Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 7d7f635d

History | View | Annotate | Download (209 KB)

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
#include "cabac.h"
35

    
36
#undef NDEBUG
37
#include <assert.h>
38

    
39
#define interlaced_dct interlaced_dct_is_a_bad_name
40
#define mb_intra mb_intra_isnt_initalized_see_mb_type
41

    
42
#define LUMA_DC_BLOCK_INDEX   25
43
#define CHROMA_DC_BLOCK_INDEX 26
44

    
45
#define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
46
#define COEFF_TOKEN_VLC_BITS           8
47
#define TOTAL_ZEROS_VLC_BITS           9
48
#define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
49
#define RUN_VLC_BITS                   3
50
#define RUN7_VLC_BITS                  6
51

    
52
#define MAX_SPS_COUNT 32
53
#define MAX_PPS_COUNT 256
54

    
55
#define MAX_MMCO_COUNT 66
56

    
57
/**
58
 * Sequence parameter set
59
 */
60
typedef struct SPS{
61
    
62
    int profile_idc;
63
    int level_idc;
64
    int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
65
    int poc_type;                      ///< pic_order_cnt_type
66
    int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
67
    int delta_pic_order_always_zero_flag;
68
    int offset_for_non_ref_pic;
69
    int offset_for_top_to_bottom_field;
70
    int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
71
    int ref_frame_count;               ///< num_ref_frames
72
    int gaps_in_frame_num_allowed_flag;
73
    int mb_width;                      ///< frame_width_in_mbs_minus1 + 1
74
    int mb_height;                     ///< frame_height_in_mbs_minus1 + 1
75
    int frame_mbs_only_flag;
76
    int mb_aff;                        ///<mb_adaptive_frame_field_flag
77
    int direct_8x8_inference_flag;
78
    int crop;                   ///< frame_cropping_flag
79
    int crop_left;              ///< frame_cropping_rect_left_offset
80
    int crop_right;             ///< frame_cropping_rect_right_offset
81
    int crop_top;               ///< frame_cropping_rect_top_offset
82
    int crop_bottom;            ///< frame_cropping_rect_bottom_offset
83
    int vui_parameters_present_flag;
84
    AVRational sar;
85
    int timing_info_present_flag;
86
    uint32_t num_units_in_tick;
87
    uint32_t time_scale;
88
    int fixed_frame_rate_flag;
89
    short offset_for_ref_frame[256]; //FIXME dyn aloc?
90
}SPS;
91

    
92
/**
93
 * Picture parameter set
94
 */
95
typedef struct PPS{
96
    int sps_id;
97
    int cabac;                  ///< entropy_coding_mode_flag
98
    int pic_order_present;      ///< pic_order_present_flag
99
    int slice_group_count;      ///< num_slice_groups_minus1 + 1
100
    int mb_slice_group_map_type;
101
    int ref_count[2];           ///< num_ref_idx_l0/1_active_minus1 + 1
102
    int weighted_pred;          ///< weighted_pred_flag
103
    int weighted_bipred_idc;
104
    int init_qp;                ///< pic_init_qp_minus26 + 26
105
    int init_qs;                ///< pic_init_qs_minus26 + 26
106
    int chroma_qp_index_offset;
107
    int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
108
    int constrained_intra_pred; ///< constrained_intra_pred_flag
109
    int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
110
}PPS;
111

    
112
/**
113
 * Memory management control operation opcode.
114
 */
115
typedef enum MMCOOpcode{
116
    MMCO_END=0,
117
    MMCO_SHORT2UNUSED,
118
    MMCO_LONG2UNUSED,
119
    MMCO_SHORT2LONG,
120
    MMCO_SET_MAX_LONG,
121
    MMCO_RESET, 
122
    MMCO_LONG,
123
} MMCOOpcode;
124

    
125
/**
126
 * Memory management control operation.
127
 */
128
typedef struct MMCO{
129
    MMCOOpcode opcode;
130
    int short_frame_num;
131
    int long_index;
132
} MMCO;
133

    
134
/**
135
 * H264Context
136
 */
137
typedef struct H264Context{
138
    MpegEncContext s;
139
    int nal_ref_idc;        
140
    int nal_unit_type;
141
#define NAL_SLICE                1
142
#define NAL_DPA                        2
143
#define NAL_DPB                        3
144
#define NAL_DPC                        4
145
#define NAL_IDR_SLICE                5
146
#define NAL_SEI                        6
147
#define NAL_SPS                        7
148
#define NAL_PPS                        8
149
#define NAL_PICTURE_DELIMITER        9
150
#define NAL_FILTER_DATA                10
151
    uint8_t *rbsp_buffer;
152
    int rbsp_buffer_size;
153

    
154
    /**
155
      * Used to parse AVC variant of h264
156
      */
157
    int is_avc; ///< this flag is != 0 if codec is avc1
158
    int got_avcC; ///< flag used to parse avcC data only once
159
    int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
160

    
161
    int chroma_qp; //QPc
162

    
163
    int prev_mb_skiped; //FIXME remove (IMHO not used)
164

    
165
    //prediction stuff
166
    int chroma_pred_mode;
167
    int intra16x16_pred_mode;
168
    
169
    int8_t intra4x4_pred_mode_cache[5*8];
170
    int8_t (*intra4x4_pred_mode)[8];
171
    void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
172
    void (*pred8x8  [4+3])(uint8_t *src, int stride);
173
    void (*pred16x16[4+3])(uint8_t *src, int stride);
174
    unsigned int topleft_samples_available;
175
    unsigned int top_samples_available;
176
    unsigned int topright_samples_available;
177
    unsigned int left_samples_available;
178
    uint8_t (*top_border)[16+2*8];
179
    uint8_t left_border[17+2*9];
180

    
181
    /**
182
     * non zero coeff count cache.
183
     * is 64 if not available.
184
     */
185
    uint8_t non_zero_count_cache[6*8];
186
    uint8_t (*non_zero_count)[16];
187

    
188
    /**
189
     * Motion vector cache.
190
     */
191
    int16_t mv_cache[2][5*8][2];
192
    int8_t ref_cache[2][5*8];
193
#define LIST_NOT_USED -1 //FIXME rename?
194
#define PART_NOT_AVAILABLE -2
195
    
196
    /**
197
     * is 1 if the specific list MV&references are set to 0,0,-2.
198
     */
199
    int mv_cache_clean[2];
200

    
201
    int block_offset[16+8];
202
    int chroma_subblock_offset[16]; //FIXME remove
203
    
204
    uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
205
    uint16_t *mb2b8_xy;
206
    int b_stride;
207
    int b8_stride;
208

    
209
    int halfpel_flag;
210
    int thirdpel_flag;
211

    
212
    int unknown_svq3_flag;
213
    int next_slice_index;
214

    
215
    SPS sps_buffer[MAX_SPS_COUNT];
216
    SPS sps; ///< current sps
217
    
218
    PPS pps_buffer[MAX_PPS_COUNT];
219
    /**
220
     * current pps
221
     */
222
    PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
223

    
224
    int slice_num;
225
    uint8_t *slice_table_base;
226
    uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
227
    int slice_type;
228
    int slice_type_fixed;
229
    
230
    //interlacing specific flags
231
    int mb_field_decoding_flag;
232
    
233
    int sub_mb_type[4];
234
    
235
    //POC stuff
236
    int poc_lsb;
237
    int poc_msb;
238
    int delta_poc_bottom;
239
    int delta_poc[2];
240
    int frame_num;
241
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
242
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
243
    int frame_num_offset;         ///< for POC type 2
244
    int prev_frame_num_offset;    ///< for POC type 2
245
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
246

    
247
    /**
248
     * frame_num for frames or 2*frame_num for field pics.
249
     */
250
    int curr_pic_num;
251
    
252
    /**
253
     * max_frame_num or 2*max_frame_num for field pics.
254
     */
255
    int max_pic_num;
256

    
257
    //Weighted pred stuff
258
    int luma_log2_weight_denom;
259
    int chroma_log2_weight_denom;
260
    int luma_weight[2][16];
261
    int luma_offset[2][16];
262
    int chroma_weight[2][16][2];
263
    int chroma_offset[2][16][2];
264
   
265
    //deblock
266
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0 
267
    int slice_alpha_c0_offset;
268
    int slice_beta_offset;
269
     
270
    int redundant_pic_count;
271
    
272
    int direct_spatial_mv_pred;
273

    
274
    /**
275
     * num_ref_idx_l0/1_active_minus1 + 1
276
     */
277
    int ref_count[2];// FIXME split for AFF
278
    Picture *short_ref[16];
279
    Picture *long_ref[16];
280
    Picture default_ref_list[2][32];
281
    Picture ref_list[2][32]; //FIXME size?
282
    Picture field_ref_list[2][32]; //FIXME size?
283
    
284
    /**
285
     * memory management control operations buffer.
286
     */
287
    MMCO mmco[MAX_MMCO_COUNT];
288
    int mmco_index;
289
    
290
    int long_ref_count;  ///< number of actual long term references
291
    int short_ref_count; ///< number of actual short term references
292
    
293
    //data partitioning
294
    GetBitContext intra_gb;
295
    GetBitContext inter_gb;
296
    GetBitContext *intra_gb_ptr;
297
    GetBitContext *inter_gb_ptr;
298
    
299
    DCTELEM mb[16*24] __align8;
300

    
301
    /**
302
     * Cabac
303
     */
304
    CABACContext cabac;
305
    uint8_t      cabac_state[399];
306
    int          cabac_init_idc;
307

    
308
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
309
    uint16_t     *cbp_table;
310
    int top_cbp;
311
    int left_cbp;
312
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
313
    uint8_t     *chroma_pred_mode_table;
314
    int         last_qscale_diff;
315
    int16_t     (*mvd_table[2])[2];
316
    int16_t     mvd_cache[2][5*8][2];
317

    
318
}H264Context;
319

    
320
static VLC coeff_token_vlc[4];
321
static VLC chroma_dc_coeff_token_vlc;
322

    
323
static VLC total_zeros_vlc[15];
324
static VLC chroma_dc_total_zeros_vlc[3];
325

    
326
static VLC run_vlc[6];
327
static VLC run7_vlc;
328

    
329
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
330
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
331
static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
332

    
333
static inline uint32_t pack16to32(int a, int b){
334
#ifdef WORDS_BIGENDIAN
335
   return (b&0xFFFF) + (a<<16);
336
#else
337
   return (a&0xFFFF) + (b<<16);
338
#endif
339
}
340

    
341
/**
342
 * fill a rectangle.
343
 * @param h height of the recatangle, should be a constant
344
 * @param w width of the recatangle, should be a constant
345
 * @param size the size of val (1 or 4), should be a constant
346
 */
347
static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
348
    uint8_t *p= (uint8_t*)vp;
349
    assert(size==1 || size==4);
350
    
351
    w      *= size;
352
    stride *= size;
353
    
354
//FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
355
    if(w==2 && h==2){
356
        *(uint16_t*)(p + 0)=
357
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
358
    }else if(w==2 && h==4){
359
        *(uint16_t*)(p + 0*stride)=
360
        *(uint16_t*)(p + 1*stride)=
361
        *(uint16_t*)(p + 2*stride)=
362
        *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
363
    }else if(w==4 && h==1){
364
        *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
365
    }else if(w==4 && h==2){
366
        *(uint32_t*)(p + 0*stride)=
367
        *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
368
    }else if(w==4 && h==4){
369
        *(uint32_t*)(p + 0*stride)=
370
        *(uint32_t*)(p + 1*stride)=
371
        *(uint32_t*)(p + 2*stride)=
372
        *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
373
    }else if(w==8 && h==1){
374
        *(uint32_t*)(p + 0)=
375
        *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
376
    }else if(w==8 && h==2){
377
        *(uint32_t*)(p + 0 + 0*stride)=
378
        *(uint32_t*)(p + 4 + 0*stride)=
379
        *(uint32_t*)(p + 0 + 1*stride)=
380
        *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
381
    }else if(w==8 && h==4){
382
        *(uint64_t*)(p + 0*stride)=
383
        *(uint64_t*)(p + 1*stride)=
384
        *(uint64_t*)(p + 2*stride)=
385
        *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
386
    }else if(w==16 && h==2){
387
        *(uint64_t*)(p + 0+0*stride)=
388
        *(uint64_t*)(p + 8+0*stride)=
389
        *(uint64_t*)(p + 0+1*stride)=
390
        *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
391
    }else if(w==16 && h==4){
392
        *(uint64_t*)(p + 0+0*stride)=
393
        *(uint64_t*)(p + 8+0*stride)=
394
        *(uint64_t*)(p + 0+1*stride)=
395
        *(uint64_t*)(p + 8+1*stride)=
396
        *(uint64_t*)(p + 0+2*stride)=
397
        *(uint64_t*)(p + 8+2*stride)=
398
        *(uint64_t*)(p + 0+3*stride)=
399
        *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
400
    }else
401
        assert(0);
402
}
403

    
404
static inline void fill_caches(H264Context *h, int mb_type){
405
    MpegEncContext * const s = &h->s;
406
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
407
    int topleft_xy, top_xy, topright_xy, left_xy[2];
408
    int topleft_type, top_type, topright_type, left_type[2];
409
    int left_block[4];
410
    int i;
411

    
412
    //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it 
413
    
414
    if(h->sps.mb_aff){
415
    //FIXME
416
        topleft_xy = 0; /* avoid warning */
417
        top_xy = 0; /* avoid warning */
418
        topright_xy = 0; /* avoid warning */
419
    }else{
420
        topleft_xy = mb_xy-1 - s->mb_stride;
421
        top_xy     = mb_xy   - s->mb_stride;
422
        topright_xy= mb_xy+1 - s->mb_stride;
423
        left_xy[0]   = mb_xy-1;
424
        left_xy[1]   = mb_xy-1;
425
        left_block[0]= 0;
426
        left_block[1]= 1;
427
        left_block[2]= 2;
428
        left_block[3]= 3;
429
    }
430

    
431
    topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
432
    top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
433
    topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
434
    left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
435
    left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
436

    
437
    if(IS_INTRA(mb_type)){
438
        h->topleft_samples_available= 
439
        h->top_samples_available= 
440
        h->left_samples_available= 0xFFFF;
441
        h->topright_samples_available= 0xEEEA;
442

    
443
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
444
            h->topleft_samples_available= 0xB3FF;
445
            h->top_samples_available= 0x33FF;
446
            h->topright_samples_available= 0x26EA;
447
        }
448
        for(i=0; i<2; i++){
449
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
450
                h->topleft_samples_available&= 0xDF5F;
451
                h->left_samples_available&= 0x5F5F;
452
            }
453
        }
454
        
455
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
456
            h->topleft_samples_available&= 0x7FFF;
457
        
458
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
459
            h->topright_samples_available&= 0xFBFF;
460
    
461
        if(IS_INTRA4x4(mb_type)){
462
            if(IS_INTRA4x4(top_type)){
463
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
464
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
465
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
466
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
467
            }else{
468
                int pred;
469
                if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
470
                    pred= 2;
471
                else{
472
                    pred= -1;
473
                }
474
                h->intra4x4_pred_mode_cache[4+8*0]=
475
                h->intra4x4_pred_mode_cache[5+8*0]=
476
                h->intra4x4_pred_mode_cache[6+8*0]=
477
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
478
            }
479
            for(i=0; i<2; i++){
480
                if(IS_INTRA4x4(left_type[i])){
481
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
482
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
483
                }else{
484
                    int pred;
485
                    if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
486
                        pred= 2;
487
                    else{
488
                        pred= -1;
489
                    }
490
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
491
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
492
                }
493
            }
494
        }
495
    }
496
    
497
    
498
/*
499
0 . T T. T T T T 
500
1 L . .L . . . . 
501
2 L . .L . . . . 
502
3 . T TL . . . . 
503
4 L . .L . . . . 
504
5 L . .. . . . . 
505
*/
506
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
507
    if(top_type){
508
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
509
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
510
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
511
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
512
    
513
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
514
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
515
    
516
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
517
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
518
        
519
        h->top_cbp= h->cbp_table[top_xy];
520
    }else{
521
        h->non_zero_count_cache[4+8*0]=      
522
        h->non_zero_count_cache[5+8*0]=
523
        h->non_zero_count_cache[6+8*0]=
524
        h->non_zero_count_cache[7+8*0]=
525
    
526
        h->non_zero_count_cache[1+8*0]=
527
        h->non_zero_count_cache[2+8*0]=
528
    
529
        h->non_zero_count_cache[1+8*3]=
530
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
531
        
532
        if(IS_INTRA(mb_type)) h->top_cbp= 0x1C0;
533
        else                  h->top_cbp= 0;
534
    }
535
    
536
    if(left_type[0]){
537
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
538
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
539
        h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
540
        h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
541
        h->left_cbp= h->cbp_table[left_xy[0]]; //FIXME interlacing
542
    }else{
543
        h->non_zero_count_cache[3+8*1]= 
544
        h->non_zero_count_cache[3+8*2]= 
545
        h->non_zero_count_cache[0+8*1]= 
546
        h->non_zero_count_cache[0+8*4]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
547
        
