stride*5];
1152
1153         temp[4*i+0]= z0+z3;
1154         temp[4*i+1]= z1+z2;
1155         temp[4*i+2]= z1-z2;
1156         temp[4*i+3]= z0-z3;
1157     }
1158
1159     for(i=0; i<4; i++){
1160         const int offset= x_offset[i];
1161         const int z0= temp[4*0+i] + temp[4*2+i];
1162         const int z1= temp[4*0+i] - temp[4*2+i];
1163         const int z2= temp[4*1+i] - temp[4*3+i];
1164         const int z3= temp[4*1+i] + temp[4*3+i];
1165
1166         block[stride*0 +offset]= (z0 + z3)>>1;
1167         block[stride*2 +offset]= (z1 + z2)>>1;
1168         block[stride*8 +offset]= (z1 - z2)>>1;
1169         block[stride*10+offset]= (z0 - z3)>>1;
1170     }
1171 }
1172 #undef xStride
1173 #undef stride
1174
1175 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1176     const int qmul= dequant_coeff[qp][0];
1177     const int stride= 16*2;
1178     const int xStride= 16;
1179     int a,b,c,d,e;
1180
1181     a= block[stride*0 + xStride*0];
1182     b= block[stride*0 + xStride*1];
1183     c= block[stride*1 + xStride*0];
1184     d= block[stride*1 + xStride*1];
1185
1186     e= a-b;
1187     a= a+b;
1188     b= c-d;
1189     c= c+d;
1190
1191     block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1192     block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1193     block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1194     block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1195 }
1196
1197 static void chroma_dc_dct_c(DCTELEM *block){
1198     const int stride= 16*2;
1199     const int xStride= 16;
1200     int a,b,c,d,e;
1201
1202     a= block[stride*0 + xStride*0];
1203     b= block[stride*0 + xStride*1];
1204     c= block[stride*1 + xStride*0];
1205     d= block[stride*1 + xStride*1];
1206
1207     e= a-b;
1208     a= a+b;
1209     b= c-d;
1210     c= c+d;
1211
1212     block[stride*0 + xStride*0]= (a+c);
1213     block[stride*0 + xStride*1]= (e+b);
1214     block[stride*1 + xStride*0]= (a-c);
1215     block[stride*1 + xStride*1]= (e-b);
1216 }
1217
1218 /**
1219  * gets the chroma qp.
1220  */
1221 static inline int get_chroma_qp(H264Context *h, int qscale){
1222
1223     return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1224 }
1225
1226
1227 /**
1228  *
1229  */
1230 static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1231     int i;
1232     uint8_t *cm = cropTbl + MAX_NEG_CROP;
1233
1234     block[0] += 32;
1235 #if 1
1236     for(i=0; i<4; i++){
1237         const int z0=  block[i + 4*0]     +  block[i + 4*2];
1238         const int z1=  block[i + 4*0]     -  block[i + 4*2];
1239         const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
1240         const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
1241
1242         block[i + 4*0]= z0 + z3;
1243         block[i + 4*1]= z1 + z2;
1244         block[i + 4*2]= z1 - z2;
1245         block[i + 4*3]= z0 - z3;
1246     }
1247
1248     for(i=0; i<4; i++){
1249         const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
1250         const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
1251         const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
1252         const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
1253
1254         dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ];
1255         dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ];
1256         dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ];
1257         dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ];
1258     }
1259 #else
1260     for(i=0; i<4; i++){
1261         const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
1262         const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
1263         const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
1264         const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
1265
1266         block[0 + 4*i]= z0 + z3;
1267         block[1 + 4*i]= z1 + z2;
1268         block[2 + 4*i]= z1 - z2;
1269         block[3 + 4*i]= z0 - z3;
1270     }
1271
1272     for(i=0; i<4; i++){
1273         const int z0=  block[i + 4*0]     +  block[i + 4*2];
1274         const int z1=  block[i + 4*0]     -  block[i + 4*2];
1275         const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
1276         const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
1277
1278         dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1279         dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1280         dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1281         dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1282     }
1283 #endif
1284 }
1285
1286 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1287     int i;
1288     //FIXME try int temp instead of block
1289
1290     for(i=0; i<4; i++){
1291         const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1292         const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1293         const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1294         const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1295         const int z0= d0 + d3;
1296         const int z3= d0 - d3;
1297         const int z1= d1 + d2;
1298         const int z2= d1 - d2;
1299
1300         block[0 + 4*i]=   z0 +   z1;
1301         block[1 + 4*i]= 2*z3 +   z2;
1302         block[2 + 4*i]=   z0 -   z1;
1303         block[3 + 4*i]=   z3 - 2*z2;
1304     }
1305
1306     for(i=0; i<4; i++){
1307         const int z0= block[0*4 + i] + block[3*4 + i];
1308         const int z3= block[0*4 + i] - block[3*4 + i];
1309         const int z1= block[1*4 + i] + block[2*4 + i];
1310         const int z2= block[1*4 + i] - block[2*4 + i];
1311
1312         block[0*4 + i]=   z0 +   z1;
1313         block[1*4 + i]= 2*z3 +   z2;
1314         block[2*4 + i]=   z0 -   z1;
1315         block[3*4 + i]=   z3 - 2*z2;
1316     }
1317 }
1318
1319 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1320 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1321 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1322     int i;
1323     const int * const quant_table= quant_coeff[qscale];
1324     const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1325     const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1326     const unsigned int threshold2= (threshold1<<1);
1327     int last_non_zero;
1328
1329     if(seperate_dc){
1330         if(qscale<=18){
1331             //avoid overflows
1332             const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1333             const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1334             const unsigned int dc_threshold2= (dc_threshold1<<1);
1335
1336             int level= block[0]*quant_coeff[qscale+18][0];
1337             if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1338                 if(level>0){
1339                     level= (dc_bias + level)>>(QUANT_SHIFT-2);
1340                     block[0]= level;
1341                 }else{
1342                     level= (dc_bias - level)>>(QUANT_SHIFT-2);
1343                     block[0]= -level;
1344                 }
1345 //                last_non_zero = i;
1346             }else{
1347                 block[0]=0;
1348             }
1349         }else{
1350             const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1351             const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1352             const unsigned int dc_threshold2= (dc_threshold1<<1);
1353
1354             int level= block[0]*quant_table[0];
1355             if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1356                 if(level>0){
1357                     level= (dc_bias + level)>>(QUANT_SHIFT+1);
1358                     block[0]= level;
1359                 }else{
1360                     level= (dc_bias - level)>>(QUANT_SHIFT+1);
1361                     block[0]= -level;
1362                 }
1363 //                last_non_zero = i;
1364             }else{
1365                 block[0]=0;
1366             }
1367         }
1368         last_non_zero= 0;
1369         i=1;
1370     }else{
1371         last_non_zero= -1;
1372         i=0;
1373     }
1374
1375     for(; i<16; i++){
1376         const int j= scantable[i];
1377         int level= block[j]*quant_table[j];
1378
1379 //        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1380 //           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1381         if(((unsigned)(level+threshold1))>threshold2){
1382             if(level>0){
1383                 level= (bias + level)>>QUANT_SHIFT;
1384                 block[j]= level;
1385             }else{
1386                 level= (bias - level)>>QUANT_SHIFT;
1387                 block[j]= -level;
1388             }
1389             last_non_zero = i;
1390         }else{
1391             block[j]=0;
1392         }
1393     }
1394
1395     return last_non_zero;
1396 }
1397
1398 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1399     const uint32_t a= ((uint32_t*)(src-stride))[0];
1400     ((uint32_t*)(src+0*stride))[0]= a;
1401     ((uint32_t*)(src+1*stride))[0]= a;
1402     ((uint32_t*)(src+2*stride))[0]= a;
1403     ((uint32_t*)(src+3*stride))[0]= a;
1404 }
1405
1406 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1407     ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1408     ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1409     ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1410     ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1411 }
1412
1413 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1414     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1415                    + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1416
1417     ((uint32_t*)(src+0*stride))[0]=
1418     ((uint32_t*)(src+1*stride))[0]=
1419     ((uint32_t*)(src+2*stride))[0]=
1420     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1421 }
1422
1423 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1424     const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1425
1426     ((uint32_t*)(src+0*stride))[0]=
1427     ((uint32_t*)(src+1*stride))[0]=
1428     ((uint32_t*)(src+2*stride))[0]=
1429     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1430 }
1431
1432 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1433     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1434
1435     ((uint32_t*)(src+0*stride))[0]=
1436     ((uint32_t*)(src+1*stride))[0]=
1437     ((uint32_t*)(src+2*stride))[0]=
1438     ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1439 }
1440
1441 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1442     ((uint32_t*)(src+0*stride))[0]=
1443     ((uint32_t*)(src+1*stride))[0]=
1444     ((uint32_t*)(src+2*stride))[0]=
1445     ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1446 }
1447
1448
1449 #define LOAD_TOP_RIGHT_EDGE\
1450     const int t4= topright[0];\
1451     const int t5= topright[1];\
1452     const int t6= topright[2];\
1453     const int t7= topright[3];\
1454
1455 #define LOAD_LEFT_EDGE\
1456     const int l0= src[-1+0*stride];\
1457     const int l1= src[-1+1*stride];\
1458     const int l2= src[-1+2*stride];\
1459     const int l3= src[-1+3*stride];\
1460
1461 #define LOAD_TOP_EDGE\
1462     const int t0= src[ 0-1*stride];\
1463     const int t1= src[ 1-1*stride];\
1464     const int t2= src[ 2-1*stride];\
1465     const int t3= src[ 3-1*stride];\
1466
1467 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1468     const int lt= src[-1-1*stride];
1469     LOAD_TOP_EDGE
1470     LOAD_LEFT_EDGE
1471
1472     src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1473     src[0+2*stride]=
1474     src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1475     src[0+1*stride]=
1476     src[1+2*stride]=
1477     src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1478     src[0+0*stride]=
1479     src[1+1*stride]=
1480     src[2+2*stride]=
1481     src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1482     src[1+0*stride]=
1483     src[2+1*stride]=
1484     src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1485     src[2+0*stride]=
1486     src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1487     src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1488 }
1489
1490 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1491     LOAD_TOP_EDGE
1492     LOAD_TOP_RIGHT_EDGE
1493 //    LOAD_LEFT_EDGE
1494
1495     src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1496     src[1+0*stride]=
1497     src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1498     src[2+0*stride]=
1499     src[1+1*stride]=
1500     src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1501     src[3+0*stride]=
1502     src[2+1*stride]=
1503     src[1+2*stride]=
1504     src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1505     src[3+1*stride]=
1506     src[2+2*stride]=
1507     src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1508     src[3+2*stride]=
1509     src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1510     src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1511 }
1512
1513 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1514     const int lt= src[-1-1*stride];
1515     LOAD_TOP_EDGE
1516     LOAD_LEFT_EDGE
1517     const __attribute__((unused)) int unu= l3;
1518
1519     src[0+0*stride]=
1520     src[1+2*stride]=(lt + t0 + 1)>>1;
1521     src[1+0*stride]=
1522     src[2+2*stride]=(t0 + t1 + 1)>>1;
1523     src[2+0*stride]=
1524     src[3+2*stride]=(t1 + t2 + 1)>>1;
1525     src[3+0*stride]=(t2 + t3 + 1)>>1;
1526     src[0+1*stride]=
1527     src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1528     src[1+1*stride]=
1529     src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1530     src[2+1*stride]=
1531     src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1532     src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1533     src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1534     src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1535 }
1536
1537 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1538     LOAD_TOP_EDGE
1539     LOAD_TOP_RIGHT_EDGE
1540     const __attribute__((unused)) int unu= t7;
1541
1542     src[0+0*stride]=(t0 + t1 + 1)>>1;
1543     src[1+0*stride]=
1544     src[0+2*stride]=(t1 + t2 + 1)>>1;
1545     src[2+0*stride]=
1546     src[1+2*stride]=(t2 + t3 + 1)>>1;
1547     src[3+0*stride]=
1548     src[2+2*stride]=(t3 + t4+ 1)>>1;
1549     src[3+2*stride]=(t4 + t5+ 1)>>1;
1550     src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1551     src[1+1*stride]=