548
        if(IS_INTRA(mb_type)) h->left_cbp= 0x1C0;//FIXME interlacing
549
        else                  h->left_cbp= 0;
550
    }
551
    
552
    if(left_type[1]){
553
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
554
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
555
        h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
556
        h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
557
    }else{
558
        h->non_zero_count_cache[3+8*3]= 
559
        h->non_zero_count_cache[3+8*4]= 
560
        h->non_zero_count_cache[0+8*2]= 
561
        h->non_zero_count_cache[0+8*5]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
562
    }
563
    
564
#if 1
565
    if(IS_INTER(mb_type)){
566
        int list;
567
        for(list=0; list<2; list++){
568
            if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
569
                /*if(!h->mv_cache_clean[list]){
570
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
571
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
572
                    h->mv_cache_clean[list]= 1;
573
                }*/
574
                continue; //FIXME direct mode ...
575
            }
576
            h->mv_cache_clean[list]= 0;
577
            
578
            if(IS_INTER(topleft_type)){
579
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
580
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
581
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
582
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
583
            }else{
584
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
585
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
586
            }
587
            
588
            if(IS_INTER(top_type)){
589
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
590
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
591
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
592
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
593
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
594
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
595
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
596
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
597
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
598
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
599
            }else{
600
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
601
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
602
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
603
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
604
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
605
            }
606

    
607
            if(IS_INTER(topright_type)){
608
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
609
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
610
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
611
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
612
            }else{
613
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
614
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
615
            }
616
            
617
            //FIXME unify cleanup or sth
618
            if(IS_INTER(left_type[0])){
619
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
620
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
621
                *(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]];
622
                *(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]];
623
                h->ref_cache[list][scan8[0] - 1 + 0*8]= 
624
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
625
            }else{
626
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
627
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
628
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
629
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
630
            }
631
            
632
            if(IS_INTER(left_type[1])){
633
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
634
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
635
                *(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]];
636
                *(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]];
637
                h->ref_cache[list][scan8[0] - 1 + 2*8]= 
638
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
639
            }else{
640
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
641
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
642
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
643
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
644
            }
645

    
646
            h->ref_cache[list][scan8[5 ]+1] = 
647
            h->ref_cache[list][scan8[7 ]+1] = 
648
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewher else)
649
            h->ref_cache[list][scan8[4 ]] = 
650
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
651
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
652
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
653
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
654
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
655
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
656

    
657
            if( h->pps.cabac ) {
658
                /* XXX beurk, Load mvd */
659
                if(IS_INTER(topleft_type)){
660
                    const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
661
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
662
                }else{
663
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
664
                }
665

    
666
                if(IS_INTER(top_type)){
667
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
668
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
669
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
670
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
671
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
672
                }else{
673
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]= 
674
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]= 
675
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]= 
676
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
677
                }
678
                if(IS_INTER(left_type[0])){
679
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
680
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
681
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
682
                }else{
683
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
684
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
685
                }
686
                if(IS_INTER(left_type[1])){
687
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
688
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
689
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
690
                }else{
691
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
692
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
693
                }
694
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
695
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
696
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
697
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
698
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
699
            }
700
        }
701
//FIXME
702
    }
703
#endif
704
}
705

    
706
static inline void write_back_intra_pred_mode(H264Context *h){
707
    MpegEncContext * const s = &h->s;
708
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
709

    
710
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
711
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
712
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
713
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
714
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
715
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
716
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
717
}
718

    
719
/**
720
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
721
 */
722
static inline int check_intra4x4_pred_mode(H264Context *h){
723
    MpegEncContext * const s = &h->s;
724
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
725
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
726
    int i;
727
    
728
    if(!(h->top_samples_available&0x8000)){
729
        for(i=0; i<4; i++){
730
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
731
            if(status<0){
732
                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);
733
                return -1;
734
            } else if(status){
735
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
736
            }
737
        }
738
    }
739
    
740
    if(!(h->left_samples_available&0x8000)){
741
        for(i=0; i<4; i++){
742
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
743
            if(status<0){
744
                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);
745
                return -1;
746
            } else if(status){
747
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
748
            }
749
        }
750
    }
751

    
752
    return 0;
753
} //FIXME cleanup like next
754

    
755
/**
756
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
757
 */
758
static inline int check_intra_pred_mode(H264Context *h, int mode){
759
    MpegEncContext * const s = &h->s;
760
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
761
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
762
    
763
    if(mode < 0 || mode > 6)
764
        return -1;
765
    
766
    if(!(h->top_samples_available&0x8000)){
767
        mode= top[ mode ];
768
        if(mode<0){
769
            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);
770
            return -1;
771
        }
772
    }
773
    
774
    if(!(h->left_samples_available&0x8000)){
775
        mode= left[ mode ];
776
        if(mode<0){
777
            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);
778
            return -1;
779
        } 
780
    }
781

    
782
    return mode;
783
}
784

    
785
/**
786
 * gets the predicted intra4x4 prediction mode.
787
 */
788
static inline int pred_intra_mode(H264Context *h, int n){
789
    const int index8= scan8[n];
790
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
791
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
792
    const int min= FFMIN(left, top);
793

    
794
    tprintf("mode:%d %d min:%d\n", left ,top, min);
795

    
796
    if(min<0) return DC_PRED;
797
    else      return min;
798
}
799

    
800
static inline void write_back_non_zero_count(H264Context *h){
801
    MpegEncContext * const s = &h->s;
802
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
803

    
804
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
805
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
806
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
807
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
808
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
809
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
810
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
811
    
812
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
813
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
814
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
815

    
816
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
817
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
818
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
819
}
820

    
821
/**
822
 * gets the predicted number of non zero coefficients.
823
 * @param n block index
824
 */
825
static inline int pred_non_zero_count(H264Context *h, int n){
826
    const int index8= scan8[n];
827
    const int left= h->non_zero_count_cache[index8 - 1];
828
    const int top = h->non_zero_count_cache[index8 - 8];
829
    int i= left + top;
830
    
831
    if(i<64) i= (i+1)>>1;
832

    
833
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
834

    
835
    return i&31;
836
}
837

    
838
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
839
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
840

    
841
    if(topright_ref != PART_NOT_AVAILABLE){
842
        *C= h->mv_cache[list][ i - 8 + part_width ];
843
        return topright_ref;
844
    }else{
845
        tprintf("topright MV not available\n");
846

    
847
        *C= h->mv_cache[list][ i - 8 - 1 ];
848
        return h->ref_cache[list][ i - 8 - 1 ];
849
    }
850
}
851

    
852
/**
853
 * gets the predicted MV.
854
 * @param n the block index
855
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
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_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
860
    const int index8= scan8[n];
861
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
862
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
863
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
864
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
865
    const int16_t * C;
866
    int diagonal_ref, match_count;
867

    
868
    assert(part_width==1 || part_width==2 || part_width==4);
869

    
870
/* mv_cache
871
  B . . A T T T T 
872
  U . . L . . , .
873
  U . . L . . . .
874
  U . . L . . , .
875
  . . . L . . . .
876
*/
877

    
878
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
879
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
880
    if(match_count > 1){ //most common
881
        *mx= mid_pred(A[0], B[0], C[0]);
882
        *my= mid_pred(A[1], B[1], C[1]);
883
    }else if(match_count==1){
884
        if(left_ref==ref){
885
            *mx= A[0];
886
            *my= A[1];        
887
        }else if(top_ref==ref){
888
            *mx= B[0];
889
            *my= B[1];        
890
        }else{
891
            *mx= C[0];
892
            *my= C[1];        
893
        }
894
    }else{
895
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
896
            *mx= A[0];
897
            *my= A[1];        
898
        }else{
899
            *mx= mid_pred(A[0], B[0], C[0]);
900
            *my= mid_pred(A[1], B[1], C[1]);
901
        }
902
    }
903
        
904
    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);
905
}
906

    
907
/**
908
 * gets the directionally predicted 16x8 MV.
909
 * @param n the block index
910
 * @param mx the x component of the predicted motion vector
911
 * @param my the y component of the predicted motion vector
912
 */
913
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
914
    if(n==0){
915
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
916
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
917

    
918
        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);
919
        
920
        if(top_ref == ref){
921
            *mx= B[0];
922
            *my= B[1];
923
            return;
924
        }
925
    }else{
926
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
927
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
928
        
929
        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);
930

    
931
        if(left_ref == ref){
932
            *mx= A[0];
933
            *my= A[1];
934
            return;
935
        }
936
    }
937

    
938
    //RARE
939
    pred_motion(h, n, 4, list, ref, mx, my);
940
}
941

    
942
/**
943
 * gets the directionally predicted 8x16 MV.
944
 * @param n the block index
945
 * @param mx the x component of the predicted motion vector
946
 * @param my the y component of the predicted motion vector
947
 */
948
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
949
    if(n==0){
950
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
951
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
952
        
953
        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);
954

    
955
        if(left_ref == ref){
956
            *mx= A[0];
957
            *my= A[1];
958
            return;
959
        }
960
    }else{
961
        const int16_t * C;
962
        int diagonal_ref;
963

    
964
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
965
        
966
        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);
967

    
968
        if(diagonal_ref == ref){ 
969
            *mx= C[0];
970
            *my= C[1];
971
            return;
972
        }
973
    }
974

    
975
    //RARE
976
    pred_motion(h, n, 2, list, ref, mx, my);
977
}
978

    
979
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
980
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
981
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
982

    
983
    tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
984

    
985
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
986
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
987
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
988
       
989
        *mx = *my = 0;
990
        return;
991
    }
992
        
993
    pred_motion(h, 0, 4, 0, 0, mx, my);
994

    
995
    return;
996
}
997

    
998
static inline void write_back_motion(H264Context *h, int mb_type){
999
    MpegEncContext * const s = &h->s;
1000
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1001
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1002
    int list;
1003

    
1004
    for(list=0; list<2; list++){
1005
        int y;
1006
        if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
1007
            if(1){ //FIXME skip or never read if mb_type doesnt use it
1008
                for(y=0; y<4; y++){
1009
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1010
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1011
                }
1012
                if( h->pps.cabac ) {
1013
                    /* FIXME needed ? */
1014
                    for(y=0; y<4; y++){
1015
                        *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1016
                        *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1017
                    }
1018
                }
1019
                for(y=0; y<2; y++){
1020
                    *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1021
                }
1022
            }
1023
            continue; //FIXME direct mode ...
1024
        }
1025
        
1026
        for(y=0; y<4; y++){
1027
            *(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];
1028
            *(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];
1029
        }
1030
        if( h->pps.cabac ) {
1031
            for(y=0; y<4; y++){
1032
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1033
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1034
            }
1035
        }
1036
        for(y=0; y<2; y++){
1037
            s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1038
            s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1039
        }
1040
    }
1041
}
1042

    
1043
/**
1044
 * Decodes a network abstraction layer unit.
1045
 * @param consumed is the number of bytes used as input
1046
 * @param length is the length of the array
1047
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1048
 * @returns decoded bytes, might be src+1 if no escapes 
1049
 */
1050
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1051
    int i, si, di;
1052
    uint8_t *dst;
1053

    
1054
//    src[0]&0x80;                //forbidden bit
1055
    h->nal_ref_idc= src[0]>>5;
1056
    h->nal_unit_type= src[0]&0x1F;
1057

    
1058
    src++; length--;
1059
#if 0    
1060
    for(i=0; i<length; i++)
1061
        printf("%2X ", src[i]);
1062
#endif
1063
    for(i=0; i+1<length; i+=2){
1064
        if(src[i]) continue;
1065
        if(i>0 && src[i-1]==0) i--;
1066
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1067
            if(src[i+2]!=3){
1068
                /* startcode, so we must be past the end */
1069
                length=i;
1070
            }
1071
            break;
1072
        }
1073
    }
1074

    
1075
    if(i>=length-1){ //no escaped 0
1076
        *dst_length= length;
1077
        *consumed= length+1; //+1 for the header
1078
        return src; 
1079
    }
1080

    
1081
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1082
    dst= h->rbsp_buffer;
1083

    
1084
//printf("deoding esc\n");
1085
    si=di=0;
1086
    while(si<length){ 
1087
        //remove escapes (very rare 1:2^22)
1088
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1089
            if(src[si+2]==3){ //escape
1090
                dst[di++]= 0;
1091
                dst[di++]= 0;
1092
                si+=3;
1093
                continue;
1094
            }else //next start code
1095
                break;
1096
        }
1097

    
1098
        dst[di++]= src[si++];
1099
    }
1100

    
1101
    *dst_length= di;
1102
    *consumed= si + 1;//+1 for the header
1103
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1104
    return dst;
1105
}
1106

    
1107
#if 0
1108
/**
1109
 * @param src the data which should be escaped
1110
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1111
 * @param length the length of the src data
1112
 * @param dst_length the length of the dst array
1113
 * @returns length of escaped data in bytes or -1 if an error occured
1114
 */
1115
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1116
    int i, escape_count, si, di;
1117
    uint8_t *temp;
1118
    
1119
    assert(length>=0);
1120
    assert(dst_length>0);
1121
    
1122
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1123

1124
    if(length==0) return 1;
1125

1126
    escape_count= 0;
1127
    for(i=0; i<length; i+=2){
1128
        if(src[i]) continue;
1129
        if(i>0 && src[i-1]==0) 
1130
            i--;
1131
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1132
            escape_count++;
1133
            i+=2;
1134
        }
1135
    }
1136
    
1137
    if(escape_count==0){ 
1138
        if(dst+1 != src)
1139
            memcpy(dst+1, src, length);
1140
        return length + 1;
1141
    }
1142
    
1143
    if(length + escape_count + 1> dst_length)
1144
        return -1;
1145

1146
    //this should be damn rare (hopefully)
1147

1148
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1149
    temp= h->rbsp_buffer;
1150
//printf("encoding esc\n");
1151
    
1152
    si= 0;
1153
    di= 0;
1154
    while(si < length){
1155
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1156
            temp[di++]= 0; si++;
1157
            temp[di++]= 0; si++;
1158
            temp[di++]= 3; 
1159
            temp[di++]= src[si++];
1160
        }
1161
        else
1162
            temp[di++]= src[si++];
1163
    }
1164
    memcpy(dst+1, temp, length+escape_count);
1165
    
1166
    assert(di == length+escape_count);
1167
    
1168
    return di + 1;
1169
}
1170

1171
/**
1172
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1173
 */
1174
static void encode_rbsp_trailing(PutBitContext *pb){
1175
    int length;
1176
    put_bits(pb, 1, 1);
1177
    length= (-put_bits_count(pb))&7;
1178
    if(length) put_bits(pb, length, 0);
1179
}
1180
#endif
1181

    
1182
/**
1183
 * identifies the exact end of the bitstream
1184
 * @return the length of the trailing, or 0 if damaged
1185
 */
1186
static int decode_rbsp_trailing(uint8_t *src){
1187
    int v= *src;
1188
    int r;
1189

    
1190
    tprintf("rbsp trailing %X\n", v);
1191

    
1192
    for(r=1; r<9; r++){
1193
        if(v&1) return r;
1194
        v>>=1;
1195
    }
1196
    return 0;
1197
}
1198

    
1199
/**
1200
 * idct tranforms the 16 dc values and dequantize them.
1201
 * @param qp quantization parameter
1202
 */
1203
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1204
    const int qmul= dequant_coeff[qp][0];
1205
#define stride 16
1206
    int i;
1207
    int temp[16]; //FIXME check if this is a good idea
1208
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1209
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1210

    
1211
//memset(block, 64, 2*256);
1212
//return;
1213
    for(i=0; i<4; i++){
1214
        const int offset= y_offset[i];
1215
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1216
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1217
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1218
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1219

    
1220
        temp[4*i+0]= z0+z3;
1221
        temp[4*i+1]= z1+z2;
1222
        temp[4*i+2]= z1-z2;
1223
        temp[4*i+3]= z0-z3;
1224
    }
1225

    
1226
    for(i=0; i<4; i++){
1227
        const int offset= x_offset[i];
1228
        const int z0= temp[4*0+i] + temp[4*2+i];
1229
        const int z1= temp[4*0+i] - temp[4*2+i];
1230
        const int z2= temp[4*1+i] - temp[4*3+i];
1231
        const int z3= temp[4*1+i] + temp[4*3+i];
1232

    
1233
        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1234
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1235
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1236
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1237
    }
1238
}
1239

    
1240
#if 0
1241
/**
1242
 * dct tranforms the 16 dc values.
1243
 * @param qp quantization parameter ??? FIXME
1244
 */
1245
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1246
//    const int qmul= dequant_coeff[qp][0];
1247
    int i;
1248
    int temp[16]; //FIXME check if this is a good idea
1249
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1250
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1251