1552     src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1553     src[2+1*stride]=
1554     src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1555     src[3+1*stride]=
1556     src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1557     src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1558 }
1559
1560 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1561     LOAD_LEFT_EDGE
1562
1563     src[0+0*stride]=(l0 + l1 + 1)>>1;
1564     src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1565     src[2+0*stride]=
1566     src[0+1*stride]=(l1 + l2 + 1)>>1;
1567     src[3+0*stride]=
1568     src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1569     src[2+1*stride]=
1570     src[0+2*stride]=(l2 + l3 + 1)>>1;
1571     src[3+1*stride]=
1572     src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1573     src[3+2*stride]=
1574     src[1+3*stride]=
1575     src[0+3*stride]=
1576     src[2+2*stride]=
1577     src[2+3*stride]=
1578     src[3+3*stride]=l3;
1579 }
1580
1581 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1582     const int lt= src[-1-1*stride];
1583     LOAD_TOP_EDGE
1584     LOAD_LEFT_EDGE
1585     const __attribute__((unused)) int unu= t3;
1586
1587     src[0+0*stride]=
1588     src[2+1*stride]=(lt + l0 + 1)>>1;
1589     src[1+0*stride]=
1590     src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1591     src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1592     src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1593     src[0+1*stride]=
1594     src[2+2*stride]=(l0 + l1 + 1)>>1;
1595     src[1+1*stride]=
1596     src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1597     src[0+2*stride]=
1598     src[2+3*stride]=(l1 + l2+ 1)>>1;
1599     src[1+2*stride]=
1600     src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1601     src[0+3*stride]=(l2 + l3 + 1)>>1;
1602     src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1603 }
1604
1605 static void pred16x16_vertical_c(uint8_t *src, int stride){
1606     int i;
1607     const uint32_t a= ((uint32_t*)(src-stride))[0];
1608     const uint32_t b= ((uint32_t*)(src-stride))[1];
1609     const uint32_t c= ((uint32_t*)(src-stride))[2];
1610     const uint32_t d= ((uint32_t*)(src-stride))[3];
1611
1612     for(i=0; i<16; i++){
1613         ((uint32_t*)(src+i*stride))[0]= a;
1614         ((uint32_t*)(src+i*stride))[1]= b;
1615         ((uint32_t*)(src+i*stride))[2]= c;
1616         ((uint32_t*)(src+i*stride))[3]= d;
1617     }
1618 }
1619
1620 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1621     int i;
1622
1623     for(i=0; i<16; i++){
1624         ((uint32_t*)(src+i*stride))[0]=
1625         ((uint32_t*)(src+i*stride))[1]=
1626         ((uint32_t*)(src+i*stride))[2]=
1627         ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1628     }
1629 }
1630
1631 static void pred16x16_dc_c(uint8_t *src, int stride){
1632     int i, dc=0;
1633
1634     for(i=0;i<16; i++){
1635         dc+= src[-1+i*stride];
1636     }
1637
1638     for(i=0;i<16; i++){
1639         dc+= src[i-stride];
1640     }
1641
1642     dc= 0x01010101*((dc + 16)>>5);
1643
1644     for(i=0; i<16; i++){
1645         ((uint32_t*)(src+i*stride))[0]=
1646         ((uint32_t*)(src+i*stride))[1]=
1647         ((uint32_t*)(src+i*stride))[2]=
1648         ((uint32_t*)(src+i*stride))[3]= dc;
1649     }
1650 }
1651
1652 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1653     int i, dc=0;
1654
1655     for(i=0;i<16; i++){
1656         dc+= src[-1+i*stride];
1657     }
1658
1659     dc= 0x01010101*((dc + 8)>>4);
1660
1661     for(i=0; i<16; i++){
1662         ((uint32_t*)(src+i*stride))[0]=
1663         ((uint32_t*)(src+i*stride))[1]=
1664         ((uint32_t*)(src+i*stride))[2]=
1665         ((uint32_t*)(src+i*stride))[3]= dc;
1666     }
1667 }
1668
1669 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1670     int i, dc=0;
1671
1672     for(i=0;i<16; i++){
1673         dc+= src[i-stride];
1674     }
1675     dc= 0x01010101*((dc + 8)>>4);
1676
1677     for(i=0; i<16; i++){
1678         ((uint32_t*)(src+i*stride))[0]=
1679         ((uint32_t*)(src+i*stride))[1]=
1680         ((uint32_t*)(src+i*stride))[2]=
1681         ((uint32_t*)(src+i*stride))[3]= dc;
1682     }
1683 }
1684
1685 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1686     int i;
1687
1688     for(i=0; i<16; i++){
1689         ((uint32_t*)(src+i*stride))[0]=
1690         ((uint32_t*)(src+i*stride))[1]=
1691         ((uint32_t*)(src+i*stride))[2]=
1692         ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1693     }
1694 }
1695
1696 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1697   int i, j, k;
1698   int a;
1699   uint8_t *cm = cropTbl + MAX_NEG_CROP;
1700   const uint8_t * const src0 = src+7-stride;
1701   const uint8_t *src1 = src+8*stride-1;
1702   const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
1703   int H = src0[1] - src0[-1];
1704   int V = src1[0] - src2[ 0];
1705   for(k=2; k<=8; ++k) {
1706     src1 += stride; src2 -= stride;
1707     H += k*(src0[k] - src0[-k]);
1708     V += k*(src1[0] - src2[ 0]);
1709   }
1710   if(svq3){
1711     H = ( 5*(H/4) ) / 16;
1712     V = ( 5*(V/4) ) / 16;
1713
1714     /* required for 100% accuracy */
1715     i = H; H = V; V = i;
1716   }else{
1717     H = ( 5*H+32 ) >> 6;
1718     V = ( 5*V+32 ) >> 6;
1719   }
1720
1721   a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1722   for(j=16; j>0; --j) {
1723     int b = a;
1724     a += V;
1725     for(i=-16; i<0; i+=4) {
1726       src[16+i] = cm[ (b    ) >> 5 ];
1727       src[17+i] = cm[ (b+  H) >> 5 ];
1728       src[18+i] = cm[ (b+2*H) >> 5 ];
1729       src[19+i] = cm[ (b+3*H) >> 5 ];
1730       b += 4*H;
1731     }
1732     src += stride;
1733   }
1734 }
1735
1736 static void pred16x16_plane_c(uint8_t *src, int stride){
1737     pred16x16_plane_compat_c(src, stride, 0);
1738 }
1739
1740 static void pred8x8_vertical_c(uint8_t *src, int stride){
1741     int i;
1742     const uint32_t a= ((uint32_t*)(src-stride))[0];
1743     const uint32_t b= ((uint32_t*)(src-stride))[1];
1744
1745     for(i=0; i<8; i++){
1746         ((uint32_t*)(src+i*stride))[0]= a;
1747         ((uint32_t*)(src+i*stride))[1]= b;
1748     }
1749 }
1750
1751 static void pred8x8_horizontal_c(uint8_t *src, int stride){
1752     int i;
1753
1754     for(i=0; i<8; i++){
1755         ((uint32_t*)(src+i*stride))[0]=
1756         ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1757     }
1758 }
1759
1760 static void pred8x8_128_dc_c(uint8_t *src, int stride){
1761     int i;
1762
1763     for(i=0; i<4; i++){
1764         ((uint32_t*)(src+i*stride))[0]=
1765         ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1766     }
1767     for(i=4; i<8; i++){
1768         ((uint32_t*)(src+i*stride))[0]=
1769         ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1770     }
1771 }
1772
1773 static void pred8x8_left_dc_c(uint8_t *src, int stride){
1774     int i;
1775     int dc0, dc2;
1776
1777     dc0=dc2=0;
1778     for(i=0;i<4; i++){
1779         dc0+= src[-1+i*stride];
1780         dc2+= src[-1+(i+4)*stride];
1781     }
1782     dc0= 0x01010101*((dc0 + 2)>>2);
1783     dc2= 0x01010101*((dc2 + 2)>>2);
1784
1785     for(i=0; i<4; i++){
1786         ((uint32_t*)(src+i*stride))[0]=
1787         ((uint32_t*)(src+i*stride))[1]= dc0;
1788     }
1789     for(i=4; i<8; i++){
1790         ((uint32_t*)(src+i*stride))[0]=
1791         ((uint32_t*)(src+i*stride))[1]= dc2;
1792     }
1793 }
1794
1795 static void pred8x8_top_dc_c(uint8_t *src, int stride){
1796     int i;
1797     int dc0, dc1;
1798
1799     dc0=dc1=0;
1800     for(i=0;i<4; i++){
1801         dc0+= src[i-stride];
1802         dc1+= src[4+i-stride];
1803     }
1804     dc0= 0x01010101*((dc0 + 2)>>2);
1805     dc1= 0x01010101*((dc1 + 2)>>2);
1806
1807     for(i=0; i<4; i++){
1808         ((uint32_t*)(src+i*stride))[0]= dc0;
1809         ((uint32_t*)(src+i*stride))[1]= dc1;
1810     }
1811     for(i=4; i<8; i++){
1812         ((uint32_t*)(src+i*stride))[0]= dc0;
1813         ((uint32_t*)(src+i*stride))[1]= dc1;
1814     }
1815 }
1816
1817
1818 static void pred8x8_dc_c(uint8_t *src, int stride){
1819     int i;
1820     int dc0, dc1, dc2, dc3;
1821
1822     dc0=dc1=dc2=0;
1823     for(i=0;i<4; i++){
1824         dc0+= src[-1+i*stride] + src[i-stride];
1825         dc1+= src[4+i-stride];
1826         dc2+= src[-1+(i+4)*stride];
1827     }
1828     dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1829     dc0= 0x01010101*((dc0 + 4)>>3);
1830     dc1= 0x01010101*((dc1 + 2)>>2);
1831     dc2= 0x01010101*((dc2 + 2)>>2);
1832
1833     for(i=0; i<4; i++){
1834         ((uint32_t*)(src+i*stride))[0]= dc0;
1835         ((uint32_t*)(src+i*stride))[1]= dc1;
1836     }
1837     for(i=4; i<8; i++){
1838         ((uint32_t*)(src+i*stride))[0]= dc2;
1839         ((uint32_t*)(src+i*stride))[1]= dc3;
1840     }
1841 }
1842
1843 static void pred8x8_plane_c(uint8_t *src, int stride){
1844   int j, k;
1845   int a;
1846   uint8_t *cm = cropTbl + MAX_NEG_CROP;
1847   const uint8_t * const src0 = src+3-stride;
1848   const uint8_t *src1 = src+4*stride-1;
1849   const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
1850   int H = src0[1] - src0[-1];
1851   int V = src1[0] - src2[ 0];
1852   for(k=2; k<=4; ++k) {
1853     src1 += stride; src2 -= stride;
1854     H += k*(src0[k] - src0[-k]);
1855     V += k*(src1[0] - src2[ 0]);
1856   }
1857   H = ( 17*H+16 ) >> 5;
1858   V = ( 17*V+16 ) >> 5;
1859
1860   a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1861   for(j=8; j>0; --j) {
1862     int b = a;
1863     a += V;
1864     src[0] = cm[ (b    ) >> 5 ];
1865     src[1] = cm[ (b+  H) >> 5 ];
1866     src[2] = cm[ (b+2*H) >> 5 ];
1867     src[3] = cm[ (b+3*H) >> 5 ];
1868     src[4] = cm[ (b+4*H) >> 5 ];
1869     src[5] = cm[ (b+5*H) >> 5 ];
1870     src[6] = cm[ (b+6*H) >> 5 ];
1871     src[7] = cm[ (b+7*H) >> 5 ];
1872     src += stride;
1873   }
1874 }
1875
1876 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1877                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1878                            int src_x_offset, int src_y_offset,
1879                            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1880     MpegEncContext * const s = &h->s;
1881     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1882     const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1883     const int luma_xy= (mx&3) + ((my&3)<<2);
1884     uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1885     uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1886     uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1887     int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1888     int extra_height= extra_width;
1889     int emu=0;
1890     const int full_mx= mx>>2;
1891     const int full_my= my>>2;
1892
1893     assert(pic->data[0]);
1894
1895     if(mx&7) extra_width -= 3;
1896     if(my&7) extra_height -= 3;
1897
1898     if(   full_mx < 0-extra_width
1899        || full_my < 0-extra_height
1900        || full_mx + 16/*FIXME*/ > s->width + extra_width
1901        || full_my + 16/*FIXME*/ > s->height + extra_height){
1902         ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, s->width, s->height);
1903             src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1904         emu=1;
1905     }
1906
1907     qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1908     if(!square){
1909         qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1910     }
1911
1912     if(s->flags&CODEC_FLAG_GRAY) return;
1913
1914     if(emu){
1915         ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
1916             src_cb= s->edge_emu_buffer;
1917     }
1918     chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1919
1920     if(emu){
1921         ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
1922             src_cr= s->edge_emu_buffer;
1923     }
1924     chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1925 }
1926
1927 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1928                            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1929                            int x_offset, int y_offset,
1930                            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1931                            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1932                            int list0, int list1){
1933     MpegEncContext * const s = &h->s;
1934     qpel_mc_func *qpix_op=  qpix_put;
1935     h264_chroma_mc_func chroma_op= chroma_put;
1936
1937     dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
1938     dest_cb +=   x_offset +   y_offset*s->uvlinesize;
1939     dest_cr +=   x_offset +   y_offset*s->uvlinesize;
1940     x_offset += 8*s->mb_x;
1941     y_offset += 8*s->mb_y;
1942
1943     if(list0){
1944         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1945         mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1946                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
1947                            qpix_op, chroma_op);
1948
1949         qpix_op=  qpix_avg;
1950         chroma_op= chroma_avg;
1951     }
1952
1953     if(list1){
1954         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1955         mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1956                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
1957                            qpix_op, chroma_op);
1958     }
1959 }
1960
1961 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1962                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1963                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
1964     MpegEncContext * const s = &h->s;
1965     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1966     const int mb_type= s->current_picture.mb_type[mb_xy];
1967
1968     assert(IS_INTER(mb_type));
1969
1970     if(IS_16X16(mb_type)){
1971         mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1972                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1973                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1974     }else if(IS_16X8(mb_type)){
1975         mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1976                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1977                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1978         mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1979                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1980                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1981     }else if(IS_8X16(mb_type)){