1252
    for(i=0; i<4; i++){
1253
        const int offset= y_offset[i];
1254
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1255
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1256
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1257
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1258

1259
        temp[4*i+0]= z0+z3;
1260
        temp[4*i+1]= z1+z2;
1261
        temp[4*i+2]= z1-z2;
1262
        temp[4*i+3]= z0-z3;
1263
    }
1264

1265
    for(i=0; i<4; i++){
1266
        const int offset= x_offset[i];
1267
        const int z0= temp[4*0+i] + temp[4*2+i];
1268
        const int z1= temp[4*0+i] - temp[4*2+i];
1269
        const int z2= temp[4*1+i] - temp[4*3+i];
1270
        const int z3= temp[4*1+i] + temp[4*3+i];
1271

1272
        block[stride*0 +offset]= (z0 + z3)>>1;
1273
        block[stride*2 +offset]= (z1 + z2)>>1;
1274
        block[stride*8 +offset]= (z1 - z2)>>1;
1275
        block[stride*10+offset]= (z0 - z3)>>1;
1276
    }
1277
}
1278
#endif
1279

    
1280
#undef xStride
1281
#undef stride
1282

    
1283
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1284
    const int qmul= dequant_coeff[qp][0];
1285
    const int stride= 16*2;
1286
    const int xStride= 16;
1287
    int a,b,c,d,e;
1288

    
1289
    a= block[stride*0 + xStride*0];
1290
    b= block[stride*0 + xStride*1];
1291
    c= block[stride*1 + xStride*0];
1292
    d= block[stride*1 + xStride*1];
1293

    
1294
    e= a-b;
1295
    a= a+b;
1296
    b= c-d;
1297
    c= c+d;
1298

    
1299
    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1300
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1301
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1302
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1303
}
1304

    
1305
#if 0
1306
static void chroma_dc_dct_c(DCTELEM *block){
1307
    const int stride= 16*2;
1308
    const int xStride= 16;
1309
    int a,b,c,d,e;
1310

1311
    a= block[stride*0 + xStride*0];
1312
    b= block[stride*0 + xStride*1];
1313
    c= block[stride*1 + xStride*0];
1314
    d= block[stride*1 + xStride*1];
1315

1316
    e= a-b;
1317
    a= a+b;
1318
    b= c-d;
1319
    c= c+d;
1320

1321
    block[stride*0 + xStride*0]= (a+c);
1322
    block[stride*0 + xStride*1]= (e+b);
1323
    block[stride*1 + xStride*0]= (a-c);
1324
    block[stride*1 + xStride*1]= (e-b);
1325
}
1326
#endif
1327

    
1328
/**
1329
 * gets the chroma qp.
1330
 */
1331
static inline int get_chroma_qp(H264Context *h, int qscale){
1332
    
1333
    return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1334
}
1335

    
1336

    
1337
#if 0
1338
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1339
    int i;
1340
    //FIXME try int temp instead of block
1341
    
1342
    for(i=0; i<4; i++){
1343
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1344
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1345
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1346
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1347
        const int z0= d0 + d3;
1348
        const int z3= d0 - d3;
1349
        const int z1= d1 + d2;
1350
        const int z2= d1 - d2;
1351
        
1352
        block[0 + 4*i]=   z0 +   z1;
1353
        block[1 + 4*i]= 2*z3 +   z2;
1354
        block[2 + 4*i]=   z0 -   z1;
1355
        block[3 + 4*i]=   z3 - 2*z2;
1356
    }    
1357

1358
    for(i=0; i<4; i++){
1359
        const int z0= block[0*4 + i] + block[3*4 + i];
1360
        const int z3= block[0*4 + i] - block[3*4 + i];
1361
        const int z1= block[1*4 + i] + block[2*4 + i];
1362
        const int z2= block[1*4 + i] - block[2*4 + i];
1363
        
1364
        block[0*4 + i]=   z0 +   z1;
1365
        block[1*4 + i]= 2*z3 +   z2;
1366
        block[2*4 + i]=   z0 -   z1;
1367
        block[3*4 + i]=   z3 - 2*z2;
1368
    }
1369
}
1370
#endif
1371

    
1372
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1373
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1374
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1375
    int i;
1376
    const int * const quant_table= quant_coeff[qscale];
1377
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1378
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1379
    const unsigned int threshold2= (threshold1<<1);
1380
    int last_non_zero;
1381

    
1382
    if(seperate_dc){
1383
        if(qscale<=18){
1384
            //avoid overflows
1385
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1386
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1387
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1388

    
1389
            int level= block[0]*quant_coeff[qscale+18][0];
1390
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1391
                if(level>0){
1392
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1393
                    block[0]= level;
1394
                }else{
1395
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1396
                    block[0]= -level;
1397
                }
1398
//                last_non_zero = i;
1399
            }else{
1400
                block[0]=0;
1401
            }
1402
        }else{
1403
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1404
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1405
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1406

    
1407
            int level= block[0]*quant_table[0];
1408
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1409
                if(level>0){
1410
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1411
                    block[0]= level;
1412
                }else{
1413
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1414
                    block[0]= -level;
1415
                }
1416
//                last_non_zero = i;
1417
            }else{
1418
                block[0]=0;
1419
            }
1420
        }
1421
        last_non_zero= 0;
1422
        i=1;
1423
    }else{
1424
        last_non_zero= -1;
1425
        i=0;
1426
    }
1427

    
1428
    for(; i<16; i++){
1429
        const int j= scantable[i];
1430
        int level= block[j]*quant_table[j];
1431

    
1432
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1433
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1434
        if(((unsigned)(level+threshold1))>threshold2){
1435
            if(level>0){
1436
                level= (bias + level)>>QUANT_SHIFT;
1437
                block[j]= level;
1438
            }else{
1439
                level= (bias - level)>>QUANT_SHIFT;
1440
                block[j]= -level;
1441
            }
1442
            last_non_zero = i;
1443
        }else{
1444
            block[j]=0;
1445
        }
1446
    }
1447

    
1448
    return last_non_zero;
1449
}
1450

    
1451
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1452
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1453
    ((uint32_t*)(src+0*stride))[0]= a;
1454
    ((uint32_t*)(src+1*stride))[0]= a;
1455
    ((uint32_t*)(src+2*stride))[0]= a;
1456
    ((uint32_t*)(src+3*stride))[0]= a;
1457
}
1458

    
1459
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1460
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1461
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1462
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1463
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1464
}
1465

    
1466
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1467
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1468
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1469
    
1470
    ((uint32_t*)(src+0*stride))[0]= 
1471
    ((uint32_t*)(src+1*stride))[0]= 
1472
    ((uint32_t*)(src+2*stride))[0]= 
1473
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1474
}
1475

    
1476
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1477
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1478
    
1479
    ((uint32_t*)(src+0*stride))[0]= 
1480
    ((uint32_t*)(src+1*stride))[0]= 
1481
    ((uint32_t*)(src+2*stride))[0]= 
1482
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1483
}
1484

    
1485
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1486
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1487
    
1488
    ((uint32_t*)(src+0*stride))[0]= 
1489
    ((uint32_t*)(src+1*stride))[0]= 
1490
    ((uint32_t*)(src+2*stride))[0]= 
1491
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1492
}
1493

    
1494
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1495
    ((uint32_t*)(src+0*stride))[0]= 
1496
    ((uint32_t*)(src+1*stride))[0]= 
1497
    ((uint32_t*)(src+2*stride))[0]= 
1498
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1499
}
1500

    
1501

    
1502
#define LOAD_TOP_RIGHT_EDGE\
1503
    const int t4= topright[0];\
1504
    const int t5= topright[1];\
1505
    const int t6= topright[2];\
1506
    const int t7= topright[3];\
1507

    
1508
#define LOAD_LEFT_EDGE\
1509
    const int l0= src[-1+0*stride];\
1510
    const int l1= src[-1+1*stride];\
1511
    const int l2= src[-1+2*stride];\
1512
    const int l3= src[-1+3*stride];\
1513

    
1514
#define LOAD_TOP_EDGE\
1515
    const int t0= src[ 0-1*stride];\
1516
    const int t1= src[ 1-1*stride];\
1517
    const int t2= src[ 2-1*stride];\
1518
    const int t3= src[ 3-1*stride];\
1519

    
1520
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1521
    const int lt= src[-1-1*stride];
1522
    LOAD_TOP_EDGE
1523
    LOAD_LEFT_EDGE
1524

    
1525
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1526
    src[0+2*stride]=
1527
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1528
    src[0+1*stride]=
1529
    src[1+2*stride]=
1530
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1531
    src[0+0*stride]=
1532
    src[1+1*stride]=
1533
    src[2+2*stride]=
1534
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1535
    src[1+0*stride]=
1536
    src[2+1*stride]=
1537
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1538
    src[2+0*stride]=
1539
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1540
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1541
}
1542

    
1543
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1544
    LOAD_TOP_EDGE    
1545
    LOAD_TOP_RIGHT_EDGE    
1546
//    LOAD_LEFT_EDGE    
1547

    
1548
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1549
    src[1+0*stride]=
1550
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1551
    src[2+0*stride]=
1552
    src[1+1*stride]=
1553
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1554
    src[3+0*stride]=
1555
    src[2+1*stride]=
1556
    src[1+2*stride]=
1557
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1558
    src[3+1*stride]=
1559
    src[2+2*stride]=
1560
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1561
    src[3+2*stride]=
1562
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1563
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1564
}
1565

    
1566
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1567
    const int lt= src[-1-1*stride];
1568
    LOAD_TOP_EDGE    
1569
    LOAD_LEFT_EDGE    
1570
    const __attribute__((unused)) int unu= l3;
1571

    
1572
    src[0+0*stride]=
1573
    src[1+2*stride]=(lt + t0 + 1)>>1;
1574
    src[1+0*stride]=
1575
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1576
    src[2+0*stride]=
1577
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1578
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1579
    src[0+1*stride]=
1580
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1581
    src[1+1*stride]=
1582
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1583
    src[2+1*stride]=
1584
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1585
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1586
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1587
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1588
}
1589

    
1590
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1591
    LOAD_TOP_EDGE    
1592
    LOAD_TOP_RIGHT_EDGE    
1593
    const __attribute__((unused)) int unu= t7;
1594

    
1595
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1596
    src[1+0*stride]=
1597
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1598
    src[2+0*stride]=
1599
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1600
    src[3+0*stride]=
1601
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1602
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1603
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1604
    src[1+1*stride]=
1605
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1606
    src[2+1*stride]=
1607
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1608
    src[3+1*stride]=
1609
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1610
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1611
}
1612

    
1613
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1614
    LOAD_LEFT_EDGE    
1615

    
1616
    src[0+0*stride]=(l0 + l1 + 1)>>1;
1617
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1618
    src[2+0*stride]=
1619
    src[0+1*stride]=(l1 + l2 + 1)>>1;
1620
    src[3+0*stride]=
1621
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1622
    src[2+1*stride]=
1623
    src[0+2*stride]=(l2 + l3 + 1)>>1;
1624
    src[3+1*stride]=
1625
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1626
    src[3+2*stride]=
1627
    src[1+3*stride]=
1628
    src[0+3*stride]=
1629
    src[2+2*stride]=
1630
    src[2+3*stride]=
1631
    src[3+3*stride]=l3;
1632
}
1633
    
1634
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1635
    const int lt= src[-1-1*stride];
1636
    LOAD_TOP_EDGE    
1637
    LOAD_LEFT_EDGE    
1638
    const __attribute__((unused)) int unu= t3;
1639

    
1640
    src[0+0*stride]=
1641
    src[2+1*stride]=(lt + l0 + 1)>>1;
1642
    src[1+0*stride]=
1643
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1644
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1645
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1646
    src[0+1*stride]=
1647
    src[2+2*stride]=(l0 + l1 + 1)>>1;
1648
    src[1+1*stride]=
1649
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1650
    src[0+2*stride]=
1651
    src[2+3*stride]=(l1 + l2+ 1)>>1;
1652
    src[1+2*stride]=
1653
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1654
    src[0+3*stride]=(l2 + l3 + 1)>>1;
1655
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1656
}
1657

    
1658
static void pred16x16_vertical_c(uint8_t *src, int stride){
1659
    int i;
1660
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1661
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1662
    const uint32_t c= ((uint32_t*)(src-stride))[2];
1663
    const uint32_t d= ((uint32_t*)(src-stride))[3];
1664
    
1665
    for(i=0; i<16; i++){
1666
        ((uint32_t*)(src+i*stride))[0]= a;
1667
        ((uint32_t*)(src+i*stride))[1]= b;
1668
        ((uint32_t*)(src+i*stride))[2]= c;
1669
        ((uint32_t*)(src+i*stride))[3]= d;
1670
    }
1671
}
1672

    
1673
static void pred16x16_horizontal_c(uint8_t *src, int stride){
1674
    int i;
1675

    
1676
    for(i=0; i<16; i++){
1677
        ((uint32_t*)(src+i*stride))[0]=
1678
        ((uint32_t*)(src+i*stride))[1]=
1679
        ((uint32_t*)(src+i*stride))[2]=
1680
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1681
    }
1682
}
1683

    
1684
static void pred16x16_dc_c(uint8_t *src, int stride){
1685
    int i, dc=0;
1686

    
1687
    for(i=0;i<16; i++){
1688
        dc+= src[-1+i*stride];
1689
    }
1690
    
1691
    for(i=0;i<16; i++){
1692
        dc+= src[i-stride];
1693
    }
1694

    
1695
    dc= 0x01010101*((dc + 16)>>5);
1696

    
1697
    for(i=0; i<16; i++){
1698
        ((uint32_t*)(src+i*stride))[0]=
1699
        ((uint32_t*)(src+i*stride))[1]=
1700
        ((uint32_t*)(src+i*stride))[2]=
1701
        ((uint32_t*)(src+i*stride))[3]= dc;
1702
    }
1703
}
1704

    
1705
static void pred16x16_left_dc_c(uint8_t *src, int stride){
1706
    int i, dc=0;
1707

    
1708
    for(i=0;i<16; i++){
1709
        dc+= src[-1+i*stride];
1710
    }
1711
    
1712
    dc= 0x01010101*((dc + 8)>>4);
1713

    
1714
    for(i=0; i<16; i++){
1715
        ((uint32_t*)(src+i*stride))[0]=
1716
        ((uint32_t*)(src+i*stride))[1]=
1717
        ((uint32_t*)(src+i*stride))[2]=
1718
        ((uint32_t*)(src+i*stride))[3]= dc;
1719
    }
1720
}
1721

    
1722
static void pred16x16_top_dc_c(uint8_t *src, int stride){
1723
    int i, dc=0;
1724

    
1725
    for(i=0;i<16; i++){
1726
        dc+= src[i-stride];
1727
    }
1728
    dc= 0x01010101*((dc + 8)>>4);
1729

    
1730
    for(i=0; i<16; i++){
1731
        ((uint32_t*)(src+i*stride))[0]=
1732
        ((uint32_t*)(src+i*stride))[1]=
1733
        ((uint32_t*)(src+i*stride))[2]=
1734
        ((uint32_t*)(src+i*stride))[3]= dc;
1735
    }
1736
}
1737

    
1738
static void pred16x16_128_dc_c(uint8_t *src, int stride){
1739
    int i;
1740

    
1741
    for(i=0; i<16; i++){
1742
        ((uint32_t*)(src+i*stride))[0]=
1743
        ((uint32_t*)(src+i*stride))[1]=
1744
        ((uint32_t*)(src+i*stride))[2]=
1745
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1746
    }
1747
}
1748

    
1749
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1750
  int i, j, k;
1751
  int a;
1752
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
1753
  const uint8_t * const src0 = src+7-stride;
1754
  const uint8_t *src1 = src+8*stride-1;
1755
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
1756
  int H = src0[1] - src0[-1];
1757
  int V = src1[0] - src2[ 0];
1758
  for(k=2; k<=8; ++k) {
1759
    src1 += stride; src2 -= stride;
1760
    H += k*(src0[k] - src0[-k]);
1761
    V += k*(src1[0] - src2[ 0]);
1762
  }
1763
  if(svq3){
1764
    H = ( 5*(H/4) ) / 16;
1765
    V = ( 5*(V/4) ) / 16;
1766

    
1767
    /* required for 100% accuracy */
1768
    i = H; H = V; V = i;
1769
  }else{
1770
    H = ( 5*H+32 ) >> 6;
1771
    V = ( 5*V+32 ) >> 6;
1772
  }
1773

    
1774
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1775
  for(j=16; j>0; --j) {
1776
    int b = a;
1777
    a += V;
1778
    for(i=-16; i<0; i+=4) {
1779
      src[16+i] = cm[ (b    ) >> 5 ];
1780
      src[17+i] = cm[ (b+  H) >> 5 ];
1781
      src[18+i] = cm[ (b+2*H) >> 5 ];
1782
      src[19+i] = cm[ (b+3*H) >> 5 ];
1783
      b += 4*H;
1784
    }
1785
    src += stride;
1786
  }
1787
}
1788