1982         mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
1983                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1984                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1985         mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
1986                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1987                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1988     }else{
1989         int i;
1990
1991         assert(IS_8X8(mb_type));
1992
1993         for(i=0; i<4; i++){
1994             const int sub_mb_type= h->sub_mb_type[i];
1995             const int n= 4*i;
1996             int x_offset= (i&1)<<2;
1997             int y_offset= (i&2)<<1;
1998
1999             if(IS_SUB_8X8(sub_mb_type)){
2000                 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2001                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2002                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2003             }else if(IS_SUB_8X4(sub_mb_type)){
2004                 mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2005                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2006                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2007                 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2008                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2009                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2010             }else if(IS_SUB_4X8(sub_mb_type)){
2011                 mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2012                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2013                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2014                 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2015                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2016                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2017             }else{
2018                 int j;
2019                 assert(IS_SUB_4X4(sub_mb_type));
2020                 for(j=0; j<4; j++){
2021                     int sub_x_offset= x_offset + 2*(j&1);
2022                     int sub_y_offset= y_offset +   (j&2);
2023                     mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2024                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2025                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2026                 }
2027             }
2028         }
2029     }
2030 }
2031
2032 static void decode_init_vlc(H264Context *h){
2033     static int done = 0;
2034
2035     if (!done) {
2036         int i;
2037         done = 1;
2038
2039         init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2040                  &chroma_dc_coeff_token_len [0], 1, 1,
2041                  &chroma_dc_coeff_token_bits[0], 1, 1);
2042
2043         for(i=0; i<4; i++){
2044             init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2045                      &coeff_token_len [i][0], 1, 1,
2046                      &coeff_token_bits[i][0], 1, 1);
2047         }
2048
2049         for(i=0; i<3; i++){
2050             init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2051                      &chroma_dc_total_zeros_len [i][0], 1, 1,
2052                      &chroma_dc_total_zeros_bits[i][0], 1, 1);
2053         }
2054         for(i=0; i<15; i++){
2055             init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2056                      &total_zeros_len [i][0], 1, 1,
2057                      &total_zeros_bits[i][0], 1, 1);
2058         }
2059
2060         for(i=0; i<6; i++){
2061             init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2062                      &run_len [i][0], 1, 1,
2063                      &run_bits[i][0], 1, 1);
2064         }
2065         init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2066                  &run_len [6][0], 1, 1,
2067                  &run_bits[6][0], 1, 1);
2068     }
2069 }
2070
2071 /**
2072  * Sets the intra prediction function pointers.
2073  */
2074 static void init_pred_ptrs(H264Context *h){
2075 //    MpegEncContext * const s = &h->s;
2076
2077     h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2078     h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2079     h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2080     h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2081     h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2082     h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2083     h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2084     h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2085     h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2086     h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2087     h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2088     h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2089
2090     h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2091     h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2092     h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2093     h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2094     h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2095     h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2096     h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2097
2098     h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2099     h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2100     h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2101     h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2102     h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2103     h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2104     h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2105 }
2106
2107 static void free_tables(H264Context *h){
2108     av_freep(&h->intra4x4_pred_mode);
2109     av_freep(&h->non_zero_count);
2110     av_freep(&h->slice_table_base);
2111     h->slice_table= NULL;
2112
2113     av_freep(&h->mb2b_xy);
2114     av_freep(&h->mb2b8_xy);
2115 }
2116
2117 /**
2118  * allocates tables.
2119  * needs widzh/height
2120  */
2121 static int alloc_tables(H264Context *h){
2122     MpegEncContext * const s = &h->s;
2123     const int big_mb_num= s->mb_stride * (s->mb_height+1);
2124     int x,y;
2125
2126     CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2127     CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2128     CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2129
2130     memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2131     h->slice_table= h->slice_table_base + s->mb_stride + 1;
2132
2133     CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint16_t));
2134     CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2135     for(y=0; y<s->mb_height; y++){
2136         for(x=0; x<s->mb_width; x++){
2137             const int mb_xy= x + y*s->mb_stride;
2138             const int b_xy = 4*x + 4*y*h->b_stride;
2139             const int b8_xy= 2*x + 2*y*h->b8_stride;
2140
2141             h->mb2b_xy [mb_xy]= b_xy;
2142             h->mb2b8_xy[mb_xy]= b8_xy;
2143         }
2144     }
2145
2146     return 0;
2147 fail:
2148     free_tables(h);
2149     return -1;
2150 }
2151
2152 static void common_init(H264Context *h){
2153     MpegEncContext * const s = &h->s;
2154
2155     s->width = s->avctx->width;
2156     s->height = s->avctx->height;
2157     s->codec_id= s->avctx->codec->id;
2158
2159     init_pred_ptrs(h);
2160
2161     s->unrestricted_mv=1;
2162     s->decode=1; //FIXME
2163 }
2164
2165 static int decode_init(AVCodecContext *avctx){
2166     H264Context *h= avctx->priv_data;
2167     MpegEncContext * const s = &h->s;
2168
2169     s->avctx = avctx;
2170     common_init(h);
2171
2172     s->out_format = FMT_H264;
2173     s->workaround_bugs= avctx->workaround_bugs;
2174
2175     // set defaults
2176     s->progressive_sequence=1;
2177 //    s->decode_mb= ff_h263_decode_mb;
2178     s->low_delay= 1;
2179     avctx->pix_fmt= PIX_FMT_YUV420P;
2180
2181     decode_init_vlc(h);
2182
2183     return 0;
2184 }
2185
2186 static void frame_start(H264Context *h){
2187     MpegEncContext * const s = &h->s;
2188     int i;
2189
2190     MPV_frame_start(s, s->avctx);
2191     ff_er_frame_start(s);
2192     h->mmco_index=0;
2193
2194     assert(s->linesize && s->uvlinesize);
2195
2196     for(i=0; i<16; i++){
2197         h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2198         h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2199     }
2200     for(i=0; i<4; i++){
2201         h->block_offset[16+i]=
2202         h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2203     }
2204
2205 //    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2206 }
2207
2208 static void hl_decode_mb(H264Context *h){
2209     MpegEncContext * const s = &h->s;
2210     const int mb_x= s->mb_x;
2211     const int mb_y= s->mb_y;
2212     const int mb_xy= mb_x + mb_y*s->mb_stride;
2213     const int mb_type= s->current_picture.mb_type[mb_xy];
2214     uint8_t  *dest_y, *dest_cb, *dest_cr;
2215     int linesize, uvlinesize /*dct_offset*/;
2216     int i;
2217
2218     if(!s->decode)
2219         return;
2220
2221     if(s->mb_skiped){
2222     }
2223
2224     dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2225     dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2226     dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2227
2228     if (h->mb_field_decoding_flag) {
2229         linesize = s->linesize * 2;
2230         uvlinesize = s->uvlinesize * 2;
2231         if(mb_y&1){ //FIXME move out of this func?
2232             dest_y -= s->linesize*15;
2233             dest_cb-= s->linesize*7;
2234             dest_cr-= s->linesize*7;
2235         }
2236     } else {
2237         linesize = s->linesize;
2238         uvlinesize = s->uvlinesize;
2239 //        dct_offset = s->linesize * 16;
2240     }
2241
2242     if(IS_INTRA(mb_type)){
2243         if(!(s->flags&CODEC_FLAG_GRAY)){
2244             h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2245             h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2246         }
2247
2248         if(IS_INTRA4x4(mb_type)){
2249             if(!s->encoding){
2250                 for(i=0; i<16; i++){
2251                     uint8_t * const ptr= dest_y + h->block_offset[i];
2252                     uint8_t *topright= ptr + 4 - linesize;
2253                     const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2254                     const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2255                     int tr;
2256
2257                     if(!topright_avail){
2258                         tr= ptr[3 - linesize]*0x01010101;
2259                         topright= (uint8_t*) &tr;
2260                     }
2261
2262                     h->pred4x4[ dir ](ptr, topright, linesize);
2263                     if(h->non_zero_count_cache[ scan8[i] ]){
2264                         if(s->codec_id == CODEC_ID_H264)
2265                             h264_add_idct_c(ptr, h->mb + i*16, linesize);
2266                         else
2267                             svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2268                     }
2269                 }
2270             }
2271         }else{
2272             h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2273             if(s->codec_id == CODEC_ID_H264)
2274                 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2275             else
2276                 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2277         }
2278     }else if(s->codec_id == CODEC_ID_H264){
2279         hl_motion(h, dest_y, dest_cb, dest_cr,
2280                   s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2281                   s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2282     }
2283
2284
2285     if(!IS_INTRA4x4(mb_type)){
2286         if(s->codec_id == CODEC_ID_H264){
2287             for(i=0; i<16; i++){
2288                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2289                     uint8_t * const ptr= dest_y + h->block_offset[i];
2290                     h264_add_idct_c(ptr, h->mb + i*16, linesize);
2291                 }
2292             }
2293         }else{
2294             for(i=0; i<16; i++){
2295                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2296                     uint8_t * const ptr= dest_y + h->block_offset[i];
2297                     svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2298                 }
2299             }
2300         }
2301     }
2302
2303     if(!(s->flags&CODEC_FLAG_GRAY)){
2304         chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2305         chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2306         if(s->codec_id == CODEC_ID_H264){
2307             for(i=16; i<16+4; i++){
2308                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2309                     uint8_t * const ptr= dest_cb + h->block_offset[i];
2310                     h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2311                 }
2312             }
2313             for(i=20; i<20+4; i++){
2314                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2315                     uint8_t * const ptr= dest_cr + h->block_offset[i];
2316                     h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2317                 }
2318             }
2319         }else{
2320             for(i=16; i<16+4; i++){
2321                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2322                     uint8_t * const ptr= dest_cb + h->block_offset[i];
2323                     svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2324                 }
2325             }
2326             for(i=20; i<20+4; i++){
2327                 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2328                     uint8_t * const ptr= dest_cr + h->block_offset[i];
2329                     svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2330                 }
2331             }
2332         }
2333     }
2334 }
2335
2336 static void decode_mb_cabac(H264Context *h){
2337 //    MpegEncContext * const s = &h->s;
2338 }
2339
2340 /**
2341  * fills the default_ref_list.
2342  */
2343 static int fill_default_ref_list(H264Context *h){
2344     MpegEncContext * const s = &h->s;
2345     int i;
2346     Picture sorted_short_ref[16];
2347
2348     if(h->slice_type==B_TYPE){
2349         int out_i;
2350         int limit= -1;
2351
2352         for(out_i=0; out_i<h->short_ref_count; out_i++){
2353             int best_i=-1;
2354             int best_poc=-1;
2355
2356             for(i=0; i<h->short_ref_count; i++){
2357                 const int poc= h->short_ref[i]->poc;
2358                 if(poc > limit && poc < best_poc){
2359                     best_poc= poc;
2360                     best_i= i;
2361                 }
2362             }
2363
2364             assert(best_i != -1);
2365
2366             limit= best_poc;
2367             sorted_short_ref[out_i]= *h->short_ref[best_i];
2368         }
2369     }
2370
2371     if(s->picture_structure == PICT_FRAME){
2372         if(h->slice_type==B_TYPE){
2373             const int current_poc= s->current_picture_ptr->poc;
2374             int list;
2375
2376             for(list=0; list<2; list++){
2377                 int index=0;
2378
2379                 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2380                     const int i2= list ? h->short_ref_count - i - 1 : i;
2381                     const int poc= sorted_short_ref[i2].poc;
2382
2383                     if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2384
2385                     if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2386                         h->default_ref_list[list][index  ]= sorted_short_ref[i2];
2387                         h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2388                     }
2389                 }
2390
2391                 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2392                     if(h->long_ref[i]->reference != 3) continue;
2393
2394                     h->default_ref_list[ list ][index  ]= *h->long_ref[i];
2395                     h->default_ref_list[ list ][index++].pic_id= i;;
2396                 }
2397
2398                 if(h->long_ref_count > 1 && h->short_ref_count==0){
2399                     Picture temp= h->default_ref_list[1][0];
2400                     h->default_ref_list[1][0] = h->default_ref_list[1][1];
2401                     h->default_ref_list[1][0] = temp;
2402                 }
2403
2404                 if(index < h->ref_count[ list ])
2405                     memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2406             }
2407         }else{
2408             int index=0;
2409             for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2410                 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2411                 h->default_ref_list[0][index  ]= *h->short_ref[i];
2412                 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2413             }
2414             for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2415                 if(h->long_ref[i]->reference != 3) continue;
2416                 h->default_ref_list[0][index  ]= *h->long_ref[i];
2417                 h->default_ref_list[0][index++].pic_id= i;;
2418             }
2419             if(index < h->ref_count[0])
2420                 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2421         }
2422     }else{ //FIELD
2423         if(h->slice_type==B_TYPE){
2424         }else{
2425             //FIXME second field balh
2426         }
2427     }
2428     return 0;
2429 }
2430
2431 static int decode_ref_pic_list_reordering(H264Context *h){
2432     MpegEncContext * const s = &h->s;
2433     int list;
2434
2435     if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2436
2437     for(list=0; list<2; list++){
2438         memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2439
2440         if(get_bits1(&s->gb)){
2441             int pred= h->curr_pic_num;
2442             int index;
2443
2444             for(index=0; ; index++){
2445                 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2446                 int pic_id;
2447                 int i;
2448
2449
2450                 if(index >= h->ref_count[list]){
2451                     av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2452                     return -1;
2453                 }
2454
2455                 if(reordering_of_pic_nums_idc<3){
2456                     if(reordering_of_pic_nums_idc<2){
2457                         const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2458
2459                         if(abs_diff_pic_num >= h->max_pic_num){
2460                             av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2461                             return -1;
2462                         }
2463
2464                         if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2465                         else                                pred+= abs_diff_pic_num;
2466                         pred &= h->max_pic_num - 1;
2467
2468                         for(i= h->ref_count[list]-1; i>=index; i--){
2469                             if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2470                                 break;
2471                         }
2472                     }else{
2473                         pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2474
2475                         for(i= h->ref_count[list]-1; i>=index; i--){
2476                             if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2477                                 break;
2478                         }
2479                     }
2480
2481                     if(i < index){
2482                         av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2483                         memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2484                     }else if(i > index){
2485                         Picture tmp= h->ref_list[list][i];
2486                         for(; i>index; i--){
2487                             h->ref_list[list][i]= h->ref_list[list][i-1];
2488                         }
2489                         h->ref_list[list][index]= tmp;
2490                     }
2491                 }else if(reordering_of_pic_nums_idc==3)
2492                     break;
2493                 else{
2494                     av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2495                     return -1;
2496                 }
2497             }
2498         }
2499
2500         if(h->slice_type!=B_TYPE) break;
2501     }
2502     return 0;
2503 }
2504
2505 static int pred_weight_table(H264Context *h){
2506     MpegEncContext * const s = &h->s;
2507     int list, i;
2508
2509     h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2510     h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2511
2512     for(list=0; list<2; list++){
2513         for(i=0; i<h->ref_count[list]; i++){
2514             int luma_weight_flag, chroma_weight_flag;
2515
2516             luma_weight_flag= get_bits1(&s->gb);
2517             if(luma_weight_flag){
2518                 h->luma_weight[list][i]= get_se_golomb(&s->gb);
2519                 h->luma_offset[list][i]= get_se_golomb(&s->gb);
2520             }
2521
2522             chroma_weight_flag= get_bits1(&s->gb);
2523             if(chroma_weight_flag){
2524                 int j;
2525                 for(j=0; j<2; j++){
2526                     h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2527                     h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2528                 }
2529             }
2530         }
2531         if(h->slice_type != B_TYPE) break;
2532     }
2533     return 0;
2534 }
2535
2536 /**
2537  * instantaneos decoder refresh.
2538  */
2539 static void idr(H264Context *h){
2540     int i;
2541
2542     for(i=0; i<h->long_ref_count; i++){
2543         h->long_ref[i]->reference=0;
2544         h->long_ref[i]= NULL;
2545     }
2546     h->long_ref_count=0;
2547
2548     for(i=0; i<h->short_ref_count; i++){
2549         h->short_ref[i]->reference=0;
2550         h->short_ref[i]= NULL;
2551     }
2552     h->short_ref_count=0;
2553 }
2554
2555 /**
2556  *
2557  * @return the removed picture or NULL if an error occures
2558  */
2559 static Picture * remove_short(H264Context *h, int frame_num){
2560     MpegEncContext * const s = &h->s;
2561     int i;
2562
2563     if(s->avctx->debug&FF_DEBUG_MMCO)
2564         av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
2565
2566     for(i=0; i<h->short_ref_count; i++){
2567         Picture *pic= h->short_ref[i];
2568         if(s->avctx->debug&FF_DEBUG_MMCO)
2569             av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
2570         if(pic->frame_num == frame_num){
2571             h->short_ref[i]= NULL;
2572             memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2573             h->short_ref_count--;
2574             return pic;
2575         }
2576     }
2577     return NULL;
2578 }
2579
2580 /**
2581  *
2582  * @return the removed picture or NULL if an error occures
2583  */
2584 static Picture * remove_long(H264Context *h, int i){
2585     Picture *pic;
2586
2587     if(i >= h->long_ref_count) return NULL;
2588     pic= h->long_ref[i];
2589     if(pic==NULL) return NULL;
2590
2591     h->long_ref[i]= NULL;
2592     memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2593     h->long_ref_count--;
2594
2595     return pic;
2596 }
2597
2598 /**
2599  * Executes the reference picture marking (memory management control operations).
2600  */
2601 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2602     MpegEncContext * const s = &h->s;
2603     int i;
2604     int current_is_long=0;
2605     Picture *pic;
2606
2607     if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2608         av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
2609
2610     for(i=0; i<mmco_count; i++){
2611         if(s->avctx->debug&FF_DEBUG_MMCO)
2612             av_log(h->s.avctx, AV_LOG_DEBUG, "mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
2613
2614         switch(mmco[i].opcode){
2615         case MMCO_SHORT2UNUSED:
2616             pic= remove_short(h, mmco[i].short_frame_num);
2617             if(pic==NULL) return -1;
2618             pic->reference= 0;
2619             break;
2620         case MMCO_SHORT2LONG:
2621             pic= remove_long(h, mmco[i].long_index);
2622             if(pic) pic->reference=0;
2623
2624             h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2625             h->long_ref[ mmco[i].long_index ]->long_ref=1;
2626             break;
2627         case MMCO_LONG2UNUSED:
2628             pic= remove_long(h, mmco[i].long_index);
2629             if(pic==NULL) return -1;
2630             pic->reference= 0;
2631             break;
2632         case MMCO_LONG:
2633             pic= remove_long(h, mmco[i].long_index);
2634             if(pic) pic->reference=0;
2635
2636             h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2637             h->long_ref[ mmco[i].long_index ]->long_ref=1;
2638             h->long_ref_count++;
2639
2640             current_is_long=1;
2641             break;
2642         case MMCO_SET_MAX_LONG:
2643             assert(mmco[i].long_index <= 16);
2644             while(mmco[i].long_index < h->long_ref_count){
2645                 pic= remove_long(h, mmco[i].long_index);
2646                 pic->reference=0;
2647             }
2648             while(mmco[i].long_index > h->long_ref_count){
2649                 h->long_ref[ h->long_ref_count++ ]= NULL;
2650             }
2651             break;
2652         case MMCO_RESET:
2653             while(h->short_ref_count){
2654                 pic= remove_short(h, h->short_ref[0]->frame_num);
2655                 pic->reference=0;
2656             }
2657             while(h->long_ref_count){
2658                 pic= remove_long(h, h->long_ref_count-1);
2659                 pic->reference=0;
2660             }
2661             break;
2662         default: assert(0);
2663         }
2664     }
2665
2666     if(!current_is_long){
2667         pic= remove_short(h, s->current_picture_ptr->frame_num);
2668         if(pic){
2669             pic->reference=0;
2670             av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
2671         }
2672
2673         if(h->short_ref_count)
2674             memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2675
2676         h->short_ref[0]= s->current_picture_ptr;
2677         h->short_ref[0]->long_ref=0;
2678         h->short_ref_count++;
2679     }
2680
2681     return 0;
2682 }
2683
2684 static int decode_ref_pic_marking(H264Context *h){
2685     MpegEncContext * const s = &h->s;
2686     int i;
2687
2688     if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2689         s->broken_link= get_bits1(&s->gb) -1;
2690         h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2691         if(h->mmco[0].long_index == -1)
2692             h->mmco_index= 0;
2693         else{
2694             h->mmco[0].opcode= MMCO_LONG;
2695             h->mmco_index= 1;
2696         }
2697     }else{
2698         if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2699             for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) {
2700                 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2701
2702                 h->mmco[i].opcode= opcode;
2703                 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2704                     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
2705 /*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2706                         fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2707                         return -1;
2708                     }*/
2709                 }
2710                 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2711                     h->mmco[i].long_index= get_ue_golomb(&s->gb);
2712                     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){
2713                         av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
2714                         return -1;
2715                     }
2716                 }
2717
2718                 if(opcode > MMCO_LONG){
2719                     av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
2720                     return -1;
2721                 }
2722             }
2723             h->mmco_index= i;
2724         }else{
2725             assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2726
2727             if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2728                 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2729                 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2730                 h->mmco_index= 1;
2731             }else
2732                 h->mmco_index= 0;
2733         }
2734     }
2735
2736     return 0;
2737 }
2738
2739 static int init_poc(H264Context *h){
2740     MpegEncContext * const s = &h->s;
2741     const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2742     int field_poc[2];
2743
2744     if(h->nal_unit_type == NAL_IDR_SLICE){
2745         h->frame_num_offset= 0;
2746     }else{
2747         if(h->frame_num < h->prev_frame_num)
2748             h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2749         else
2750             h->frame_num_offset= h->prev_frame_num_offset;
2751     }
2752
2753     if(h->sps.poc_type==0){
2754         const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2755
2756         if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2757             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2758         else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2759             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2760         else
2761             h->poc_msb = h->prev_poc_msb;
2762 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2763         field_poc[0] =
2764         field_poc[1] = h->poc_msb + h->poc_lsb;
2765         if(s->picture_structure == PICT_FRAME)
2766             field_poc[1] += h->delta_poc_bottom;
2767     }else if(h->sps.poc_type==1){
2768         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2769         int i;
2770
2771         if(h->sps.poc_cycle_length != 0)
2772             abs_frame_num = h->frame_num_offset + h->frame_num;
2773         else
2774             abs_frame_num = 0;
2775
2776         if(h->nal_ref_idc==0 && abs_frame_num > 0)
2777             abs_frame_num--;
2778
2779         expected_delta_per_poc_cycle = 0;
2780         for(i=0; i < h->sps.poc_cycle_length; i++)
2781             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2782
2783         if(abs_frame_num > 0){
2784             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2785             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2786
2787             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2788             for(i = 0; i <= frame_num_in_poc_cycle; i++)
2789                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2790         } else
2791             expectedpoc = 0;
2792
2793         if(h->nal_ref_idc == 0)
2794             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2795
2796         field_poc[0] = expectedpoc + h->delta_poc[0];
2797         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2798
2799         if(s->picture_structure == PICT_FRAME)
2800             field_poc[1] += h->delta_poc[1];
2801     }else{
2802         int poc;
2803         if(h->nal_unit_type == NAL_IDR_SLICE){
2804             poc= 0;
2805         }else{
2806             if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2807             else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2808         }
2809         field_poc[0]= poc;
2810         field_poc[1]= poc;
2811     }
2812
2813     if(s->picture_structure != PICT_BOTTOM_FIELD)
2814         s->current_picture_ptr->field_poc[0]= field_poc[0];
2815     if(s->picture_structure != PICT_TOP_FIELD)
2816         s->current_picture_ptr->field_poc[1]= field_poc[1];
2817     if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2818         s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2819
2820     return 0;
2821 }
2822
2823 /**
2824  * decodes a slice header.