    
1789
static void pred16x16_plane_c(uint8_t *src, int stride){
1790
    pred16x16_plane_compat_c(src, stride, 0);
1791
}
1792

    
1793
static void pred8x8_vertical_c(uint8_t *src, int stride){
1794
    int i;
1795
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1796
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1797
    
1798
    for(i=0; i<8; i++){
1799
        ((uint32_t*)(src+i*stride))[0]= a;
1800
        ((uint32_t*)(src+i*stride))[1]= b;
1801
    }
1802
}
1803

    
1804
static void pred8x8_horizontal_c(uint8_t *src, int stride){
1805
    int i;
1806

    
1807
    for(i=0; i<8; i++){
1808
        ((uint32_t*)(src+i*stride))[0]=
1809
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1810
    }
1811
}
1812

    
1813
static void pred8x8_128_dc_c(uint8_t *src, int stride){
1814
    int i;
1815

    
1816
    for(i=0; i<4; i++){
1817
        ((uint32_t*)(src+i*stride))[0]= 
1818
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1819
    }
1820
    for(i=4; i<8; i++){
1821
        ((uint32_t*)(src+i*stride))[0]= 
1822
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1823
    }
1824
}
1825

    
1826
static void pred8x8_left_dc_c(uint8_t *src, int stride){
1827
    int i;
1828
    int dc0, dc2;
1829

    
1830
    dc0=dc2=0;
1831
    for(i=0;i<4; i++){
1832
        dc0+= src[-1+i*stride];
1833
        dc2+= src[-1+(i+4)*stride];
1834
    }
1835
    dc0= 0x01010101*((dc0 + 2)>>2);
1836
    dc2= 0x01010101*((dc2 + 2)>>2);
1837

    
1838
    for(i=0; i<4; i++){
1839
        ((uint32_t*)(src+i*stride))[0]=
1840
        ((uint32_t*)(src+i*stride))[1]= dc0;
1841
    }
1842
    for(i=4; i<8; i++){
1843
        ((uint32_t*)(src+i*stride))[0]=
1844
        ((uint32_t*)(src+i*stride))[1]= dc2;
1845
    }
1846
}
1847

    
1848
static void pred8x8_top_dc_c(uint8_t *src, int stride){
1849
    int i;
1850
    int dc0, dc1;
1851

    
1852
    dc0=dc1=0;
1853
    for(i=0;i<4; i++){
1854
        dc0+= src[i-stride];
1855
        dc1+= src[4+i-stride];
1856
    }
1857
    dc0= 0x01010101*((dc0 + 2)>>2);
1858
    dc1= 0x01010101*((dc1 + 2)>>2);
1859

    
1860
    for(i=0; i<4; i++){
1861
        ((uint32_t*)(src+i*stride))[0]= dc0;
1862
        ((uint32_t*)(src+i*stride))[1]= dc1;
1863
    }
1864
    for(i=4; i<8; i++){
1865
        ((uint32_t*)(src+i*stride))[0]= dc0;
1866
        ((uint32_t*)(src+i*stride))[1]= dc1;
1867
    }
1868
}
1869

    
1870

    
1871
static void pred8x8_dc_c(uint8_t *src, int stride){
1872
    int i;
1873
    int dc0, dc1, dc2, dc3;
1874

    
1875
    dc0=dc1=dc2=0;
1876
    for(i=0;i<4; i++){
1877
        dc0+= src[-1+i*stride] + src[i-stride];
1878
        dc1+= src[4+i-stride];
1879
        dc2+= src[-1+(i+4)*stride];
1880
    }
1881
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1882
    dc0= 0x01010101*((dc0 + 4)>>3);
1883
    dc1= 0x01010101*((dc1 + 2)>>2);
1884
    dc2= 0x01010101*((dc2 + 2)>>2);
1885

    
1886
    for(i=0; i<4; i++){
1887
        ((uint32_t*)(src+i*stride))[0]= dc0;
1888
        ((uint32_t*)(src+i*stride))[1]= dc1;
1889
    }
1890
    for(i=4; i<8; i++){
1891
        ((uint32_t*)(src+i*stride))[0]= dc2;
1892
        ((uint32_t*)(src+i*stride))[1]= dc3;
1893
    }
1894
}
1895

    
1896
static void pred8x8_plane_c(uint8_t *src, int stride){
1897
  int j, k;
1898
  int a;
1899
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
1900
  const uint8_t * const src0 = src+3-stride;
1901
  const uint8_t *src1 = src+4*stride-1;
1902
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
1903
  int H = src0[1] - src0[-1];
1904
  int V = src1[0] - src2[ 0];
1905
  for(k=2; k<=4; ++k) {
1906
    src1 += stride; src2 -= stride;
1907
    H += k*(src0[k] - src0[-k]);
1908
    V += k*(src1[0] - src2[ 0]);
1909
  }
1910
  H = ( 17*H+16 ) >> 5;
1911
  V = ( 17*V+16 ) >> 5;
1912

    
1913
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1914
  for(j=8; j>0; --j) {
1915
    int b = a;
1916
    a += V;
1917
    src[0] = cm[ (b    ) >> 5 ];
1918
    src[1] = cm[ (b+  H) >> 5 ];
1919
    src[2] = cm[ (b+2*H) >> 5 ];
1920
    src[3] = cm[ (b+3*H) >> 5 ];
1921
    src[4] = cm[ (b+4*H) >> 5 ];
1922
    src[5] = cm[ (b+5*H) >> 5 ];
1923
    src[6] = cm[ (b+6*H) >> 5 ];
1924
    src[7] = cm[ (b+7*H) >> 5 ];
1925
    src += stride;
1926
  }
1927
}
1928

    
1929
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1930
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1931
                           int src_x_offset, int src_y_offset,
1932
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1933
    MpegEncContext * const s = &h->s;
1934
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1935
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1936
    const int luma_xy= (mx&3) + ((my&3)<<2);
1937
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1938
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1939
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1940
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1941
    int extra_height= extra_width;
1942
    int emu=0;
1943
    const int full_mx= mx>>2;
1944
    const int full_my= my>>2;
1945
    
1946
    assert(pic->data[0]);
1947
    
1948
    if(mx&7) extra_width -= 3;
1949
    if(my&7) extra_height -= 3;
1950
    
1951
    if(   full_mx < 0-extra_width 
1952
       || full_my < 0-extra_height 
1953
       || full_mx + 16/*FIXME*/ > s->width + extra_width 
1954
       || full_my + 16/*FIXME*/ > s->height + extra_height){
1955
        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);
1956
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1957
        emu=1;
1958
    }
1959
    
1960
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1961
    if(!square){
1962
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1963
    }
1964
    
1965
    if(s->flags&CODEC_FLAG_GRAY) return;
1966
    
1967
    if(emu){
1968
        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);
1969
            src_cb= s->edge_emu_buffer;
1970
    }
1971
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1972

    
1973
    if(emu){
1974
        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);
1975
            src_cr= s->edge_emu_buffer;
1976
    }
1977
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1978
}
1979

    
1980
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1981
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1982
                           int x_offset, int y_offset,
1983
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1984
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1985
                           int list0, int list1){
1986
    MpegEncContext * const s = &h->s;
1987
    qpel_mc_func *qpix_op=  qpix_put;
1988
    h264_chroma_mc_func chroma_op= chroma_put;
1989
    
1990
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
1991
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
1992
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
1993
    x_offset += 8*s->mb_x;
1994
    y_offset += 8*s->mb_y;
1995
    
1996
    if(list0){
1997
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1998
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1999
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2000
                           qpix_op, chroma_op);
2001

    
2002
        qpix_op=  qpix_avg;
2003
        chroma_op= chroma_avg;
2004
    }
2005

    
2006
    if(list1){
2007
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2008
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2009
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2010
                           qpix_op, chroma_op);
2011
    }
2012
}
2013

    
2014
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2015
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2016
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
2017
    MpegEncContext * const s = &h->s;
2018
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2019
    const int mb_type= s->current_picture.mb_type[mb_xy];
2020
    
2021
    assert(IS_INTER(mb_type));
2022
    
2023
    if(IS_16X16(mb_type)){
2024
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2025
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2026
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2027
    }else if(IS_16X8(mb_type)){
2028
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2029
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2030
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2031
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2032
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2033
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2034
    }else if(IS_8X16(mb_type)){
2035
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2036
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2037
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2038
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2039
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2040
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2041
    }else{
2042
        int i;
2043
        
2044
        assert(IS_8X8(mb_type));
2045

    
2046
        for(i=0; i<4; i++){
2047
            const int sub_mb_type= h->sub_mb_type[i];
2048
            const int n= 4*i;
2049
            int x_offset= (i&1)<<2;
2050
            int y_offset= (i&2)<<1;
2051

    
2052
            if(IS_SUB_8X8(sub_mb_type)){
2053
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2054
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2055
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2056
            }else if(IS_SUB_8X4(sub_mb_type)){
2057
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2058
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2059
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2060
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2061
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2062
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2063
            }else if(IS_SUB_4X8(sub_mb_type)){
2064
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2065
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2066
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2067
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2068
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2069
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2070
            }else{
2071
                int j;
2072
                assert(IS_SUB_4X4(sub_mb_type));
2073
                for(j=0; j<4; j++){
2074
                    int sub_x_offset= x_offset + 2*(j&1);
2075
                    int sub_y_offset= y_offset +   (j&2);
2076
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2077
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2078
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2079
                }
2080
            }
2081
        }
2082
    }
2083
}
2084

    
2085
static void decode_init_vlc(H264Context *h){
2086
    static int done = 0;
2087

    
2088
    if (!done) {
2089
        int i;
2090
        done = 1;
2091

    
2092
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2093
                 &chroma_dc_coeff_token_len [0], 1, 1,
2094
                 &chroma_dc_coeff_token_bits[0], 1, 1);
2095

    
2096
        for(i=0; i<4; i++){
2097
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2098
                     &coeff_token_len [i][0], 1, 1,
2099
                     &coeff_token_bits[i][0], 1, 1);
2100
        }
2101

    
2102
        for(i=0; i<3; i++){
2103
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2104
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2105
                     &chroma_dc_total_zeros_bits[i][0], 1, 1);
2106
        }
2107
        for(i=0; i<15; i++){
2108
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2109
                     &total_zeros_len [i][0], 1, 1,
2110
                     &total_zeros_bits[i][0], 1, 1);
2111
        }
2112

    
2113
        for(i=0; i<6; i++){
2114
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2115
                     &run_len [i][0], 1, 1,
2116
                     &run_bits[i][0], 1, 1);
2117
        }
2118
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2119
                 &run_len [6][0], 1, 1,
2120
                 &run_bits[6][0], 1, 1);
2121
    }
2122
}
2123

    
2124
/**
2125
 * Sets the intra prediction function pointers.
2126
 */
2127
static void init_pred_ptrs(H264Context *h){
2128
//    MpegEncContext * const s = &h->s;
2129

    
2130
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2131
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2132
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2133
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2134
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2135
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2136
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2137
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2138
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2139
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2140
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2141
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2142

    
2143
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2144
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2145
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2146
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2147
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2148
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2149
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2150

    
2151
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2152
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2153
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2154
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2155
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2156
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2157
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2158
}
2159

    
2160
static void free_tables(H264Context *h){
2161
    av_freep(&h->intra4x4_pred_mode);
2162
    av_freep(&h->chroma_pred_mode_table);
2163
    av_freep(&h->cbp_table);
2164
    av_freep(&h->mvd_table[0]);
2165
    av_freep(&h->mvd_table[1]);
2166
    av_freep(&h->non_zero_count);
2167
    av_freep(&h->slice_table_base);
2168
    av_freep(&h->top_border);
2169
    h->slice_table= NULL;
2170

    
2171
    av_freep(&h->mb2b_xy);
2172
    av_freep(&h->mb2b8_xy);
2173
}
2174

    
2175
/**
2176
 * allocates tables.
2177
 * needs widzh/height
2178
 */
2179
static int alloc_tables(H264Context *h){
2180
    MpegEncContext * const s = &h->s;
2181
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2182
    int x,y;
2183

    
2184
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2185

    
2186
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2187
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2188
    CHECKED_ALLOCZ(h->top_border       , s->mb_width * (16+8+8) * sizeof(uint8_t))
2189
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2190

    
2191
    if( h->pps.cabac ) {
2192
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2193
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2194
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2195
    }
2196

    
2197
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2198
    h->slice_table= h->slice_table_base + s->mb_stride + 1;
2199

    
2200
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint16_t));
2201
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2202
    for(y=0; y<s->mb_height; y++){
2203
        for(x=0; x<s->mb_width; x++){
2204
            const int mb_xy= x + y*s->mb_stride;
2205
            const int b_xy = 4*x + 4*y*h->b_stride;
2206
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2207
        
2208
            h->mb2b_xy [mb_xy]= b_xy;
2209
            h->mb2b8_xy[mb_xy]= b8_xy;
2210
        }
2211
    }
2212
    
2213
    return 0;
2214
fail:
2215
    free_tables(h);
2216
    return -1;
2217
}
2218

    
2219
static void common_init(H264Context *h){
2220
    MpegEncContext * const s = &h->s;
2221

    
2222
    s->width = s->avctx->width;
2223
    s->height = s->avctx->height;
2224
    s->codec_id= s->avctx->codec->id;
2225
    
2226
    init_pred_ptrs(h);
2227

    
2228
    s->unrestricted_mv=1;
2229
    s->decode=1; //FIXME
2230
}
2231

    
2232
static int decode_init(AVCodecContext *avctx){
2233
    H264Context *h= avctx->priv_data;
2234
    MpegEncContext * const s = &h->s;
2235

    
2236
    MPV_decode_defaults(s);
2237
    
2238
    s->avctx = avctx;
2239
    common_init(h);
2240

    
2241
    s->out_format = FMT_H264;
2242
    s->workaround_bugs= avctx->workaround_bugs;
2243

    
2244
    // set defaults
2245
//    s->decode_mb= ff_h263_decode_mb;
2246
    s->low_delay= 1;
2247
    avctx->pix_fmt= PIX_FMT_YUV420P;
2248

    
2249
    decode_init_vlc(h);
2250
    
2251
    if(avctx->codec_tag != 0x31637661) // avc1
2252
        h->is_avc = 0;
2253
    else {
2254
        if((avctx->extradata_size == 0) || (avctx->extradata == NULL)) {
2255
            av_log(avctx, AV_LOG_ERROR, "AVC codec requires avcC data\n");
2256
            return -1;
2257
        }
2258
        h->is_avc = 1;
2259
        h->got_avcC = 0;
2260
    }
2261

    
2262
    return 0;
2263
}
2264

    
2265
static void frame_start(H264Context *h){
2266
    MpegEncContext * const s = &h->s;
2267
    int i;
2268

    
2269
    MPV_frame_start(s, s->avctx);
2270
    ff_er_frame_start(s);
2271
    h->mmco_index=0;
2272

    
2273
    assert(s->linesize && s->uvlinesize);
2274

    
2275
    for(i=0; i<16; i++){
2276
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2277
        h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2278
    }
2279
    for(i=0; i<4; i++){
2280
        h->block_offset[16+i]=
2281
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2282
    }
2283

    
2284
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2285
}
2286

    
2287
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2288
    MpegEncContext * const s = &h->s;
2289
    int i;
2290
    
2291
    src_y  -=   linesize;
2292
    src_cb -= uvlinesize;
2293
    src_cr -= uvlinesize;
2294

    
2295
    h->left_border[0]= h->top_border[s->mb_x][15];
2296
    for(i=1; i<17; i++){
2297
        h->left_border[i]= src_y[15+i*  linesize];
2298
    }
2299
    
2300
    *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2301
    *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2302

    
2303
    if(!(s->flags&CODEC_FLAG_GRAY)){
2304
        h->left_border[17  ]= h->top_border[s->mb_x][16+7];
2305
        h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2306
        for(i=1; i<9; i++){
2307
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
2308
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2309
        }
2310
        *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2311
        *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2312
    }
2313
}
2314

    
2315
static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
2316
    MpegEncContext * const s = &h->s;
2317
    int temp8, i;
2318
    uint64_t temp64;
2319
    int deblock_left = (s->mb_x > 0);
2320
    int deblock_top  = (s->mb_y > 0);
2321

    
2322
    src_y  -=   linesize + 1;
2323
    src_cb -= uvlinesize + 1;
2324
    src_cr -= uvlinesize + 1;
2325

    
2326
#define XCHG(a,b,t,xchg)\
2327
t= a;\
2328
if(xchg)\
2329
    a= b;\
2330
b= t;
2331

    
2332
    if(deblock_left){
2333
        for(i = !deblock_top; i<17; i++){
2334
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2335
        }
2336
    }
2337

    
2338
    if(deblock_top){
2339
        XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2340
        XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2341
    }
2342