2825  * this will allso call MPV_common_init() and frame_start() as needed
2826  */
2827 static int decode_slice_header(H264Context *h){
2828     MpegEncContext * const s = &h->s;
2829     int first_mb_in_slice, pps_id;
2830     int num_ref_idx_active_override_flag;
2831     static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2832
2833     s->current_picture.reference= h->nal_ref_idc != 0;
2834
2835     first_mb_in_slice= get_ue_golomb(&s->gb);
2836
2837     h->slice_type= get_ue_golomb(&s->gb);
2838     if(h->slice_type > 9){
2839         av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
2840     }
2841     if(h->slice_type > 4){
2842         h->slice_type -= 5;
2843         h->slice_type_fixed=1;
2844     }else
2845         h->slice_type_fixed=0;
2846
2847     h->slice_type= slice_type_map[ h->slice_type ];
2848
2849     s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
2850
2851     pps_id= get_ue_golomb(&s->gb);
2852     if(pps_id>255){
2853         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2854         return -1;
2855     }
2856     h->pps= h->pps_buffer[pps_id];
2857     if(h->pps.slice_group_count == 0){
2858         av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
2859         return -1;
2860     }
2861
2862     h->sps= h->sps_buffer[ h->pps.sps_id ];
2863     if(h->sps.log2_max_frame_num == 0){
2864         av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
2865         return -1;
2866     }
2867
2868     s->mb_width= h->sps.mb_width;
2869     s->mb_height= h->sps.mb_height;
2870
2871     h->b_stride=  s->mb_width*4;
2872     h->b8_stride= s->mb_width*2;
2873
2874     s->mb_x = first_mb_in_slice % s->mb_width;
2875     s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
2876
2877     s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
2878     if(h->sps.frame_mbs_only_flag)
2879         s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
2880     else
2881         s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
2882
2883     if (s->context_initialized
2884         && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
2885         free_tables(h);
2886         MPV_common_end(s);
2887     }
2888     if (!s->context_initialized) {
2889         if (MPV_common_init(s) < 0)
2890             return -1;
2891
2892         alloc_tables(h);
2893
2894         s->avctx->width = s->width;
2895         s->avctx->height = s->height;
2896         s->avctx->sample_aspect_ratio= h->sps.sar;
2897     }
2898
2899     if(first_mb_in_slice == 0){
2900         frame_start(h);
2901     }
2902
2903     s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
2904     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2905
2906     if(h->sps.frame_mbs_only_flag){
2907         s->picture_structure= PICT_FRAME;
2908     }else{
2909         if(get_bits1(&s->gb)) //field_pic_flag
2910             s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2911         else
2912             s->picture_structure= PICT_FRAME;
2913     }
2914
2915     if(s->picture_structure==PICT_FRAME){
2916         h->curr_pic_num=   h->frame_num;
2917         h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2918     }else{
2919         h->curr_pic_num= 2*h->frame_num;
2920         h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2921     }
2922
2923     if(h->nal_unit_type == NAL_IDR_SLICE){
2924         get_ue_golomb(&s->gb); /* idr_pic_id */
2925     }
2926
2927     if(h->sps.poc_type==0){
2928         h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2929
2930         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2931             h->delta_poc_bottom= get_se_golomb(&s->gb);
2932         }
2933     }
2934
2935     if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2936         h->delta_poc[0]= get_se_golomb(&s->gb);
2937
2938         if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2939             h->delta_poc[1]= get_se_golomb(&s->gb);
2940     }
2941
2942     init_poc(h);
2943
2944     if(h->pps.redundant_pic_cnt_present){
2945         h->redundant_pic_count= get_ue_golomb(&s->gb);
2946     }
2947
2948     //set defaults, might be overriden a few line later
2949     h->ref_count[0]= h->pps.ref_count[0];
2950     h->ref_count[1]= h->pps.ref_count[1];
2951
2952     if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
2953         if(h->slice_type == B_TYPE){
2954             h->direct_spatial_mv_pred= get_bits1(&s->gb);
2955         }
2956         num_ref_idx_active_override_flag= get_bits1(&s->gb);
2957
2958         if(num_ref_idx_active_override_flag){
2959             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2960             if(h->slice_type==B_TYPE)
2961                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2962
2963             if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
2964                 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2965                 return -1;
2966             }
2967         }
2968     }
2969
2970     if(first_mb_in_slice == 0){
2971         fill_default_ref_list(h);
2972     }
2973
2974     decode_ref_pic_list_reordering(h);
2975
2976     if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
2977        || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
2978         pred_weight_table(h);
2979
2980     if(s->current_picture.reference)
2981         decode_ref_pic_marking(h);
2982     //FIXME CABAC stuff
2983
2984     s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta
2985     //FIXME qscale / qp ... stuff
2986     if(h->slice_type == SP_TYPE){
2987         get_bits1(&s->gb); /* sp_for_switch_flag */
2988     }
2989     if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
2990         get_se_golomb(&s->gb); /* slice_qs_delta */
2991     }
2992
2993     if( h->pps.deblocking_filter_parameters_present ) {
2994         h->disable_deblocking_filter_idc= get_ue_golomb(&s->gb);
2995         if( h->disable_deblocking_filter_idc  !=  1 ) {
2996             h->slice_alpha_c0_offset_div2= get_se_golomb(&s->gb);
2997             h->slice_beta_offset_div2= get_se_golomb(&s->gb);
2998         }
2999     }else
3000         h->disable_deblocking_filter_idc= 0;
3001
3002 #if 0 //FMO
3003     if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3004         slice_group_change_cycle= get_bits(&s->gb, ?);
3005 #endif
3006
3007     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3008         av_log(h->s.avctx, AV_LOG_DEBUG, "mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n",
3009                first_mb_in_slice,
3010                av_get_pict_type_char(h->slice_type),
3011                pps_id, h->frame_num,
3012                s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3013                h->ref_count[0], h->ref_count[1],
3014                s->qscale,
3015                h->disable_deblocking_filter_idc
3016                );
3017     }
3018
3019     return 0;
3020 }
3021
3022 /**
3023  *
3024  */
3025 static inline int get_level_prefix(GetBitContext *gb){
3026     unsigned int buf;
3027     int log;
3028
3029     OPEN_READER(re, gb);
3030     UPDATE_CACHE(re, gb);
3031     buf=GET_CACHE(re, gb);
3032
3033     log= 32 - av_log2(buf);
3034 #ifdef TRACE
3035     print_bin(buf>>(32-log), log);
3036     printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3037 #endif
3038
3039     LAST_SKIP_BITS(re, gb, log);
3040     CLOSE_READER(re, gb);
3041
3042     return log-1;
3043 }
3044
3045 /**
3046  * decodes a residual block.
3047  * @param n block index
3048  * @param scantable scantable
3049  * @param max_coeff number of coefficients in the block
3050  * @return <0 if an error occured
3051  */
3052 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3053     MpegEncContext * const s = &h->s;
3054     const uint16_t *qmul= dequant_coeff[qp];
3055     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};
3056     int level[16], run[16];
3057     int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3058
3059     //FIXME put trailing_onex into the context
3060
3061     if(n == CHROMA_DC_BLOCK_INDEX){
3062         coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3063         total_coeff= coeff_token>>2;
3064     }else{
3065         if(n == LUMA_DC_BLOCK_INDEX){
3066             total_coeff= pred_non_zero_count(h, 0);
3067             coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3068             total_coeff= coeff_token>>2;
3069         }else{
3070             total_coeff= pred_non_zero_count(h, n);
3071             coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3072             total_coeff= coeff_token>>2;
3073             h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3074         }
3075     }
3076
3077     //FIXME set last_non_zero?
3078
3079     if(total_coeff==0)
3080         return 0;
3081
3082     trailing_ones= coeff_token&3;
3083     tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3084     assert(total_coeff<=16);
3085
3086     for(i=0; i<trailing_ones; i++){
3087         level[i]= 1 - 2*get_bits1(gb);
3088     }
3089
3090     suffix_length= total_coeff > 10 && trailing_ones < 3;
3091
3092     for(; i<total_coeff; i++){
3093         const int prefix= get_level_prefix(gb);
3094         int level_code, mask;
3095
3096         if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3097             if(suffix_length)
3098                 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3099             else
3100                 level_code= (prefix<<suffix_length); //part
3101         }else if(prefix==14){
3102             if(suffix_length)
3103                 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3104             else
3105                 level_code= prefix + get_bits(gb, 4); //part
3106         }else if(prefix==15){
3107             level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3108             if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3109         }else{
3110             av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3111             return -1;
3112         }
3113
3114         if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3115
3116         mask= -(level_code&1);
3117         level[i]= (((2+level_code)>>1) ^ mask) - mask;
3118
3119         if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3120
3121 #if 1
3122         if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3123 #else
3124         if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3125         ? == prefix > 2 or sth
3126 #endif
3127         tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3128     }
3129
3130     if(total_coeff == max_coeff)
3131         zeros_left=0;
3132     else{
3133         if(n == CHROMA_DC_BLOCK_INDEX)
3134             zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3135         else
3136             zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3137     }
3138
3139     for(i=0; i<total_coeff-1; i++){
3140         if(zeros_left <=0)
3141             break;
3142         else if(zeros_left < 7){
3143             run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3144         }else{
3145             run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3146         }
3147         zeros_left -= run[i];
3148     }
3149
3150     if(zeros_left<0){
3151         av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3152         return -1;
3153     }
3154
3155     for(; i<total_coeff-1; i++){
3156         run[i]= 0;
3157     }
3158
3159     run[i]= zeros_left;
3160
3161     coeff_num=-1;
3162     if(n > 24){
3163         for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3164             int j;
3165
3166             coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3167             j= scantable[ coeff_num ];
3168
3169             block[j]= level[i];
3170         }
3171     }else{
3172         for(i=total_coeff-1; i>=0; i--){ //FIXME merge into  rundecode?
3173             int j;
3174
3175             coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3176             j= scantable[ coeff_num ];
3177
3178             block[j]= level[i] * qmul[j];
3179 //            printf("%d %d  ", block[j], qmul[j]);
3180         }
3181     }
3182     return 0;
3183 }
3184
3185 /**
3186  * decodes a macroblock
3187  * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3188  */
3189 static int decode_mb(H264Context *h){
3190     MpegEncContext * const s = &h->s;
3191     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3192     int mb_type, partition_count, cbp;
3193
3194     s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?
3195
3196     tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3197     cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3198                 down the code */
3199     if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3200         if(s->mb_skip_run==-1)
3201             s->mb_skip_run= get_ue_golomb(&s->gb);
3202
3203         if (s->mb_skip_run--) {
3204             int mx, my;
3205             /* skip mb */
3206 //FIXME b frame
3207             mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3208
3209             memset(h->non_zero_count[mb_xy], 0, 16);
3210             memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3211
3212             if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3213                 h->mb_field_decoding_flag= get_bits1(&s->gb);
3214             }
3215
3216             if(h->mb_field_decoding_flag)
3217                 mb_type|= MB_TYPE_INTERLACED;
3218
3219             fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3220             pred_pskip_motion(h, &mx, &my);
3221             fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3222             fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3223             write_back_motion(h, mb_type);
3224
3225             s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3226             h->slice_table[ mb_xy ]= h->slice_num;
3227
3228             h->prev_mb_skiped= 1;
3229             return 0;
3230         }
3231     }
3232     if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3233         if((s->mb_y&1)==0)
3234             h->mb_field_decoding_flag = get_bits1(&s->gb);
3235     }else
3236         h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3237
3238     h->prev_mb_skiped= 0;
3239
3240     mb_type= get_ue_golomb(&s->gb);
3241     if(h->slice_type == B_TYPE){
3242         if(mb_type < 23){
3243             partition_count= b_mb_type_info[mb_type].partition_count;
3244             mb_type=         b_mb_type_info[mb_type].type;
3245         }else{
3246             mb_type -= 23;
3247             goto decode_intra_mb;
3248         }
3249     }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3250         if(mb_type < 5){
3251             partition_count= p_mb_type_info[mb_type].partition_count;
3252             mb_type=         p_mb_type_info[mb_type].type;
3253         }else{
3254             mb_type -= 5;
3255             goto decode_intra_mb;
3256         }
3257     }else{
3258        assert(h->slice_type == I_TYPE);
3259 decode_intra_mb:
3260         if(mb_type > 25){
3261             av_log(h->s.avctx, AV_LOG_ERROR, "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);
3262             return -1;
3263         }
3264         partition_count=0;
3265         cbp= i_mb_type_info[mb_type].cbp;
3266         h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3267         mb_type= i_mb_type_info[mb_type].type;
3268     }
3269
3270     if(h->mb_field_decoding_flag)
3271         mb_type |= MB_TYPE_INTERLACED;
3272
3273     s->current_picture.mb_type[mb_xy]= mb_type;
3274     h->slice_table[ mb_xy ]= h->slice_num;
3275
3276     if(IS_INTRA_PCM(mb_type)){
3277         const uint8_t *ptr;
3278         int x, y;
3279
3280         // we assume these blocks are very rare so we dont optimize it
3281         align_get_bits(&s->gb);
3282
3283         ptr= s->gb.buffer + get_bits_count(&s->gb);
3284
3285         for(y=0; y<16; y++){
3286             const int index= 4*(y&3) + 64*(y>>2);
3287             for(x=0; x<16; x++){
3288                 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3289             }
3290         }
3291         for(y=0; y<8; y++){
3292             const int index= 256 + 4*(y&3) + 32*(y>>2);
3293             for(x=0; x<8; x++){
3294                 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3295             }
3296         }
3297         for(y=0; y<8; y++){
3298             const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3299             for(x=0; x<8; x++){
3300                 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3301             }
3302         }
3303
3304         skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3305
3306         memset(h->non_zero_count[mb_xy], 16, 16);
3307
3308         return 0;
3309     }
3310
3311     fill_caches(h, mb_type);
3312
3313     //mb_pred
3314     if(IS_INTRA(mb_type)){
3315 //            init_top_left_availability(h);
3316             if(IS_INTRA4x4(mb_type)){
3317                 int i;
3318
3319 //                fill_intra4x4_pred_table(h);
3320                 for(i=0; i<16; i++){
3321                     const int mode_coded= !get_bits1(&s->gb);
3322                     const int predicted_mode=  pred_intra_mode(h, i);
3323                     int mode;
3324
3325                     if(mode_coded){
3326                         const int rem_mode= get_bits(&s->gb, 3);
3327                         if(rem_mode<predicted_mode)
3328                             mode= rem_mode;
3329                         else
3330                             mode= rem_mode + 1;
3331                     }else{
3332                         mode= predicted_mode;
3333                     }
3334
3335                     h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3336                 }
3337                 write_back_intra_pred_mode(h);
3338                 if( check_intra4x4_pred_mode(h) < 0)
3339                     return -1;
3340             }else{
3341                 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3342                 if(h->intra16x16_pred_mode < 0)
3343                     return -1;
3344             }
3345             h->chroma_pred_mode= get_ue_golomb(&s->gb);
3346
3347             h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3348             if(h->chroma_pred_mode < 0)
3349                 return -1;
3350     }else if(partition_count==4){
3351         int i, j, sub_partition_count[4], list, ref[2][4];
3352
3353         if(h->slice_type == B_TYPE){
3354             for(i=0; i<4; i++){
3355                 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3356                 if(h->sub_mb_type[i] >=13){
3357                     av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
3358                     return -1;
3359                 }
3360                 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3361                 h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3362             }
3363         }else{
3364             assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3365             for(i=0; i<4; i++){
3366                 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3367                 if(h->sub_mb_type[i] >=4){
3368                     av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
3369                     return -1;
3370                 }
3371                 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3372                 h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3373             }
3374         }
3375
3376         for(list=0; list<2; list++){
3377             const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3378             if(ref_count == 0) continue;
3379             for(i=0; i<4; i++){
3380                 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3381                     ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3382                 }else{
3383                  //FIXME
3384                     ref[list][i] = -1;
3385                 }
3386             }
3387         }
3388
3389         for(list=0; list<2; list++){
3390             const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3391             if(ref_count == 0) continue;
3392
3393             for(i=0; i<4; i++){
3394                 h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3395                 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3396
3397                 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3398                     const int sub_mb_type= h->sub_mb_type[i];
3399                     const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3400                     for(j=0; j<sub_partition_count[i]; j++){
3401                         int mx, my;
3402                         const int index= 4*i + block_width*j;
3403                         int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3404                         pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3405                         mx += get_se_golomb(&s->gb);
3406                         my += get_se_golomb(&s->gb);
3407                         tprintf("final mv:%d %d\n", mx, my);
3408
3409                         if(IS_SUB_8X8(sub_mb_type)){
3410                             mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
3411                             mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3412                             mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
3413                             mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3414                         }else if(IS_SUB_8X4(sub_mb_type)){
3415                             mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3416                             mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3417                         }else if(IS_SUB_4X8(sub_mb_type)){
3418                             mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3419                             mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3420                         }else{
3421                             assert(IS_SUB_4X4(sub_mb_type));
3422                             mv_cache[ 0 ][0]= mx;
3423                             mv_cache[ 0 ][1]= my;
3424                         }
3425                     }
3426                 }else{
3427                     uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3428                     p[0] = p[1]=
3429                     p[8] = p[9]= 0;
3430                 }
3431             }
3432         }
3433     }else if(!IS_DIRECT(mb_type)){
3434         int list, mx, my, i;
3435          //FIXME we should set ref_idx_l? to 0 if we use that later ...
3436         if(IS_16X16(mb_type)){
3437             for(list=0; list<2; list++){
3438                 if(h->ref_count[0]>0){
3439                     if(IS_DIR(mb_type, 0, list)){
3440                         const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3441                         fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3442                     }
3443                 }
3444             }
3445             for(list=0; list<2; list++){
3446                 if(IS_DIR(mb_type, 0, list)){
3447                     pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3448                     mx += get_se_golomb(&s->gb);
3449                     my += get_se_golomb(&s->gb);
3450                     tprintf("final mv:%d %d\n", mx, my);
3451
3452                     fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
3453                 }
3454             }
3455         }
3456         else if(IS_16X8(mb_type)){
3457             for(list=0; list<2; list++){
3458                 if(h->ref_count[list]>0){
3459                     for(i=0; i<2; i++){
3460                         if(IS_DIR(mb_type, i, list)){
3461                             const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3462                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3463                         }
3464                     }
3465                 }
3466             }
3467             for(list=0; list<2; list++){
3468                 for(i=0; i<2; i++){
3469                     if(IS_DIR(mb_type, i, list)){
3470                         pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3471                         mx += get_se_golomb(&s->gb);
3472                         my += get_se_golomb(&s->gb);
3473                         tprintf("final mv:%d %d\n", mx, my);
3474
3475                         fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
3476                     }
3477                 }
3478             }
3479         }else{
3480             assert(IS_8X16(mb_type));
3481             for(list=0; list<2; list++){
3482                 if(h->ref_count[list]>0){
3483                     for(i=0; i<2; i++){
3484                         if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3485                             const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3486                             fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3487                         }
3488                     }
3489                 }
3490             }
3491             for(list=0; list<2; list++){
3492                 for(i=0; i<2; i++){
3493                     if(IS_DIR(mb_type, i, list)){
3494                         pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3495                         mx += get_se_golomb(&s->gb);
3496                         my += get_se_golomb(&s->gb);
3497                         tprintf("final mv:%d %d\n", mx, my);
3498
3499                         fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
3500                     }
3501                 }
3502             }
3503         }
3504     }
3505
3506     if(IS_INTER(mb_type))
3507         write_back_motion(h, mb_type);
3508
3509     if(!IS_INTRA16x16(mb_type)){
3510         cbp= get_ue_golomb(&s->gb);
3511         if(cbp > 47){
3512             av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3513             return -1;
3514         }
3515
3516         if(IS_INTRA4x4(mb_type))
3517             cbp= golomb_to_intra4x4_cbp[cbp];
3518         else
3519             cbp= golomb_to_inter_cbp[cbp];
3520     }
3521
3522     if(cbp || IS_INTRA16x16(mb_type)){
3523         int i8x8, i4x4, chroma_idx;
3524         int chroma_qp, dquant;
3525         GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3526         const uint8_t *scan, *dc_scan;
3527
3528 //        fill_non_zero_count_cache(h);
3529
3530         if(IS_INTERLACED(mb_type)){
3531             scan= field_scan;
3532             dc_scan= luma_dc_field_scan;
3533         }else{
3534             scan= zigzag_scan;
3535             dc_scan= luma_dc_zigzag_scan;
3536         }
3537
3538         dquant= get_se_golomb(&s->gb);
3539
3540         if( dquant > 25 || dquant < -26 ){
3541             av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3542             return -1;
3543         }
3544
3545         s->qscale += dquant;
3546         if(((unsigned)s->qscale) > 51){
3547             if(s->qscale<0) s->qscale+= 52;
3548             else            s->qscale-= 52;
3549         }
3550
3551         h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3552         if(IS_INTRA16x16(mb_type)){
3553             if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3554                 return -1; //FIXME continue if partotioned and other retirn -1 too
3555             }
3556
3557             assert((cbp&15) == 0 || (cbp&15) == 15);
3558
3559             if(cbp&15){
3560                 for(i8x8=0; i8x8<4; i8x8++){
3561                     for(i4x4=0; i4x4<4; i4x4++){
3562                         const int index= i4x4 + 4*i8x8;
3563                         if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3564                             return -1;
3565                         }
3566                     }
3567                 }
3568             }else{
3569                 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
3570             }
3571         }else{
3572             for(i8x8=0; i8x8<4; i8x8++){
3573                 if(cbp & (1<<i8x8)){
3574                     for(i4x4=0; i4x4<4; i4x4++){
3575                         const int index= i4x4 + 4*i8x8;
3576
3577                         if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3578                             return -1;
3579                         }