    
2343
    if(!(s->flags&CODEC_FLAG_GRAY)){
2344
        if(deblock_left){
2345
            for(i = !deblock_top; i<9; i++){
2346
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
2347
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2348
            }
2349
        }
2350
        if(deblock_top){
2351
            XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2352
            XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2353
        }
2354
    }
2355
}
2356

    
2357
static void hl_decode_mb(H264Context *h){
2358
    MpegEncContext * const s = &h->s;
2359
    const int mb_x= s->mb_x;
2360
    const int mb_y= s->mb_y;
2361
    const int mb_xy= mb_x + mb_y*s->mb_stride;
2362
    const int mb_type= s->current_picture.mb_type[mb_xy];
2363
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2364
    int linesize, uvlinesize /*dct_offset*/;
2365
    int i;
2366

    
2367
    if(!s->decode)
2368
        return;
2369

    
2370
    if(s->mb_skiped){
2371
    }
2372

    
2373
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2374
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2375
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2376

    
2377
    if (h->mb_field_decoding_flag) {
2378
        linesize = s->linesize * 2;
2379
        uvlinesize = s->uvlinesize * 2;
2380
        if(mb_y&1){ //FIXME move out of this func?
2381
            dest_y -= s->linesize*15;
2382
            dest_cb-= s->linesize*7;
2383
            dest_cr-= s->linesize*7;
2384
        }
2385
    } else {
2386
        linesize = s->linesize;
2387
        uvlinesize = s->uvlinesize;
2388
//        dct_offset = s->linesize * 16;
2389
    }
2390

    
2391
    if(IS_INTRA(mb_type)){
2392
        if(h->deblocking_filter)
2393
            xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2394

    
2395
        if(!(s->flags&CODEC_FLAG_GRAY)){
2396
            h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2397
            h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2398
        }
2399

    
2400
        if(IS_INTRA4x4(mb_type)){
2401
            if(!s->encoding){
2402
                for(i=0; i<16; i++){
2403
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2404
                    uint8_t *topright= ptr + 4 - linesize;
2405
                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2406
                    const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2407
                    int tr;
2408

    
2409
                    if(!topright_avail){
2410
                        tr= ptr[3 - linesize]*0x01010101;
2411
                        topright= (uint8_t*) &tr;
2412
                    }else if(i==5 && h->deblocking_filter){
2413
                        tr= *(uint32_t*)h->top_border[mb_x+1];
2414
                        topright= (uint8_t*) &tr;
2415
                    }
2416

    
2417
                    h->pred4x4[ dir ](ptr, topright, linesize);
2418
                    if(h->non_zero_count_cache[ scan8[i] ]){
2419
                        if(s->codec_id == CODEC_ID_H264)
2420
                            s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2421
                        else
2422
                            svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2423
                    }
2424
                }
2425
            }
2426
        }else{
2427
            h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2428
            if(s->codec_id == CODEC_ID_H264)
2429
                h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2430
            else
2431
                svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2432
        }
2433
        if(h->deblocking_filter)
2434
            xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2435
    }else if(s->codec_id == CODEC_ID_H264){
2436
        hl_motion(h, dest_y, dest_cb, dest_cr,
2437
                  s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
2438
                  s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2439
    }
2440

    
2441

    
2442
    if(!IS_INTRA4x4(mb_type)){
2443
        if(s->codec_id == CODEC_ID_H264){
2444
            for(i=0; i<16; i++){
2445
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2446
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2447
                    s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2448
                }
2449
            }
2450
        }else{
2451
            for(i=0; i<16; i++){
2452
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2453
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2454
                    svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2455
                }
2456
            }
2457
        }
2458
    }
2459

    
2460
    if(!(s->flags&CODEC_FLAG_GRAY)){
2461
        chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2462
        chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2463
        if(s->codec_id == CODEC_ID_H264){
2464
            for(i=16; i<16+4; i++){
2465
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2466
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2467
                    s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2468
                }
2469
            }
2470
            for(i=20; i<20+4; i++){
2471
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2472
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2473
                    s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2474
                }
2475
            }
2476
        }else{
2477
            for(i=16; i<16+4; i++){
2478
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2479
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2480
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2481
                }
2482
            }
2483
            for(i=20; i<20+4; i++){
2484
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2485
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2486
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2487
                }
2488
            }
2489
        }
2490
    }
2491
    if(h->deblocking_filter) {
2492
        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2493
        filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2494
    }
2495
}
2496

    
2497
/**
2498
 * fills the default_ref_list.
2499
 */
2500
static int fill_default_ref_list(H264Context *h){
2501
    MpegEncContext * const s = &h->s;
2502
    int i;
2503
    Picture sorted_short_ref[16];
2504
    
2505
    if(h->slice_type==B_TYPE){
2506
        int out_i;
2507
        int limit= -1;
2508

    
2509
        for(out_i=0; out_i<h->short_ref_count; out_i++){
2510
            int best_i=-1;
2511
            int best_poc=INT_MAX;
2512

    
2513
            for(i=0; i<h->short_ref_count; i++){
2514
                const int poc= h->short_ref[i]->poc;
2515
                if(poc > limit && poc < best_poc){
2516
                    best_poc= poc;
2517
                    best_i= i;
2518
                }
2519
            }
2520
            
2521
            assert(best_i != -1);
2522
            
2523
            limit= best_poc;
2524
            sorted_short_ref[out_i]= *h->short_ref[best_i];
2525
        }
2526
    }
2527

    
2528
    if(s->picture_structure == PICT_FRAME){
2529
        if(h->slice_type==B_TYPE){
2530
            const int current_poc= s->current_picture_ptr->poc;
2531
            int list;
2532

    
2533
            for(list=0; list<2; list++){
2534
                int index=0;
2535

    
2536
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2537
                    const int i2= list ? h->short_ref_count - i - 1 : i;
2538
                    const int poc= sorted_short_ref[i2].poc;
2539
                    
2540
                    if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2541

    
2542
                    if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2543
                        h->default_ref_list[list][index  ]= sorted_short_ref[i2];
2544
                        h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2545
                    }
2546
                }
2547

    
2548
                for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2549
                    if(h->long_ref[i]->reference != 3) continue;
2550

    
2551
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
2552
                    h->default_ref_list[ list ][index++].pic_id= i;;
2553
                }
2554
                
2555
                if(h->long_ref_count > 1 && h->short_ref_count==0){
2556
                    Picture temp= h->default_ref_list[1][0];
2557
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
2558
                    h->default_ref_list[1][0] = temp;
2559
                }
2560

    
2561
                if(index < h->ref_count[ list ])
2562
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2563
            }
2564
        }else{
2565
            int index=0;
2566
            for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2567
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2568
                h->default_ref_list[0][index  ]= *h->short_ref[i];
2569
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2570
            }
2571
            for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2572
                if(h->long_ref[i]->reference != 3) continue;
2573
                h->default_ref_list[0][index  ]= *h->long_ref[i];
2574
                h->default_ref_list[0][index++].pic_id= i;;
2575
            }
2576
            if(index < h->ref_count[0])
2577
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2578
        }
2579
    }else{ //FIELD
2580
        if(h->slice_type==B_TYPE){
2581
        }else{
2582
            //FIXME second field balh
2583
        }
2584
    }
2585
    return 0;
2586
}
2587

    
2588
static int decode_ref_pic_list_reordering(H264Context *h){
2589
    MpegEncContext * const s = &h->s;
2590
    int list;
2591
    
2592
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2593
    
2594
    for(list=0; list<2; list++){
2595
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2596

    
2597
        if(get_bits1(&s->gb)){
2598
            int pred= h->curr_pic_num;
2599
            int index;
2600

    
2601
            for(index=0; ; index++){
2602
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2603
                int pic_id;
2604
                int i;
2605
                
2606
                if(reordering_of_pic_nums_idc==3) 
2607
                    break;
2608
                
2609
                if(index >= h->ref_count[list]){
2610
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2611
                    return -1;
2612
                }
2613
                
2614
                if(reordering_of_pic_nums_idc<3){
2615
                    if(reordering_of_pic_nums_idc<2){
2616
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2617

    
2618
                        if(abs_diff_pic_num >= h->max_pic_num){
2619
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2620
                            return -1;
2621
                        }
2622

    
2623
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2624
                        else                                pred+= abs_diff_pic_num;
2625
                        pred &= h->max_pic_num - 1;
2626
                    
2627
                        for(i= h->ref_count[list]-1; i>=index; i--){
2628
                            if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2629
                                break;
2630
                        }
2631
                    }else{
2632
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2633

    
2634
                        for(i= h->ref_count[list]-1; i>=index; i--){
2635
                            if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2636
                                break;
2637
                        }
2638
                    }
2639

    
2640
                    if(i < index){
2641
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2642
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2643
                    }else if(i > index){
2644
                        Picture tmp= h->ref_list[list][i];
2645
                        for(; i>index; i--){
2646
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2647
                        }
2648
                        h->ref_list[list][index]= tmp;
2649
                    }
2650
                }else{
2651
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2652
                    return -1;
2653
                }
2654
            }
2655
        }
2656

    
2657
        if(h->slice_type!=B_TYPE) break;
2658
    }
2659
    return 0;    
2660
}
2661

    
2662
static int pred_weight_table(H264Context *h){
2663
    MpegEncContext * const s = &h->s;
2664
    int list, i;
2665
    
2666
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2667
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2668

    
2669
    for(list=0; list<2; list++){
2670
        for(i=0; i<h->ref_count[list]; i++){
2671
            int luma_weight_flag, chroma_weight_flag;
2672
            
2673
            luma_weight_flag= get_bits1(&s->gb);
2674
            if(luma_weight_flag){
2675
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
2676
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
2677
            }
2678

    
2679
            chroma_weight_flag= get_bits1(&s->gb);
2680
            if(chroma_weight_flag){
2681
                int j;
2682
                for(j=0; j<2; j++){
2683
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2684
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2685
                }
2686
            }
2687
        }
2688
        if(h->slice_type != B_TYPE) break;
2689
    }
2690
    return 0;
2691
}
2692

    
2693
/**
2694
 * instantaneos decoder refresh.
2695
 */
2696
static void idr(H264Context *h){
2697
    int i;
2698

    
2699
    for(i=0; i<h->long_ref_count; i++){
2700
        h->long_ref[i]->reference=0;
2701
        h->long_ref[i]= NULL;
2702
    }
2703
    h->long_ref_count=0;
2704

    
2705
    for(i=0; i<h->short_ref_count; i++){
2706
        h->short_ref[i]->reference=0;
2707
        h->short_ref[i]= NULL;
2708
    }
2709
    h->short_ref_count=0;
2710
}
2711

    
2712
/**
2713
 *
2714
 * @return the removed picture or NULL if an error occures
2715
 */
2716
static Picture * remove_short(H264Context *h, int frame_num){
2717
    MpegEncContext * const s = &h->s;
2718
    int i;
2719
    
2720
    if(s->avctx->debug&FF_DEBUG_MMCO)
2721
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
2722
    
2723
    for(i=0; i<h->short_ref_count; i++){
2724
        Picture *pic= h->short_ref[i];
2725
        if(s->avctx->debug&FF_DEBUG_MMCO)
2726
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
2727
        if(pic->frame_num == frame_num){
2728
            h->short_ref[i]= NULL;
2729
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2730
            h->short_ref_count--;
2731
            return pic;
2732
        }
2733
    }
2734
    return NULL;
2735
}
2736

    
2737
/**
2738
 *
2739
 * @return the removed picture or NULL if an error occures
2740
 */
2741
static Picture * remove_long(H264Context *h, int i){
2742
    Picture *pic;
2743

    
2744
    if(i >= h->long_ref_count) return NULL;
2745
    pic= h->long_ref[i];
2746
    if(pic==NULL) return NULL;
2747
    
2748
    h->long_ref[i]= NULL;
2749
    memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2750
    h->long_ref_count--;
2751

    
2752
    return pic;
2753
}
2754

    
2755
/**
2756
 * Executes the reference picture marking (memory management control operations).
2757
 */
2758
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2759
    MpegEncContext * const s = &h->s;
2760
    int i;
2761
    int current_is_long=0;
2762
    Picture *pic;
2763
    
2764
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2765
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
2766
        
2767
    for(i=0; i<mmco_count; i++){
2768
        if(s->avctx->debug&FF_DEBUG_MMCO)
2769
            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);
2770

    
2771
        switch(mmco[i].opcode){
2772
        case MMCO_SHORT2UNUSED:
2773
            pic= remove_short(h, mmco[i].short_frame_num);
2774
            if(pic==NULL) return -1;
2775
            pic->reference= 0;
2776
            break;
2777
        case MMCO_SHORT2LONG:
2778
            pic= remove_long(h, mmco[i].long_index);
2779
            if(pic) pic->reference=0;
2780
            
2781
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2782
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
2783
            break;
2784
        case MMCO_LONG2UNUSED:
2785
            pic= remove_long(h, mmco[i].long_index);
2786
            if(pic==NULL) return -1;
2787
            pic->reference= 0;
2788
            break;
2789
        case MMCO_LONG:
2790
            pic= remove_long(h, mmco[i].long_index);
2791
            if(pic) pic->reference=0;
2792
            
2793
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2794
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
2795
            h->long_ref_count++;
2796
            
2797
            current_is_long=1;
2798
            break;
2799
        case MMCO_SET_MAX_LONG:
2800
            assert(mmco[i].long_index <= 16);
2801
            while(mmco[i].long_index < h->long_ref_count){
2802
                pic= remove_long(h, mmco[i].long_index);
2803
                pic->reference=0;
2804
            }
2805
            while(mmco[i].long_index > h->long_ref_count){
2806
                h->long_ref[ h->long_ref_count++ ]= NULL;
2807
            }
2808
            break;
2809
        case MMCO_RESET:
2810
            while(h->short_ref_count){
2811
                pic= remove_short(h, h->short_ref[0]->frame_num);
2812
                pic->reference=0;
2813
            }
2814
            while(h->long_ref_count){
2815
                pic= remove_long(h, h->long_ref_count-1);
2816
                pic->reference=0;
2817
            }
2818
            break;
2819
        default: assert(0);
2820
        }
2821
    }
2822
    
2823
    if(!current_is_long){
2824
        pic= remove_short(h, s->current_picture_ptr->frame_num);
2825
        if(pic){
2826
            pic->reference=0;
2827
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
2828
        }
2829
        
2830
        if(h->short_ref_count)
2831
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2832

    
2833
        h->short_ref[0]= s->current_picture_ptr;
2834
        h->short_ref[0]->long_ref=0;
2835
        h->short_ref_count++;
2836
    }
2837
    
2838
    return 0; 
2839
}
2840

    
2841
static int decode_ref_pic_marking(H264Context *h){
2842
    MpegEncContext * const s = &h->s;
2843
    int i;
2844
    
2845
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2846
        s->broken_link= get_bits1(&s->gb) -1;
2847
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2848
        if(h->mmco[0].long_index == -1)
2849
            h->mmco_index= 0;
2850
        else{
2851
            h->mmco[0].opcode= MMCO_LONG;
2852
            h->mmco_index= 1;
2853
        } 
2854
    }else{
2855
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2856
            for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) { 
2857
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2858

    
2859
                h->mmco[i].opcode= opcode;
2860
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2861
                    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
2862
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2863
                        fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2864
                        return -1;
2865
                    }*/
2866
                }
2867
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2868
                    h->mmco[i].long_index= get_ue_golomb(&s->gb);
2869
                    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){
2870
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
2871
                        return -1;
2872
                    }
2873
                }
2874
                    
2875
                if(opcode > MMCO_LONG){
2876
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
2877
                    return -1;
2878
                }
2879
                if(opcode == MMCO_END)
2880
                    break;
2881
            }
2882
            h->mmco_index= i;
2883
        }else{
2884
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2885

    
2886
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2887
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2888
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2889
                h->mmco_index= 1;
2890
            }else
2891
                h->mmco_index= 0;
2892
        }
2893
    }
2894
    
2895
    return 0; 
2896
}
2897

    
2898
static int init_poc(H264Context *h){
2899
    MpegEncContext * const s = &h->s;
2900
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2901
    int field_poc[2];
2902

    
2903
    if(h->nal_unit_type == NAL_IDR_SLICE){
2904
        h->frame_num_offset= 0;
2905
    }else{
2906
        if(h->frame_num < h->prev_frame_num)
2907
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2908
        else
2909
            h->frame_num_offset= h->prev_frame_num_offset;
2910
    }
2911

    
2912
    if(h->sps.poc_type==0){
2913
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2914

    
2915
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2916
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2917
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2918
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2919
        else
2920
            h->poc_msb = h->prev_poc_msb;
2921
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2922
        field_poc[0] = 
2923
        field_poc[1] = h->poc_msb + h->poc_lsb;
2924
        if(s->picture_structure == PICT_FRAME) 
2925
            field_poc[1] += h->delta_poc_bottom;
2926
    }else if(h->sps.poc_type==1){
2927
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2928
        int i;
2929

    
2930
        if(h->sps.poc_cycle_length != 0)
2931
            abs_frame_num = h->frame_num_offset + h->frame_num;
2932
        else
2933
            abs_frame_num = 0;
2934

    
2935
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
2936
            abs_frame_num--;
2937
            
2938
        expected_delta_per_poc_cycle = 0;
2939
        for(i=0; i < h->sps.poc_cycle_length; i++)
2940
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2941

    
2942
        if(abs_frame_num > 0){
2943
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2944
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2945

    
2946
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2947
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
2948
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2949
        } else
2950
            expectedpoc = 0;
2951

    
2952
        if(h->nal_ref_idc == 0) 
2953
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2954
        
2955
        field_poc[0] = expectedpoc + h->delta_poc[0];
2956
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2957

    
2958
        if(s->picture_structure == PICT_FRAME)
2959
            field_poc[1] += h->delta_poc[1];
2960
    }else{
2961
        int poc;
2962
        if(h->nal_unit_type == NAL_IDR_SLICE){
2963
            poc= 0;
2964
        }else{
2965
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2966
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2967
        }
2968
        field_poc[0]= poc;
2969
        field_poc[1]= poc;
2970
    }
2971
    
2972
    if(s->picture_structure != PICT_BOTTOM_FIELD)
2973
        s->current_picture_ptr->field_poc[0]= field_poc[0];
2974
    if(s->picture_structure != PICT_TOP_FIELD)
2975
        s->current_picture_ptr->field_poc[1]= field_poc[1];
2976
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2977
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2978

    
2979
    return 0;
2980
}
2981

    
2982
/**
2983
 * decodes a slice header.
2984
 * this will allso call MPV_common_init() and frame_start() as needed
2985
 */
2986
static int decode_slice_header(H264Context *h){
2987
    MpegEncContext * const s = &h->s;
2988
    int first_mb_in_slice, pps_id;
2989
    int num_ref_idx_active_override_flag;
2990
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2991

    
2992
    s->current_picture.reference= h->nal_ref_idc != 0;
2993

    
2994
    first_mb_in_slice= get_ue_golomb(&s->gb);
2995

    
2996
    h->slice_type= get_ue_golomb(&s->gb);
2997
    if(h->slice_type > 9){
2998
        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);
2999
    }
3000
    if(h->slice_type > 4){
3001
        h->slice_type -= 5;
3002
        h->slice_type_fixed=1;
3003
    }else
3004
        h->slice_type_fixed=0;
3005
    
3006
    h->slice_type= slice_type_map[ h->slice_type ];
3007
    
3008
    s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
3009
        
3010
    pps_id= get_ue_golomb(&s->gb);
3011
    if(pps_id>255){
3012
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3013
        return -1;
3014
    }
3015
    h->pps= h->pps_buffer[pps_id];
3016
    if(h->pps.slice_group_count == 0){
3017
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3018
        return -1;
3019
    }
3020

    
3021
    h->sps= h->sps_buffer[ h->pps.sps_id ];
3022
    if(h->sps.log2_max_frame_num == 0){
3023
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3024
        return -1;
3025
    }
3026
    
3027
    s->mb_width= h->sps.mb_width;
3028
    s->mb_height= h->sps.mb_height;
3029
    
3030
    h->b_stride=  s->mb_width*4;
3031
    h->b8_stride= s->mb_width*2;
3032

    
3033
    s->mb_x = first_mb_in_slice % s->mb_width;
3034
    s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
3035
    
3036
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3037
    if(h->sps.frame_mbs_only_flag)
3038
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
3039
    else
3040
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
3041
    
3042
    if (s->context_initialized 
3043
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3044
        free_tables(h);
3045
        MPV_common_end(s);
3046
    }
3047
    if (!s->context_initialized) {
3048
        if (MPV_common_init(s) < 0)
3049
            return -1;
3050

    
3051
        alloc_tables(h);
3052

    
3053
        s->avctx->width = s->width;
3054
        s->avctx->height = s->height;
3055
        s->avctx->sample_aspect_ratio= h->sps.sar;
3056

    
3057
        if(h->sps.timing_info_present_flag && h->sps.fixed_frame_rate_flag){
3058
            s->avctx->frame_rate = h->sps.time_scale;
3059
            s->avctx->frame_rate_base = h->sps.num_units_in_tick;
3060
        }
3061
    }
3062

    
3063
    if(first_mb_in_slice == 0){
3064
        frame_start(h);
3065
    }
3066

    
3067
    s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3068
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3069

    
3070
    if(h->sps.frame_mbs_only_flag){
3071
        s->picture_structure= PICT_FRAME;
3072
    }else{
3073
        if(get_bits1(&s->gb)) //field_pic_flag
3074
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3075
        else
3076
            s->picture_structure= PICT_FRAME;
3077
    }
3078

    
3079
    if(s->picture_structure==PICT_FRAME){
3080
        h->curr_pic_num=   h->frame_num;
3081
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3082
    }else{
3083
        h->curr_pic_num= 2*h->frame_num;
3084
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3085
    }
3086
        
3087
    if(h->nal_unit_type == NAL_IDR_SLICE){
3088
        get_ue_golomb(&s->gb); /* idr_pic_id */
3089
    }
3090
   
3091
    if(h->sps.poc_type==0){
3092
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3093
        
3094
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3095
            h->delta_poc_bottom= get_se_golomb(&s->gb);
3096
        }
3097
    }
3098
    
3099
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3100
        h->delta_poc[0]= get_se_golomb(&s->gb);
3101
        
3102
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3103
            h->delta_poc[1]= get_se_golomb(&s->gb);
3104
    }
3105
    
3106
    init_poc(h);
3107
    
3108
    if(h->pps.redundant_pic_cnt_present){
3109
        h->redundant_pic_count= get_ue_golomb(&s->gb);
3110
    }
3111

    
3112
    //set defaults, might be overriden a few line later
3113
    h->ref_count[0]= h->pps.ref_count[0];
3114
    h->ref_count[1]= h->pps.ref_count[1];
3115

    
3116
    if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3117
        if(h->slice_type == B_TYPE){
3118
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
3119
        }
3120
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
3121
    
3122
        if(num_ref_idx_active_override_flag){
3123
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3124
            if(h->slice_type==B_TYPE)
3125
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3126

    
3127
            if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3128
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3129
                return -1;
3130
            }
3131
        }
3132
    }
3133

    
3134
    if(first_mb_in_slice == 0){
3135
        fill_default_ref_list(h);
3136
    }
3137

    
3138
    decode_ref_pic_list_reordering(h);
3139

    
3140
    if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) 
3141
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3142
        pred_weight_table(h);
3143
    
3144
    if(s->current_picture.reference)
3145
        decode_ref_pic_marking(h);
3146

    
3147
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3148
        h->cabac_init_idc = get_ue_golomb(&s->gb);
3149

    
3150
    h->last_qscale_diff = 0;
3151
    s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3152
    if(s->qscale<0 || s->qscale>51){
3153
        av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3154
        return -1;
3155
    }
3156
    //FIXME qscale / qp ... stuff
3157
    if(h->slice_type == SP_TYPE){
3158
        get_bits1(&s->gb); /* sp_for_switch_flag */
3159
    }
3160
    if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3161
        get_se_golomb(&s->gb); /* slice_qs_delta */
3162
    }
3163

    
3164
    h->deblocking_filter = 1;
3165
    h->slice_alpha_c0_offset = 0;
3166
    h->slice_beta_offset = 0;
3167
    if( h->pps.deblocking_filter_parameters_present ) {
3168
        h->deblocking_filter= get_ue_golomb(&s->gb);
3169
        if(h->deblocking_filter < 2) 
3170
            h->deblocking_filter^= 1; // 1<->0
3171

    
3172
        if( h->deblocking_filter ) {
3173
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3174
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3175
        }
3176
    }
3177

    
3178
#if 0 //FMO
3179
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3180
        slice_group_change_cycle= get_bits(&s->gb, ?);
3181
#endif
3182

    
3183
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3184
        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", 
3185
               first_mb_in_slice, 
3186
               av_get_pict_type_char(h->slice_type),
3187
               pps_id, h->frame_num,
3188
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3189
               h->ref_count[0], h->ref_count[1],
3190
               s->qscale,
3191
               h->deblocking_filter
3192
               );
3193
    }
3194

    
3195
    return 0;
3196
}
3197

    
3198
/**
3199
 *
3200
 */
3201
static inline int get_level_prefix(GetBitContext *gb){
3202
    unsigned int buf;
3203
    int log;
3204
    
3205
    OPEN_READER(re, gb);
3206
    UPDATE_CACHE(re, gb);
3207
    buf=GET_CACHE(re, gb);
3208
    
3209
    log= 32 - av_log2(buf);
3210
#ifdef TRACE
3211
    print_bin(buf>>(32-log), log);
3212
    av_log(NULL, AV_LOG_DEBUG, "%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3213
#endif
3214

    
3215
    LAST_SKIP_BITS(re, gb, log);
3216
    CLOSE_READER(re, gb);
3217

    
3218
    return log-1;
3219
}
3220

    
3221
/**
3222
 * decodes a residual block.
3223
 * @param n block index
3224
 * @param scantable scantable
3225
 * @param max_coeff number of coefficients in the block
3226
 * @return <0 if an error occured
3227
 */
3228
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3229
    MpegEncContext * const s = &h->s;
3230
    const uint16_t *qmul= dequant_coeff[qp];
3231
    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};
3232
    int level[16], run[16];
3233
    int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3234

    
3235
    //FIXME put trailing_onex into the context
3236

    
3237
    if(n == CHROMA_DC_BLOCK_INDEX){
3238
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3239
        total_coeff= coeff_token>>2;
3240
    }else{    
3241
        if(n == LUMA_DC_BLOCK_INDEX){
3242
            total_coeff= pred_non_zero_count(h, 0);
3243
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3244
            total_coeff= coeff_token>>2;
3245
        }else{
3246
            total_coeff= pred_non_zero_count(h, n);
3247
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3248
            total_coeff= coeff_token>>2;
3249
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3250
        }
3251
    }
3252

    
3253
    //FIXME set last_non_zero?
3254

    
3255
    if(total_coeff==0)
3256
        return 0;
3257
        
3258
    trailing_ones= coeff_token&3;
3259
    tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3260
    assert(total_coeff<=16);
3261
    
3262
    for(i=0; i<trailing_ones; i++){
3263
        level[i]= 1 - 2*get_bits1(gb);
3264
    }
3265

    
3266
    suffix_length= total_coeff > 10 && trailing_ones < 3;
3267

    
3268
    for(; i<total_coeff; i++){
3269
        const int prefix= get_level_prefix(gb);
3270
        int level_code, mask;
3271

    
3272
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3273
            if(suffix_length)
3274
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3275
            else
3276
                level_code= (prefix<<suffix_length); //part
3277
        }else if(prefix==14){
3278
            if(suffix_length)
3279
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3280
            else
3281
                level_code= prefix + get_bits(gb, 4); //part
3282
        }else if(prefix==15){
3283
            level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3284
            if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3285
        }else{
3286
            av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3287
            return -1;
3288
        }
3289

    
3290
        if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3291

    
3292
        mask= -(level_code&1);
3293
        level[i]= (((2+level_code)>>1) ^ mask) - mask;
3294

    
3295
        if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3296

    
3297
#if 1
3298
        if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3299
#else        
3300
        if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3301
        /* ? == prefix > 2 or sth */
3302
#endif
3303
        tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3304
    }
3305

    
3306
    if(total_coeff == max_coeff)
3307
        zeros_left=0;
3308
    else{
3309
        if(n == CHROMA_DC_BLOCK_INDEX)
3310
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3311
        else
3312
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3313
    }
3314
    
3315
    for(i=0; i<total_coeff-1; i++){
3316
        if(zeros_left <=0)
3317
            break;
3318
        else if(zeros_left < 7){
3319
            run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3320
        }else{
3321
            run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3322
        }
3323
        zeros_left -= run[i];
3324
    }
3325

    
3326
    if(zeros_left<0){
3327
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3328
        return -1;
3329
    }
3330
    
3331
    for(; i<total_coeff-1; i++){
3332
        run[i]= 0;
3333
    }
3334

    
3335
    run[i]= zeros_left;
3336

    
3337
    coeff_num=-1;
3338
    if(n > 24){
3339
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3340
            int j;
3341

    
3342
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3343
            j= scantable[ coeff_num ];
3344

    
3345
            block[j]= level[i];
3346
        }
3347
    }else{
3348
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into  rundecode?
3349
            int j;
3350

    
3351
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3352
            j= scantable[ coeff_num ];
3353

    
3354
            block[j]= level[i] * qmul[j];
3355
//            printf("%d %d  ", block[j], qmul[j]);
3356
        }
3357
    }
3358
    return 0;
3359
}
3360

    
3361
/**
3362
 * decodes a macroblock
3363
 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3364
 */
3365
static int decode_mb_cavlc(H264Context *h){
3366
    MpegEncContext * const s = &h->s;
3367
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3368
    int mb_type, partition_count, cbp;
3369

    
3370
    s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?    
3371

    
3372
    tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3373
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3374
                down the code */
3375
    if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3376
        if(s->mb_skip_run==-1)
3377
            s->mb_skip_run= get_ue_golomb(&s->gb);
3378
        
3379
        if (s->mb_skip_run--) {
3380
            int mx, my;
3381
            /* skip mb */
3382
//FIXME b frame
3383
            mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3384

    
3385
            memset(h->non_zero_count[mb_xy], 0, 16);
3386
            memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3387

    
3388
            if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3389
                h->mb_field_decoding_flag= get_bits1(&s->gb);
3390
            }
3391

    
3392
            if(h->mb_field_decoding_flag)
3393
                mb_type|= MB_TYPE_INTERLACED;
3394
            
3395
            fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3396
            pred_pskip_motion(h, &mx, &my);
3397
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3398
            fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3399
            write_back_motion(h, mb_type);
3400

    
3401
            s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3402
            s->current_picture.qscale_table[mb_xy]= s->qscale;
3403
            h->slice_table[ mb_xy ]= h->slice_num;
3404

    
3405
            h->prev_mb_skiped= 1;
3406
            return 0;
3407
        }
3408
    }
3409
    if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3410
        if((s->mb_y&1)==0)
3411
            h->mb_field_decoding_flag = get_bits1(&s->gb);
3412
    }else
3413
        h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3414
    
3415
    h->prev_mb_skiped= 0;
3416
    
3417
    mb_type= get_ue_golomb(&s->gb);
3418
    if(h->slice_type == B_TYPE){
3419
        if(mb_type < 23){
3420
            partition_count= b_mb_type_info[mb_type].partition_count;
3421
            mb_type=         b_mb_type_info[mb_type].type;
3422
        }else{
3423
            mb_type -= 23;
3424
            goto decode_intra_mb;
3425
        }
3426
    }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3427
        if(mb_type < 5){
3428
            partition_count= p_mb_type_info[mb_type].partition_count;
3429
            mb_type=         p_mb_type_info[mb_type].type;
3430
        }else{
3431
            mb_type -= 5;
3432
            goto decode_intra_mb;
3433
        }
3434
    }else{
3435
       assert(h->slice_type == I_TYPE);
3436
decode_intra_mb:
3437
        if(mb_type > 25){
3438
            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);
3439
            return -1;
3440
        }
3441
        partition_count=0;
3442
        cbp= i_mb_type_info[mb_type].cbp;
3443
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3444
        mb_type= i_mb_type_info[mb_type].type;
3445
    }
3446

    
3447
    if(h->mb_field_decoding_flag)
3448
        mb_type |= MB_TYPE_INTERLACED;
3449

    
3450
    s->current_picture.mb_type[mb_xy]= mb_type;
3451
    h->slice_table[ mb_xy ]= h->slice_num;
3452
    
3453
    if(IS_INTRA_PCM(mb_type)){
3454
        const uint8_t *ptr;
3455
        int x, y;
3456
        
3457
        // we assume these blocks are very rare so we dont optimize it
3458
        align_get_bits(&s->gb);
3459
        
3460
        ptr= s->gb.buffer + get_bits_count(&s->gb);
3461
    
3462
        for(y=0; y<16; y++){
3463
            const int index= 4*(y&3) + 64*(y>>2);
3464
            for(x=0; x<16; x++){
3465
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3466
            }
3467
        }
3468
        for(y=0; y<8; y++){
3469
            const int index= 256 + 4*(y&3) + 32*(y>>2);
3470
            for(x=0; x<8; x++){
3471
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3472
            }
3473
        }
3474
        for(y=0; y<8; y++){
3475
            const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3476
            for(x=0; x<8; x++){
3477
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3478
            }
3479
        }
3480
    