3580                     }
3581                 }else{
3582                     uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3583                     nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3584                 }
3585             }
3586         }
3587
3588         if(cbp&0x30){
3589             for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3590                 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3591                     return -1;
3592                 }
3593         }
3594
3595         if(cbp&0x20){
3596             for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3597                 for(i4x4=0; i4x4<4; i4x4++){
3598                     const int index= 16 + 4*chroma_idx + i4x4;
3599                     if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3600                         return -1;
3601                     }
3602                 }
3603             }
3604         }else{
3605             uint8_t * const nnz= &h->non_zero_count_cache[0];
3606             nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3607             nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3608         }
3609     }else{
3610         memset(&h->non_zero_count_cache[8], 0, 8*5);
3611     }
3612     write_back_non_zero_count(h);
3613
3614     return 0;
3615 }
3616
3617 static int decode_slice(H264Context *h){
3618     MpegEncContext * const s = &h->s;
3619     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3620
3621     s->mb_skip_run= -1;
3622
3623 #if 1
3624     for(;;){
3625         int ret= decode_mb(h);
3626
3627         hl_decode_mb(h);
3628
3629         if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
3630             s->mb_y++;
3631             ret= decode_mb(h);
3632
3633             hl_decode_mb(h);
3634             s->mb_y--;
3635         }
3636
3637         if(ret<0){
3638             av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3639             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);
3640
3641             return -1;
3642         }
3643
3644         if(++s->mb_x >= s->mb_width){
3645             s->mb_x=0;
3646             ff_draw_horiz_band(s, 16*s->mb_y, 16);
3647             if(++s->mb_y >= s->mb_height){
3648                 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3649
3650                 if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3651                     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);
3652
3653                     return 0;
3654                 }else{
3655                     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);
3656
3657                     return -1;
3658                 }
3659             }
3660         }
3661
3662         if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3663             if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3664                 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);
3665
3666                 return 0;
3667             }else{
3668                 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);
3669
3670                 return -1;
3671             }
3672         }
3673     }
3674 #endif
3675 #if 0
3676     for(;s->mb_y < s->mb_height; s->mb_y++){
3677         for(;s->mb_x < s->mb_width; s->mb_x++){
3678             int ret= decode_mb(h);
3679
3680             hl_decode_mb(h);
3681
3682             if(ret<0){
3683                 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3684                 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);
3685
3686                 return -1;
3687             }
3688
3689             if(++s->mb_x >= s->mb_width){
3690                 s->mb_x=0;
3691                 if(++s->mb_y >= s->mb_height){
3692                     if(get_bits_count(s->gb) == s->gb.size_in_bits){
3693                         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);
3694
3695                         return 0;
3696                     }else{
3697                         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);
3698
3699                         return -1;
3700                     }
3701                 }
3702             }
3703
3704             if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
3705                 if(get_bits_count(s->gb) == s->gb.size_in_bits){
3706                     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);
3707
3708                     return 0;
3709                 }else{
3710                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3711
3712                     return -1;
3713                 }
3714             }
3715         }
3716         s->mb_x=0;
3717         ff_draw_horiz_band(s, 16*s->mb_y, 16);
3718     }
3719 #endif
3720     return -1; //not reached
3721 }
3722
3723 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
3724     MpegEncContext * const s = &h->s;
3725     int aspect_ratio_info_present_flag, aspect_ratio_idc;
3726
3727     aspect_ratio_info_present_flag= get_bits1(&s->gb);
3728
3729     if( aspect_ratio_info_present_flag ) {
3730         aspect_ratio_idc= get_bits(&s->gb, 8);
3731         if( aspect_ratio_idc == EXTENDED_SAR ) {
3732             sps->sar.num= get_bits(&s->gb, 16);
3733             sps->sar.den= get_bits(&s->gb, 16);
3734         }else if(aspect_ratio_idc < 16){
3735             sps->sar=  pixel_aspect[aspect_ratio_idc];
3736         }else{
3737             av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
3738             return -1;
3739         }
3740     }else{
3741         sps->sar.num=
3742         sps->sar.den= 0;
3743     }
3744 //            s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
3745 #if 0
3746 | overscan_info_present_flag                        |0  |u(1)    |
3747 | if( overscan_info_present_flag )                  |   |        |
3748 |  overscan_appropriate_flag                        |0  |u(1)    |
3749 | video_signal_type_present_flag                    |0  |u(1)    |
3750 | if( video_signal_type_present_flag ) {            |   |        |
3751 |  video_format                                     |0  |u(3)    |
3752 |  video_full_range_flag                            |0  |u(1)    |
3753 |  colour_description_present_flag                  |0  |u(1)    |
3754 |  if( colour_description_present_flag ) {          |   |        |
3755 |   colour_primaries                                |0  |u(8)    |
3756 |   transfer_characteristics                        |0  |u(8)    |
3757 |   matrix_coefficients                             |0  |u(8)    |
3758 |  }                                                |   |        |
3759 | }                                                 |   |        |
3760 | chroma_location_info_present_flag                 |0  |u(1)    |
3761 | if ( chroma_location_info_present_flag ) {        |   |        |
3762 |  chroma_sample_location_type_top_field            |0  |ue(v)   |
3763 |  chroma_sample_location_type_bottom_field         |0  |ue(v)   |
3764 | }                                                 |   |        |
3765 | timing_info_present_flag                          |0  |u(1)    |
3766 | if( timing_info_present_flag ) {                  |   |        |
3767 |  num_units_in_tick                                |0  |u(32)   |
3768 |  time_scale                                       |0  |u(32)   |
3769 |  fixed_frame_rate_flag                            |0  |u(1)    |
3770 | }                                                 |   |        |
3771 | nal_hrd_parameters_present_flag                   |0  |u(1)    |
3772 | if( nal_hrd_parameters_present_flag  = =  1)      |   |        |
3773 |  hrd_parameters( )                                |   |        |
3774 | vcl_hrd_parameters_present_flag                   |0  |u(1)    |
3775 | if( vcl_hrd_parameters_present_flag  = =  1)      |   |        |
3776 |  hrd_parameters( )                                |   |        |
3777 | if( ( nal_hrd_parameters_present_flag  = =  1  | ||   |        |
3778 |                                                   |   |        |
3779 |( vcl_hrd_parameters_present_flag  = =  1 ) )      |   |        |
3780 |  low_delay_hrd_flag                               |0  |u(1)    |
3781 | bitstream_restriction_flag                        |0  |u(1)    |
3782 | if( bitstream_restriction_flag ) {                |0  |u(1)    |
3783 |  motion_vectors_over_pic_boundaries_flag          |0  |u(1)    |
3784 |  max_bytes_per_pic_denom                          |0  |ue(v)   |
3785 |  max_bits_per_mb_denom                            |0  |ue(v)   |
3786 |  log2_max_mv_length_horizontal                    |0  |ue(v)   |
3787 |  log2_max_mv_length_vertical                      |0  |ue(v)   |
3788 |  num_reorder_frames                               |0  |ue(v)   |
3789 |  max_dec_frame_buffering                          |0  |ue(v)   |
3790 | }                                                 |   |        |
3791 |}                                                  |   |        |
3792 #endif
3793     return 0;
3794 }
3795
3796 static inline int decode_seq_parameter_set(H264Context *h){
3797     MpegEncContext * const s = &h->s;
3798     int profile_idc, level_idc;
3799     int sps_id, i;
3800     SPS *sps;
3801
3802     profile_idc= get_bits(&s->gb, 8);
3803     get_bits1(&s->gb);   //constraint_set0_flag
3804     get_bits1(&s->gb);   //constraint_set1_flag
3805     get_bits1(&s->gb);   //constraint_set2_flag
3806     get_bits(&s->gb, 5); // reserved
3807     level_idc= get_bits(&s->gb, 8);
3808     sps_id= get_ue_golomb(&s->gb);
3809
3810     sps= &h->sps_buffer[ sps_id ];
3811     sps->profile_idc= profile_idc;
3812     sps->level_idc= level_idc;
3813
3814     sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
3815     sps->poc_type= get_ue_golomb(&s->gb);
3816
3817     if(sps->poc_type == 0){ //FIXME #define
3818         sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
3819     } else if(sps->poc_type == 1){//FIXME #define
3820         sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
3821         sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
3822         sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
3823         sps->poc_cycle_length= get_ue_golomb(&s->gb);
3824
3825         for(i=0; i<sps->poc_cycle_length; i++)
3826             sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
3827     }
3828     if(sps->poc_type > 2){
3829         av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
3830         return -1;
3831     }
3832
3833     sps->ref_frame_count= get_ue_golomb(&s->gb);
3834     sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
3835     sps->mb_width= get_ue_golomb(&s->gb) + 1;
3836     sps->mb_height= get_ue_golomb(&s->gb) + 1;
3837     sps->frame_mbs_only_flag= get_bits1(&s->gb);
3838     if(!sps->frame_mbs_only_flag)
3839         sps->mb_aff= get_bits1(&s->gb);
3840     else
3841         sps->mb_aff= 0;
3842
3843     sps->direct_8x8_inference_flag= get_bits1(&s->gb);
3844
3845     sps->crop= get_bits1(&s->gb);
3846     if(sps->crop){
3847         sps->crop_left  = get_ue_golomb(&s->gb);
3848         sps->crop_right = get_ue_golomb(&s->gb);
3849         sps->crop_top   = get_ue_golomb(&s->gb);
3850         sps->crop_bottom= get_ue_golomb(&s->gb);
3851         if(sps->crop_left || sps->crop_top){
3852             av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completly supported, this could look slightly wrong ...\n");
3853         }
3854     }else{
3855         sps->crop_left  =
3856         sps->crop_right =
3857         sps->crop_top   =
3858         sps->crop_bottom= 0;
3859     }
3860
3861     sps->vui_parameters_present_flag= get_bits1(&s->gb);
3862     if( sps->vui_parameters_present_flag )
3863         decode_vui_parameters(h, sps);
3864
3865     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3866         av_log(h->s.avctx, AV_LOG_DEBUG, "sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s crop:%d/%d/%d/%d %s\n",
3867                sps_id, sps->profile_idc, sps->level_idc,
3868                sps->poc_type,
3869                sps->ref_frame_count,
3870                sps->mb_width, sps->mb_height,
3871                sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
3872                sps->direct_8x8_inference_flag ? "8B8" : "",
3873                sps->crop_left, sps->crop_right,
3874                sps->crop_top, sps->crop_bottom,
3875                sps->vui_parameters_present_flag ? "VUI" : ""
3876                );
3877     }
3878     return 0;
3879 }
3880
3881 static inline int decode_picture_parameter_set(H264Context *h){
3882     MpegEncContext * const s = &h->s;
3883     int pps_id= get_ue_golomb(&s->gb);
3884     PPS *pps= &h->pps_buffer[pps_id];
3885
3886     pps->sps_id= get_ue_golomb(&s->gb);
3887     pps->cabac= get_bits1(&s->gb);
3888     pps->pic_order_present= get_bits1(&s->gb);
3889     pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
3890     if(pps->slice_group_count > 1 ){
3891         pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
3892         av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
3893         switch(pps->mb_slice_group_map_type){
3894         case 0:
3895 #if 0
3896 |   for( i = 0; i <= num_slice_groups_minus1; i++ ) |   |        |
3897 |    run_length[ i ]                                |1  |ue(v)   |
3898 #endif
3899             break;
3900         case 2:
3901 #if 0
3902 |   for( i = 0; i < num_slice_groups_minus1; i++ )  |   |        |
3903 |{                                                  |   |        |
3904 |    top_left_mb[ i ]                               |1  |ue(v)   |
3905 |    bottom_right_mb[ i ]                           |1  |ue(v)   |
3906 |   }                                               |   |        |
3907 #endif
3908             break;
3909         case 3:
3910         case 4:
3911         case 5:
3912 #if 0
3913 |   slice_group_change_direction_flag               |1  |u(1)    |
3914 |   slice_group_change_rate_minus1                  |1  |ue(v)   |
3915 #endif
3916             break;
3917         case 6:
3918 #if 0
3919 |   slice_group_id_cnt_minus1                       |1  |ue(v)   |
3920 |   for( i = 0; i <= slice_group_id_cnt_minus1; i++ |   |        |
3921 |)                                                  |   |        |
3922 |    slice_group_id[ i ]                            |1  |u(v)    |
3923 #endif
3924             break;
3925         }
3926     }
3927     pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3928     pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3929     if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
3930         av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
3931         return -1;
3932     }
3933
3934     pps->weighted_pred= get_bits1(&s->gb);
3935     pps->weighted_bipred_idc= get_bits(&s->gb, 2);
3936     pps->init_qp= get_se_golomb(&s->gb) + 26;
3937     pps->init_qs= get_se_golomb(&s->gb) + 26;
3938     pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
3939     pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
3940     pps->constrained_intra_pred= get_bits1(&s->gb);
3941     pps->redundant_pic_cnt_present = get_bits1(&s->gb);
3942
3943     if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3944         av_log(h->s.avctx, AV_LOG_DEBUG, "pps:%d sps:%d %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s\n",
3945                pps_id, pps->sps_id,
3946                pps->cabac ? "CABAC" : "CAVLC",
3947                pps->slice_group_count,
3948                pps->ref_count[0], pps->ref_count[1],
3949                pps->weighted_pred ? "weighted" : "",
3950                pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
3951                pps->deblocking_filter_parameters_present ? "LPAR" : "",
3952                pps->constrained_intra_pred ? "CONSTR" : "",
3953                pps->redundant_pic_cnt_present ? "REDU" : ""
3954                );
3955     }
3956
3957     return 0;
3958 }
3959
3960 /**
3961  * finds the end of the current frame in the bitstream.