3481
        skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3482
        
3483
        //FIXME deblock filter, non_zero_count_cache init ...
3484
        memset(h->non_zero_count[mb_xy], 16, 16);
3485
        s->current_picture.qscale_table[mb_xy]= s->qscale;
3486
        
3487
        return 0;
3488
    }
3489
        
3490
    fill_caches(h, mb_type);
3491

    
3492
    //mb_pred
3493
    if(IS_INTRA(mb_type)){
3494
//            init_top_left_availability(h);
3495
            if(IS_INTRA4x4(mb_type)){
3496
                int i;
3497

    
3498
//                fill_intra4x4_pred_table(h);
3499
                for(i=0; i<16; i++){
3500
                    const int mode_coded= !get_bits1(&s->gb);
3501
                    const int predicted_mode=  pred_intra_mode(h, i);
3502
                    int mode;
3503

    
3504
                    if(mode_coded){
3505
                        const int rem_mode= get_bits(&s->gb, 3);
3506
                        if(rem_mode<predicted_mode)
3507
                            mode= rem_mode;
3508
                        else
3509
                            mode= rem_mode + 1;
3510
                    }else{
3511
                        mode= predicted_mode;
3512
                    }
3513
                    
3514
                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3515
                }
3516
                write_back_intra_pred_mode(h);
3517
                if( check_intra4x4_pred_mode(h) < 0)
3518
                    return -1;
3519
            }else{
3520
                h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3521
                if(h->intra16x16_pred_mode < 0)
3522
                    return -1;
3523
            }
3524
            h->chroma_pred_mode= get_ue_golomb(&s->gb);
3525

    
3526
            h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3527
            if(h->chroma_pred_mode < 0)
3528
                return -1;
3529
    }else if(partition_count==4){
3530
        int i, j, sub_partition_count[4], list, ref[2][4];
3531
        
3532
        if(h->slice_type == B_TYPE){
3533
            for(i=0; i<4; i++){
3534
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3535
                if(h->sub_mb_type[i] >=13){
3536
                    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);
3537
                    return -1;
3538
                }
3539
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3540
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3541
            }
3542
        }else{
3543
            assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3544
            for(i=0; i<4; i++){
3545
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3546
                if(h->sub_mb_type[i] >=4){
3547
                    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);
3548
                    return -1;
3549
                }
3550
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3551
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3552
            }
3553
        }
3554
        
3555
        for(list=0; list<2; list++){
3556
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3557
            if(ref_count == 0) continue;
3558
            for(i=0; i<4; i++){
3559
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3560
                    ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3561
                }else{
3562
                 //FIXME
3563
                    ref[list][i] = -1;
3564
                }
3565
            }
3566
        }
3567
        
3568
        for(list=0; list<2; list++){
3569
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3570
            if(ref_count == 0) continue;
3571

    
3572
            for(i=0; i<4; i++){
3573
                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3574
                h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3575

    
3576
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3577
                    const int sub_mb_type= h->sub_mb_type[i];
3578
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3579
                    for(j=0; j<sub_partition_count[i]; j++){
3580
                        int mx, my;
3581
                        const int index= 4*i + block_width*j;
3582
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3583
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3584
                        mx += get_se_golomb(&s->gb);
3585
                        my += get_se_golomb(&s->gb);
3586
                        tprintf("final mv:%d %d\n", mx, my);
3587

    
3588
                        if(IS_SUB_8X8(sub_mb_type)){
3589
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= 
3590
                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3591
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= 
3592
                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3593
                        }else if(IS_SUB_8X4(sub_mb_type)){
3594
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3595
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3596
                        }else if(IS_SUB_4X8(sub_mb_type)){
3597
                            mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3598
                            mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3599
                        }else{
3600
                            assert(IS_SUB_4X4(sub_mb_type));
3601
                            mv_cache[ 0 ][0]= mx;
3602
                            mv_cache[ 0 ][1]= my;
3603
                        }
3604
                    }
3605
                }else{
3606
                    uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3607
                    p[0] = p[1]=
3608
                    p[8] = p[9]= 0;
3609
                }
3610
            }
3611
        }
3612
    }else if(!IS_DIRECT(mb_type)){
3613
        int list, mx, my, i;
3614
         //FIXME we should set ref_idx_l? to 0 if we use that later ...
3615
        if(IS_16X16(mb_type)){
3616
            for(list=0; list<2; list++){
3617
                if(h->ref_count[0]>0){
3618
                    if(IS_DIR(mb_type, 0, list)){
3619
                        const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3620
                        fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3621
                    }
3622
                }
3623
            }
3624
            for(list=0; list<2; list++){
3625
                if(IS_DIR(mb_type, 0, list)){
3626
                    pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3627
                    mx += get_se_golomb(&s->gb);
3628
                    my += get_se_golomb(&s->gb);
3629
                    tprintf("final mv:%d %d\n", mx, my);
3630

    
3631
                    fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
3632
                }
3633
            }
3634
        }
3635
        else if(IS_16X8(mb_type)){
3636
            for(list=0; list<2; list++){
3637
                if(h->ref_count[list]>0){
3638
                    for(i=0; i<2; i++){
3639
                        if(IS_DIR(mb_type, i, list)){
3640
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3641
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3642
                        }
3643
                    }
3644
                }
3645
            }
3646
            for(list=0; list<2; list++){
3647
                for(i=0; i<2; i++){
3648
                    if(IS_DIR(mb_type, i, list)){
3649
                        pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3650
                        mx += get_se_golomb(&s->gb);
3651
                        my += get_se_golomb(&s->gb);
3652
                        tprintf("final mv:%d %d\n", mx, my);
3653

    
3654
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
3655
                    }
3656
                }
3657
            }
3658
        }else{
3659
            assert(IS_8X16(mb_type));
3660
            for(list=0; list<2; list++){
3661
                if(h->ref_count[list]>0){
3662
                    for(i=0; i<2; i++){
3663
                        if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3664
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3665
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3666
                        }
3667
                    }
3668
                }
3669
            }
3670
            for(list=0; list<2; list++){
3671
                for(i=0; i<2; i++){
3672
                    if(IS_DIR(mb_type, i, list)){
3673
                        pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3674
                        mx += get_se_golomb(&s->gb);
3675
                        my += get_se_golomb(&s->gb);
3676
                        tprintf("final mv:%d %d\n", mx, my);
3677

    
3678
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
3679
                    }
3680
                }
3681
            }
3682
        }
3683
    }
3684
    
3685
    if(IS_INTER(mb_type))
3686
        write_back_motion(h, mb_type);
3687
    
3688
    if(!IS_INTRA16x16(mb_type)){
3689
        cbp= get_ue_golomb(&s->gb);
3690
        if(cbp > 47){
3691
            av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3692
            return -1;
3693
        }
3694
        
3695
        if(IS_INTRA4x4(mb_type))
3696
            cbp= golomb_to_intra4x4_cbp[cbp];
3697
        else
3698
            cbp= golomb_to_inter_cbp[cbp];
3699
    }
3700

    
3701
    if(cbp || IS_INTRA16x16(mb_type)){
3702
        int i8x8, i4x4, chroma_idx;
3703
        int chroma_qp, dquant;
3704
        GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3705
        const uint8_t *scan, *dc_scan;
3706
        
3707
//        fill_non_zero_count_cache(h);
3708

    
3709
        if(IS_INTERLACED(mb_type)){
3710
            scan= field_scan;
3711
            dc_scan= luma_dc_field_scan;
3712
        }else{
3713
            scan= zigzag_scan;
3714
            dc_scan= luma_dc_zigzag_scan;
3715
        }
3716

    
3717
        dquant= get_se_golomb(&s->gb);
3718

    
3719
        if( dquant > 25 || dquant < -26 ){
3720
            av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3721
            return -1;
3722
        }
3723
        
3724
        s->qscale += dquant;
3725
        if(((unsigned)s->qscale) > 51){
3726
            if(s->qscale<0) s->qscale+= 52;
3727
            else            s->qscale-= 52;
3728
        }
3729
        
3730
        h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3731
        if(IS_INTRA16x16(mb_type)){
3732
            if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3733
                return -1; //FIXME continue if partotioned and other retirn -1 too
3734
            }
3735

    
3736
            assert((cbp&15) == 0 || (cbp&15) == 15);
3737

    
3738
            if(cbp&15){
3739
                for(i8x8=0; i8x8<4; i8x8++){
3740
                    for(i4x4=0; i4x4<4; i4x4++){
3741
                        const int index= i4x4 + 4*i8x8;
3742
                        if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3743
                            return -1;
3744
                        }
3745
                    }
3746
                }
3747
            }else{
3748
                fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
3749
            }
3750
        }else{
3751
            for(i8x8=0; i8x8<4; i8x8++){
3752
                if(cbp & (1<<i8x8)){
3753
                    for(i4x4=0; i4x4<4; i4x4++){
3754
                        const int index= i4x4 + 4*i8x8;
3755
                        
3756
                        if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3757
                            return -1;
3758
                        }
3759
                    }
3760
                }else{
3761
                    uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3762
                    nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3763
                }
3764
            }
3765
        }
3766
        
3767
        if(cbp&0x30){
3768
            for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3769
                if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3770
                    return -1;
3771
                }
3772
        }
3773

    
3774
        if(cbp&0x20){
3775
            for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3776
                for(i4x4=0; i4x4<4; i4x4++){
3777
                    const int index= 16 + 4*chroma_idx + i4x4;
3778
                    if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3779
                        return -1;
3780
                    }
3781
                }
3782
            }
3783
        }else{
3784
            uint8_t * const nnz= &h->non_zero_count_cache[0];
3785
            nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3786
            nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3787
        }
3788
    }else{
3789
        uint8_t * const nnz= &h->non_zero_count_cache[0];
3790
        fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
3791
        nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3792
        nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3793
    }
3794
    s->current_picture.qscale_table[mb_xy]= s->qscale;
3795
    write_back_non_zero_count(h);
3796

    
3797
    return 0;
3798
}
3799

    
3800
static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {
3801
    uint8_t *state= &h->cabac_state[ctx_base];
3802
    int mb_type;
3803
    
3804
    if(intra_slice){
3805
        MpegEncContext * const s = &h->s;
3806
        const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3807
        int ctx=0;
3808
        if( s->mb_x > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-1] ) )
3809
            ctx++;
3810
        if( s->mb_y > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
3811
            ctx++;
3812
        if( get_cabac( &h->cabac, &state[ctx] ) == 0 )
3813
            return 0;   /* I4x4 */
3814
        state += 2;
3815
    }else{
3816
        if( get_cabac( &h->cabac, &state[0] ) == 0 )
3817
            return 0;   /* I4x4 */
3818
    }
3819

    
3820
    if( get_cabac_terminate( &h->cabac ) )
3821
        return 25;  /* PCM */
3822

    
3823
    mb_type = 1; /* I16x16 */
3824
    if( get_cabac( &h->cabac, &state[1] ) )
3825
        mb_type += 12;  /* cbp_luma != 0 */
3826

    
3827
    if( get_cabac( &h->cabac, &state[2] ) ) {
3828
        if( get_cabac( &h->cabac, &state[2+intra_slice] ) )
3829
            mb_type += 4 * 2;   /* cbp_chroma == 2 */
3830
        else
3831
            mb_type += 4 * 1;   /* cbp_chroma == 1 */
3832
    }
3833
    if( get_cabac( &h->cabac, &state[3+intra_slice] ) )
3834
        mb_type += 2;
3835
    if( get_cabac( &h->cabac, &state[3+2*intra_slice] ) )
3836
        mb_type += 1;
3837
    return mb_type;
3838
}
3839

    
3840
static int decode_cabac_mb_type( H264Context *h ) {
3841
    MpegEncContext * const s = &h->s;
3842

    
3843
    if( h->slice_type == I_TYPE ) {
3844
        return decode_cabac_intra_mb_type(h, 3, 1);
3845
    } else if( h->slice_type == P_TYPE ) {
3846
        if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
3847
            /* P-type */
3848
            if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
3849
                if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
3850
                    return 0; /* P_L0_D16x16; */
3851
                else
3852
                    return 3; /* P_8x8; */
3853
            } else {
3854
                if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
3855
                    return 2; /* P_L0_D8x16; */
3856
                else
3857
                    return 1; /* P_L0_D16x8; */
3858
            }
3859
        } else {
3860
            return decode_cabac_intra_mb_type(h, 17, 0) + 5;
3861
        }
3862
    } else if( h->slice_type == B_TYPE ) {
3863
        const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3864
        int ctx = 0;
3865
        int bits;
3866

    
3867
        if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mb_xy-1] )
3868
                      && !IS_DIRECT( s->current_picture.mb_type[mb_xy-1] ) )
3869
            ctx++;
3870
        if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mb_xy-s->mb_stride] )
3871
                      && !IS_DIRECT( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
3872
            ctx++;
3873

    
3874
        if( !get_cabac( &h->cabac, &h->cabac_state[27+ctx] ) )
3875
            return 0; /* B_Direct_16x16 */
3876

    
3877
        if( !get_cabac( &h->cabac, &h->cabac_state[27+3] ) ) {
3878
            return 1 + get_cabac( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */
3879
        }
3880

    
3881
        bits = get_cabac( &h->cabac, &h->cabac_state[27+4] ) << 3;
3882
        bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 2;
3883
        bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 1;
3884
        bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] );
3885
        if( bits < 8 )
3886
            return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
3887
        else if( bits == 13 ) {
3888
            return decode_cabac_intra_mb_type(h, 32, 0) + 23;
3889
        } else if( bits == 14 )
3890
            return 11; /* B_L1_L0_8x16 */
3891
        else if( bits == 15 )
3892
            return 22; /* B_8x8 */
3893

    
3894
        bits= ( bits<<1 ) | get_cabac( &h->cabac, &h->cabac_state[27+5] );
3895
        return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */
3896
    } else {
3897
        /* TODO SI/SP frames? */
3898
        return -1;
3899
    }
3900
}
3901

    
3902
static int decode_cabac_mb_skip( H264Context *h) {
3903
    MpegEncContext * const s = &h->s;
3904
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3905
    const int mba_xy = mb_xy - 1;
3906
    const int mbb_xy = mb_xy - s->mb_stride;
3907
    int ctx = 0;
3908

    
3909
    if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mba_xy] ) )
3910
        ctx++;
3911
    if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ) )
3912
        ctx++;
3913

    
3914
    if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
3915
        return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
3916
    else /* B-frame */
3917
        return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
3918
}
3919

    
3920
static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
3921
    int mode = 0;
3922

    
3923
    if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
3924
        return pred_mode;
3925

    
3926
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3927
        mode += 1;
3928
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3929
        mode += 2;
3930
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3931
        mode += 4;
3932
    if( mode >= pred_mode )
3933
        return mode + 1;
3934
    else
3935
        return mode;
3936
}
3937

    
3938
static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
3939
    MpegEncContext * const s = &h->s;
3940
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3941
    const int mba_xy = mb_xy - 1;
3942
    const int mbb_xy = mb_xy - s->mb_stride;
3943

    
3944
    int ctx = 0;
3945

    
3946
    /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
3947
    if( s->mb_x > 0 && h->chroma_pred_mode_table[mba_xy] != 0 )
3948
        ctx++;
3949

    
3950
    if( s->mb_y > 0 && h->chroma_pred_mode_table[mbb_xy] != 0 )
3951
        ctx++;
3952

    
3953
    if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
3954
        return 0;
3955

    
3956
    if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
3957
        return 1;
3958
    if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
3959
        return 2;
3960
    else
3961
        return 3;
3962
}
3963

    
3964
static const uint8_t block_idx_x[16] = {
3965
    0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
3966
};
3967
static const uint8_t block_idx_y[16] = {
3968
    0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
3969
};
3970
static const uint8_t block_idx_xy[4][4] = {
3971
    { 0, 2, 8,  10},
3972
    { 1, 3, 9,  11},
3973
    { 4, 6, 12, 14},
3974
    { 5, 7, 13, 15}
3975
};
3976

    
3977
static int decode_cabac_mb_cbp_luma( H264Context *h) {
3978
    MpegEncContext * const s = &h->s;
3979
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3980

    
3981
    int cbp = 0;
3982
    int i8x8;
3983

    
3984
    h->cbp_table[mb_xy] = 0;  /* FIXME aaahahahah beurk */
3985

    
3986
    for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
3987
        int mba_xy = -1;
3988
        int mbb_xy = -1;
3989
        int x, y;
3990
        int ctx = 0;
3991

    
3992
        x = block_idx_x[4*i8x8];
3993
        y = block_idx_y[4*i8x8];
3994

    
3995
        if( x > 0 )
3996
            mba_xy = mb_xy;
3997
        else if( s->mb_x > 0 )
3998
            mba_xy = mb_xy - 1;
3999

    
4000
        if( y > 0 )
4001
            mbb_xy = mb_xy;
4002
        else if( s->mb_y > 0 )
4003
            mbb_xy = mb_xy - s->mb_stride;
4004