3962  * @return the position of the first byte of the next frame, or -1
3963  */
3964 static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){
3965     ParseContext *pc= &s->parse_context;
3966     int i;
3967     uint32_t state;
3968 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
3969 //    mb_addr= pc->mb_addr - 1;
3970     state= pc->state;
3971     //FIXME this will fail with slices
3972     for(i=0; i<buf_size; i++){
3973         state= (state<<8) | buf[i];
3974         if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
3975             if(pc->frame_start_found){
3976                 pc->state=-1;
3977                 pc->frame_start_found= 0;
3978                 return i-3;
3979             }
3980             pc->frame_start_found= 1;
3981         }
3982     }
3983
3984     pc->state= state;
3985     return END_NOT_FOUND;
3986 }
3987
3988 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
3989     MpegEncContext * const s = &h->s;
3990     AVCodecContext * const avctx= s->avctx;
3991     int buf_index=0;
3992 #if 0
3993     int i;
3994     for(i=0; i<32; i++){
3995         printf("%X ", buf[i]);
3996     }
3997 #endif
3998     for(;;){
3999         int consumed;
4000         int dst_length;
4001         int bit_length;
4002         uint8_t *ptr;
4003
4004         // start code prefix search
4005         for(; buf_index + 3 < buf_size; buf_index++){
4006             // this should allways succeed in the first iteration
4007             if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
4008                 break;
4009         }
4010
4011         if(buf_index+3 >= buf_size) break;
4012
4013         buf_index+=3;
4014
4015         ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index);
4016         if(ptr[dst_length - 1] == 0) dst_length--;
4017         bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
4018
4019         if(s->avctx->debug&FF_DEBUG_STARTCODE){
4020             av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
4021         }
4022
4023         buf_index += consumed;
4024
4025         if(h->nal_ref_idc < s->hurry_up)
4026             continue;
4027
4028         switch(h->nal_unit_type){
4029         case NAL_IDR_SLICE:
4030             idr(h); //FIXME ensure we dont loose some frames if there is reordering
4031         case NAL_SLICE:
4032             init_get_bits(&s->gb, ptr, bit_length);
4033             h->intra_gb_ptr=
4034             h->inter_gb_ptr= &s->gb;
4035             s->data_partitioning = 0;
4036
4037             if(decode_slice_header(h) < 0) return -1;
4038             if(h->redundant_pic_count==0)
4039                 decode_slice(h);
4040             break;
4041         case NAL_DPA:
4042             init_get_bits(&s->gb, ptr, bit_length);
4043             h->intra_gb_ptr=
4044             h->inter_gb_ptr= NULL;
4045             s->data_partitioning = 1;
4046
4047             if(decode_slice_header(h) < 0) return -1;
4048             break;
4049         case NAL_DPB:
4050             init_get_bits(&h->intra_gb, ptr, bit_length);
4051             h->intra_gb_ptr= &h->intra_gb;
4052             break;
4053         case NAL_DPC:
4054             init_get_bits(&h->inter_gb, ptr, bit_length);
4055             h->inter_gb_ptr= &h->inter_gb;
4056
4057             if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning)
4058                 decode_slice(h);
4059             break;
4060         case NAL_SEI:
4061             break;
4062         case NAL_SPS:
4063             init_get_bits(&s->gb, ptr, bit_length);
4064             decode_seq_parameter_set(h);
4065
4066             if(s->flags& CODEC_FLAG_LOW_DELAY)
4067                 s->low_delay=1;
4068
4069             avctx->has_b_frames= !s->low_delay;
4070             break;
4071         case NAL_PPS:
4072             init_get_bits(&s->gb, ptr, bit_length);
4073
4074             decode_picture_parameter_set(h);
4075
4076             break;
4077         case NAL_PICTURE_DELIMITER:
4078             break;
4079         case NAL_FILTER_DATA:
4080             break;
4081         }
4082
4083         //FIXME move after where irt is set
4084         s->current_picture.pict_type= s->pict_type;
4085         s->current_picture.key_frame= s->pict_type == I_TYPE;
4086     }
4087
4088     if(!s->current_picture_ptr) return buf_index; //no frame
4089
4090     h->prev_frame_num_offset= h->frame_num_offset;
4091     h->prev_frame_num= h->frame_num;
4092     if(s->current_picture_ptr->reference){
4093         h->prev_poc_msb= h->poc_msb;
4094         h->prev_poc_lsb= h->poc_lsb;
4095     }
4096     if(s->current_picture_ptr->reference)
4097         execute_ref_pic_marking(h, h->mmco, h->mmco_index);
4098     else
4099         assert(h->mmco_index==0);
4100
4101     ff_er_frame_end(s);
4102     MPV_frame_end(s);
4103
4104     return buf_index;
4105 }
4106
4107 /**
4108  * retunrs the number of bytes consumed for building the current frame
4109  */
4110 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
4111     if(s->flags&CODEC_FLAG_TRUNCATED){
4112         pos -= s->parse_context.last_index;
4113         if(pos<0) pos=0; // FIXME remove (uneeded?)
4114
4115         return pos;
4116     }else{
4117         if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
4118         if(pos+10>buf_size) pos=buf_size; // oops ;)
4119
4120         return pos;
4121     }
4122 }
4123
4124 static int decode_frame(AVCodecContext *avctx,
4125                              void *data, int *data_size,
4126                              uint8_t *buf, int buf_size)
4127 {
4128     H264Context *h = avctx->priv_data;
4129     MpegEncContext *s = &h->s;
4130     AVFrame *pict = data;
4131     int buf_index;
4132
4133     s->flags= avctx->flags;
4134     s->flags2= avctx->flags2;
4135
4136     *data_size = 0;
4137
4138    /* no supplementary picture */
4139     if (buf_size == 0) {
4140         return 0;
4141     }
4142
4143     if(s->flags&CODEC_FLAG_TRUNCATED){
4144         int next= find_frame_end(s, buf, buf_size);
4145
4146         if( ff_combine_frame(s, next, &buf, &buf_size) < 0 )
4147             return buf_size;
4148 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
4149     }
4150
4151     if(s->avctx->extradata_size && s->picture_number==0){
4152         if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
4153             return -1;
4154     }
4155
4156     buf_index=decode_nal_units(h, buf, buf_size);
4157     if(buf_index < 0)
4158         return -1;
4159
4160     //FIXME do something with unavailable reference frames
4161
4162 //    if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
4163 #if 0
4164     if(s->pict_type==B_TYPE || s->low_delay){
4165         *pict= *(AVFrame*)&s->current_picture;
4166     } else {
4167         *pict= *(AVFrame*)&s->last_picture;
4168     }
4169 #endif
4170     if(!s->current_picture_ptr){
4171         av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");
4172         return -1;
4173     }
4174
4175     *pict= *(AVFrame*)&s->current_picture; //FIXME
4176     ff_print_debug_info(s, pict);
4177     assert(pict->data[0]);
4178 //printf("out %d\n", (int)pict->data[0]);
4179 #if 0 //?
4180
4181     /* Return the Picture timestamp as the frame number */
4182     /* we substract 1 because it is added on utils.c    */
4183     avctx->frame_number = s->picture_number - 1;
4184 #endif
4185 #if 0
4186     /* dont output the last pic after seeking */
4187     if(s->last_picture_ptr || s->low_delay)
4188     //Note this isnt a issue as a IDR pic should flush teh buffers
4189 #endif
4190         *data_size = sizeof(AVFrame);
4191     return get_consumed_bytes(s, buf_index, buf_size);
4192 }
4193 #if 0
4194 static inline void fill_mb_avail(H264Context *h){
4195     MpegEncContext * const s = &h->s;
4196     const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4197
4198     if(s->mb_y){
4199         h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
4200         h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
4201         h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
4202     }else{
4203         h->mb_avail[0]=
4204         h->mb_avail[1]=
4205         h->mb_avail[2]= 0;
4206     }
4207     h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
4208     h->mb_avail[4]= 1; //FIXME move out
4209     h->mb_avail[5]= 0; //FIXME move out
4210 }
4211 #endif
4212
4213 #if 0 //selftest
4214 #define COUNT 8000
4215 #define SIZE (COUNT*40)
4216 int main(){
4217     int i;
4218     uint8_t temp[SIZE];
4219     PutBitContext pb;
4220     GetBitContext gb;
4221 //    int int_temp[10000];
4222     DSPContext dsp;
4223     AVCodecContext avctx;
4224
4225     dsputil_init(&dsp, &avctx);
4226
4227     init_put_bits(&pb, temp, SIZE);
4228     printf("testing unsigned exp golomb\n");
4229     for(i=0; i<COUNT; i++){
4230         START_TIMER
4231         set_ue_golomb(&pb, i);
4232         STOP_TIMER("set_ue_golomb");
4233     }
4234     flush_put_bits(&pb);
4235
4236     init_get_bits(&gb, temp, 8*SIZE);
4237     for(i=0; i<COUNT; i++){
4238         int j, s;
4239
4240         s= show_bits(&gb, 24);
4241
4242         START_TIMER
4243         j= get_ue_golomb(&gb);
4244         if(j != i){
4245             printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4246 //            return -1;
4247         }
4248         STOP_TIMER("get_ue_golomb");
4249     }
4250
4251
4252     init_put_bits(&pb, temp, SIZE);
4253     printf("testing signed exp golomb\n");
4254     for(i=0; i<COUNT; i++){
4255         START_TIMER
4256         set_se_golomb(&pb, i - COUNT/2);
4257         STOP_TIMER("set_se_golomb");
4258     }
4259     flush_put_bits(&pb);
4260
4261     init_get_bits(&gb, temp, 8*SIZE);
4262     for(i=0; i<COUNT; i++){
4263         int j, s;
4264
4265         s= show_bits(&gb, 24);
4266
4267         START_TIMER
4268         j= get_se_golomb(&gb);
4269         if(j != i - COUNT/2){
4270             printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4271 //            return -1;
4272         }
4273         STOP_TIMER("get_se_golomb");
4274     }
4275
4276     printf("testing 4x4 (I)DCT\n");
4277
4278     DCTELEM block[16];
4279     uint8_t src[16], ref[16];
4280     uint64_t error= 0, max_error=0;
4281
4282     for(i=0; i<COUNT; i++){
4283         int j;
4284 //        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
4285         for(j=0; j<16; j++){
4286             ref[j]= random()%255;
4287             src[j]= random()%255;
4288         }
4289
4290         h264_diff_dct_c(block, src, ref, 4);
4291
4292         //normalize
4293         for(j=0; j<16; j++){
4294 //            printf("%d ", block[j]);
4295             block[j]= block[j]*4;
4296             if(j&1) block[j]= (block[j]*4 + 2)/5;
4297             if(j&4) block[j]= (block[j]*4 + 2)/5;
4298         }
4299 //        printf("\n");
4300
4301         h264_add_idct_c(ref, block, 4);
4302 /*        for(j=0; j<16; j++){
4303             printf("%d ", ref[j]);
4304         }
4305         printf("\n");*/
4306
4307         for(j=0; j<16; j++){
4308             int diff= ABS(src[j] - ref[j]);
4309
4310             error+= diff*diff;
4311             max_error= FFMAX(max_error, diff);
4312         }
4313     }
4314     printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
4315 #if 0
4316     printf("testing quantizer\n");
4317     for(qp=0; qp<52; qp++){
4318         for(i=0; i<16; i++)
4319             src1_block[i]= src2_block[i]= random()%255;
4320
4321     }
4322 #endif
4323     printf("Testing NAL layer\n");
4324
4325     uint8_t bitstream[COUNT];
4326     uint8_t nal[COUNT*2];
4327     H264Context h;
4328     memset(&h, 0, sizeof(H264Context));
4329
4330     for(i=0; i<COUNT; i++){
4331         int zeros= i;
4332         int nal_length;
4333         int consumed;
4334         int out_length;
4335         uint8_t *out;
4336         int j;
4337
4338         for(j=0; j<COUNT; j++){
4339             bitstream[j]= (random() % 255) + 1;
4340         }
4341
4342         for(j=0; j<zeros; j++){
4343             int pos= random() % COUNT;
4344             while(bitstream[pos] == 0){
4345                 pos++;
4346                 pos %= COUNT;
4347             }
4348             bitstream[pos]=0;
4349         }
4350
4351         START_TIMER
4352
4353         nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
4354         if(nal_length<0){
4355             printf("encoding failed\n");
4356             return -1;
4357         }
4358
4359         out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
4360
4361         STOP_TIMER("NAL")
4362
4363         if(out_length != COUNT){
4364             printf("incorrect length %d %d\n", out_length, COUNT);
4365             return -1;
4366         }
4367
4368         if(consumed != nal_length){
4369             printf("incorrect consumed length %d %d\n", nal_length, consumed);
4370             return -1;
4371         }
4372
4373         if(memcmp(bitstream, out, COUNT)){
4374             printf("missmatch\n");
4375             return -1;
4376         }
4377     }
4378
4379     printf("Testing RBSP\n");
4380
4381
4382     return 0;
4383 }
4384 #endif
4385
4386
4387 static int decode_end(AVCodecContext *avctx)
4388 {
4389     H264Context *h = avctx->priv_data;
4390     MpegEncContext *s = &h->s;
4391
4392     free_tables(h); //FIXME cleanup init stuff perhaps
4393     MPV_common_end(s);
4394
4395 //    memset(h, 0, sizeof(H264Context));
4396
4397     return 0;
4398 }
4399
4400
4401 AVCodec h264_decoder = {
4402     "h264",
4403     CODEC_TYPE_VIDEO,
4404     CODEC_ID_H264,
4405     sizeof(H264Context),
4406     decode_init,
4407     NULL,
4408     decode_end,
4409     decode_frame,
4410     /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,
4411 };
4412
4413 #include "svq3.c"