    
4005
        /* No need to test for skip as we put 0 for skip block */
4006
        if( mba_xy >= 0 ) {
4007
            int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4008
            if( ((h->cbp_table[mba_xy] >> i8x8a)&0x01) == 0 )
4009
                ctx++;
4010
        }
4011

    
4012
        if( mbb_xy >= 0 ) {
4013
            int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4014
            if( ((h->cbp_table[mbb_xy] >> i8x8b)&0x01) == 0 )
4015
                ctx += 2;
4016
        }
4017

    
4018
        if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
4019
            cbp |= 1 << i8x8;
4020
            h->cbp_table[mb_xy] = cbp;  /* FIXME aaahahahah beurk */
4021
        }
4022
    }
4023
    return cbp;
4024
}
4025
static int decode_cabac_mb_cbp_chroma( H264Context *h) {
4026
    int ctx;
4027
    int cbp_a, cbp_b;
4028

    
4029
    cbp_a = (h->left_cbp>>4)&0x03;
4030
    cbp_b = (h-> top_cbp>>4)&0x03;
4031

    
4032
    ctx = 0;
4033
    if( cbp_a > 0 ) ctx++;
4034
    if( cbp_b > 0 ) ctx += 2;
4035
    if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
4036
        return 0;
4037

    
4038
    ctx = 4;
4039
    if( cbp_a == 2 ) ctx++;
4040
    if( cbp_b == 2 ) ctx += 2;
4041
    return 1 + get_cabac( &h->cabac, &h->cabac_state[77 + ctx] );
4042
}
4043
static int decode_cabac_mb_dqp( H264Context *h) {
4044
    MpegEncContext * const s = &h->s;
4045
    int mbn_xy;
4046
    int   ctx = 0;
4047
    int   val = 0;
4048

    
4049
    if( s->mb_x > 0 )
4050
        mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;
4051
    else
4052
        mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;
4053

    
4054
    if( mbn_xy >= 0 && h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
4055
        ctx++;
4056

    
4057
    while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
4058
        if( ctx < 2 )
4059
            ctx = 2;
4060
        else
4061
            ctx = 3;
4062
        val++;
4063
    }
4064

    
4065
    if( val&0x01 )
4066
        return (val + 1)/2;
4067
    else
4068
        return -(val + 1)/2;
4069
}
4070
static int decode_cabac_p_mb_sub_type( H264Context *h ) {
4071
    if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
4072
        return 0;   /* 8x8 */
4073
    if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
4074
        return 1;   /* 8x4 */
4075
    if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
4076
        return 2;   /* 4x8 */
4077
    return 3;       /* 4x4 */
4078
}
4079
static int decode_cabac_b_mb_sub_type( H264Context *h ) {
4080
    int type;
4081
    if( !get_cabac( &h->cabac, &h->cabac_state[36] ) )
4082
        return 0;   /* B_Direct_8x8 */
4083
    if( !get_cabac( &h->cabac, &h->cabac_state[37] ) )
4084
        return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */
4085
    type = 3;
4086
    if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) {
4087
        if( get_cabac( &h->cabac, &h->cabac_state[39] ) )
4088
            return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */
4089
        type += 4;
4090
    }
4091
    type += 2*get_cabac( &h->cabac, &h->cabac_state[39] );
4092
    type +=   get_cabac( &h->cabac, &h->cabac_state[39] );
4093
    return type;
4094
}
4095

    
4096
static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
4097
    int refa = h->ref_cache[list][scan8[n] - 1];
4098
    int refb = h->ref_cache[list][scan8[n] - 8];
4099
    int ref  = 0;
4100
    int ctx  = 0;
4101

    
4102
    if( refa > 0 )
4103
        ctx++;
4104
    if( refb > 0 )
4105
        ctx += 2;
4106

    
4107
    while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
4108
        ref++;
4109
        if( ctx < 4 )
4110
            ctx = 4;
4111
        else
4112
            ctx = 5;
4113
    }
4114
    return ref;
4115
}
4116

    
4117
static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
4118
    int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
4119
               abs( h->mvd_cache[list][scan8[n] - 8][l] );
4120
    int ctxbase = (l == 0) ? 40 : 47;
4121
    int ctx, mvd;
4122

    
4123
    if( amvd < 3 )
4124
        ctx = 0;
4125
    else if( amvd > 32 )
4126
        ctx = 2;
4127
    else
4128
        ctx = 1;
4129

    
4130
    if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+ctx]))
4131
        return 0;
4132

    
4133
    mvd= 1;
4134
    ctx= 3;
4135
    while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
4136
        mvd++;
4137
        if( ctx < 6 )
4138
            ctx++;
4139
    }
4140

    
4141
    if( mvd >= 9 ) {
4142
        int k = 3;
4143
        while( get_cabac_bypass( &h->cabac ) ) {
4144
            mvd += 1 << k;
4145
            k++;
4146
        }
4147
        while( k-- ) {
4148
            if( get_cabac_bypass( &h->cabac ) )
4149
                mvd += 1 << k;
4150
        }
4151
    }
4152
    if( get_cabac_bypass( &h->cabac ) )  return -mvd;
4153
    else                                 return  mvd;
4154
}
4155

    
4156
static int inline get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {
4157
    int nza, nzb;
4158
    int ctx = 0;
4159

    
4160
    if( cat == 0 ) {
4161
        nza = h->left_cbp&0x100;
4162
        nzb = h-> top_cbp&0x100;
4163
    } else if( cat == 1 || cat == 2 ) {
4164
        nza = h->non_zero_count_cache[scan8[idx] - 1];
4165
        nzb = h->non_zero_count_cache[scan8[idx] - 8];
4166
    } else if( cat == 3 ) {
4167
        nza = (h->left_cbp>>(6+idx))&0x01;
4168
        nzb = (h-> top_cbp>>(6+idx))&0x01;
4169
    } else {
4170
        assert(cat == 4);
4171
        nza = h->non_zero_count_cache[scan8[16+idx] - 1];
4172
        nzb = h->non_zero_count_cache[scan8[16+idx] - 8];
4173
    }
4174

    
4175
    if( nza > 0 )
4176
        ctx++;
4177

    
4178
    if( nzb > 0 )
4179
        ctx += 2;
4180

    
4181
    return ctx + 4 * cat;
4182
}
4183

    
4184
static int inline decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, int qp, int max_coeff) {
4185
    const int mb_xy  = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
4186
    const uint16_t *qmul= dequant_coeff[qp];
4187
    static const int significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
4188
    static const int coeff_abs_level_m1_offset[5] = {227+ 0, 227+10, 227+20, 227+30, 227+39 };
4189

    
4190
    int index[16];
4191

    
4192
    int i, last;
4193
    int coeff_count = 0;
4194

    
4195
    int abslevel1 = 1;
4196
    int abslevelgt1 = 0;
4197

    
4198
    /* cat: 0-> DC 16x16  n = 0
4199
     *      1-> AC 16x16  n = luma4x4idx
4200
     *      2-> Luma4x4   n = luma4x4idx
4201
     *      3-> DC Chroma n = iCbCr
4202
     *      4-> AC Chroma n = 4 * iCbCr + chroma4x4idx
4203
     */
4204

    
4205
    /* read coded block flag */
4206
    if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {
4207
        if( cat == 1 || cat == 2 )
4208
            h->non_zero_count_cache[scan8[n]] = 0;
4209
        else if( cat == 4 )
4210
            h->non_zero_count_cache[scan8[16+n]] = 0;
4211

    
4212
        return 0;
4213
    }
4214

    
4215
    for(last= 0; last < max_coeff - 1; last++) {
4216
        if( get_cabac( &h->cabac, &h->cabac_state[105+significant_coeff_flag_offset[cat]+last] )) {
4217
            index[coeff_count++] = last;
4218
            if( get_cabac( &h->cabac, &h->cabac_state[166+significant_coeff_flag_offset[cat]+last] ) ) {
4219
                last= max_coeff;
4220
                break;
4221
            }
4222
        }
4223
    }
4224
    if( last == max_coeff -1 ) {
4225
        index[coeff_count++] = last;
4226
    }
4227
    assert(coeff_count > 0);
4228

    
4229
    if( cat == 0 )
4230
        h->cbp_table[mb_xy] |= 0x100;
4231
    else if( cat == 1 || cat == 2 )
4232
        h->non_zero_count_cache[scan8[n]] = coeff_count;
4233
    else if( cat == 3 )
4234
        h->cbp_table[mb_xy] |= 0x40 << n;
4235
    else {
4236
        assert( cat == 4 );
4237
        h->non_zero_count_cache[scan8[16+n]] = coeff_count;
4238
    }
4239

    
4240
    for( i = coeff_count - 1; i >= 0; i-- ) {
4241
        int ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 )) + coeff_abs_level_m1_offset[cat];
4242
        int j= scantable[index[i]];
4243

    
4244
        if( get_cabac( &h->cabac, &h->cabac_state[ctx] ) == 0 ) {
4245
            if( cat == 0 || cat == 3 ) {
4246
                if( get_cabac_bypass( &h->cabac ) ) block[j] = -1;
4247
                else                                block[j] =  1;
4248
            }else{
4249
                if( get_cabac_bypass( &h->cabac ) ) block[j] = -qmul[j];
4250
                else                                block[j] =  qmul[j];
4251
            }
4252
    
4253
            abslevel1++;
4254
        } else {
4255
            int coeff_abs = 2;
4256
            ctx = 5 + FFMIN( 4, abslevelgt1 ) + coeff_abs_level_m1_offset[cat];
4257
            while( coeff_abs < 15 && get_cabac( &h->cabac, &h->cabac_state[ctx] ) ) {
4258
                coeff_abs++;
4259
            }
4260

    
4261
            if( coeff_abs >= 15 ) {
4262
                int j = 0;
4263
                while( get_cabac_bypass( &h->cabac ) ) {
4264
                    coeff_abs += 1 << j;
4265
                    j++;
4266
                }
4267
    
4268
                while( j-- ) {
4269
                    if( get_cabac_bypass( &h->cabac ) )
4270
                        coeff_abs += 1 << j ;
4271
                }
4272
            }
4273

    
4274
            if( cat == 0 || cat == 3 ) {
4275
                if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs;
4276
                else                                block[j] =  coeff_abs;
4277
            }else{
4278
                if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs * qmul[j];
4279
                else                                block[j] =  coeff_abs * qmul[j];
4280
            }
4281
    
4282
            abslevelgt1++;
4283
        }
4284
    }
4285
    return 0;
4286
}
4287

    
4288
/**
4289
 * decodes a macroblock
4290
 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4291
 */
4292
static int decode_mb_cabac(H264Context *h) {
4293
    MpegEncContext * const s = &h->s;
4294
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4295
    int mb_type, partition_count, cbp = 0;
4296

    
4297
    s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?)
4298

    
4299
    if( h->sps.mb_aff ) {
4300
        av_log( h->s.avctx, AV_LOG_ERROR, "Fields not supported with CABAC\n" );
4301
        return -1;
4302
    }
4303

    
4304
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
4305
        /* read skip flags */
4306
        if( decode_cabac_mb_skip( h ) ) {
4307
            int mx, my;
4308

    
4309
//FIXME b frame
4310
            /* skip mb */
4311
            mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4312

    
4313
            memset(h->non_zero_count[mb_xy], 0, 16);
4314
            memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4315
#if 0
4316
            if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
4317
                h->mb_field_decoding_flag= get_bits1(&s->gb);
4318
            }
4319
            if(h->mb_field_decoding_flag)
4320
                mb_type|= MB_TYPE_INTERLACED;
4321
#endif
4322

    
4323
            fill_caches(h, mb_type); //FIXME check what is needed and what not ...
4324
            pred_pskip_motion(h, &mx, &my);
4325
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4326
            fill_rectangle(  h->mvd_cache[0][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
4327
            fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4328
            write_back_motion(h, mb_type);
4329

    
4330
            s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
4331
            s->current_picture.qscale_table[mb_xy]= s->qscale;
4332
            h->slice_table[ mb_xy ]= h->slice_num;
4333
            h->cbp_table[mb_xy] = 0;
4334
            h->chroma_pred_mode_table[mb_xy] = 0;
4335
            h->last_qscale_diff = 0;
4336

    
4337
            h->prev_mb_skiped= 1;
4338

    
4339
            return 0;
4340

    
4341
        }
4342
    }
4343
    h->prev_mb_skiped = 0;
4344

    
4345
    if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {
4346
        av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );
4347
        return -1;
4348
    }
4349

    
4350
    if( h->slice_type == B_TYPE ) {
4351
        if( mb_type < 23 ){
4352
            partition_count= b_mb_type_info[mb_type].partition_count;
4353
            mb_type=         b_mb_type_info[mb_type].type;
4354
        }else{
4355
            mb_type -= 23;
4356
            goto decode_intra_mb;
4357
        }
4358
    } else if( h->slice_type == P_TYPE ) {
4359
        if( mb_type < 5) {
4360
            partition_count= p_mb_type_info[mb_type].partition_count;
4361
            mb_type=         p_mb_type_info[mb_type].type;
4362
        } else {
4363
            mb_type -= 5;
4364
            goto decode_intra_mb;
4365
        }
4366
    } else {
4367
       assert(h->slice_type == I_TYPE);
4368
decode_intra_mb:
4369
        partition_count = 0;
4370
        cbp= i_mb_type_info[mb_type].cbp;
4371
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4372
        mb_type= i_mb_type_info[mb_type].type;
4373
    }
4374
#if 0
4375
    if(h->mb_field_decoding_flag)
4376
        mb_type |= MB_TYPE_INTERLACED;
4377
#endif
4378

    
4379
    s->current_picture.mb_type[mb_xy]= mb_type;
4380
    h->slice_table[ mb_xy ]= h->slice_num;
4381

    
4382
    if(IS_INTRA_PCM(mb_type)) {
4383
        /* TODO */
4384
        assert(0);
4385
        h->cbp_table[mb_xy] = 0xf +4*2; //FIXME ?!
4386
        h->cbp_table[mb_xy] |= 0x1C0;
4387
        h->chroma_pred_mode_table[mb_xy] = 0;
4388
        s->current_picture.qscale_table[mb_xy]= s->qscale;
4389
        return -1;
4390
    }
4391

    
4392
    fill_caches(h, mb_type);
4393

    
4394
    if( IS_INTRA( mb_type ) ) {
4395
        if( IS_INTRA4x4( mb_type ) ) {
4396
            int i;
4397
            for( i = 0; i < 16; i++ ) {
4398
                int pred = pred_intra_mode( h, i );
4399
                h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
4400

    
4401
                //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
4402
            }
4403
            write_back_intra_pred_mode(h);
4404
            if( check_intra4x4_pred_mode(h) < 0 ) return -1;
4405
        } else {
4406
            h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
4407
            if( h->intra16x16_pred_mode < 0 ) return -1;
4408
        }
4409
        h->chroma_pred_mode_table[mb_xy] =
4410
            h->chroma_pred_mode          = decode_cabac_mb_chroma_pre_mode( h );
4411

    
4412
        h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );
4413
        if( h->chroma_pred_mode < 0 ) return -1;
4414
    } else if( partition_count == 4 ) {
4415
        int i, j, sub_partition_count[4], list, ref[2][4];
4416

    
4417
        if( h->slice_type == B_TYPE ) {
4418
            for( i = 0; i < 4; i++ ) {
4419
                h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h );
4420
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4421
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4422
            }
4423
        } else {
4424
            for( i = 0; i < 4; i++ ) {
4425
                h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h );
4426
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4427
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4428
            }
4429
        }
4430

    
4431
        for( list = 0; list < 2; list++ ) {
4432
            if( h->ref_count[list] > 0 ) {
4433
                for( i = 0; i < 4; i++ ) {
4434
                    if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
4435
                        if( h->ref_count[list] > 1 )
4436
                            ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
4437
                        else
4438
                            ref[list][i] = 0;
4439
                    } else {
4440
                        ref[list][i] = -1;
4441
                    }
4442
                                                       h->ref_cache[list][ scan8[4*i]+1 ]=
4443
                    h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4444
                }
4445
            }
4446
        }
4447

    
4448
        for(list=0; list<2; list++){
4449
            for(i=0; i<4; i++){
4450
                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ];
4451

    
4452
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
4453
                    const int sub_mb_type= h->sub_mb_type[i];
4454
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4455
                    for(j=0; j<sub_partition_count[i]; j++){
4456
                        int mpx, mpy;
4457
                        int mx, my;
4458
                        const int index= 4*i + block_width*j;
4459
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4460
                        int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
4461
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
4462

    
4463
                        mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
4464
                        my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
4465
                        tprintf("final mv:%d %d\n", mx, my);
4466

    
4467
                        if(IS_SUB_8X8(sub_mb_type)){
4468
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
4469
                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4470
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
4471
                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4472

    
4473
                            mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=
4474
                            mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
4475
                            mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=
4476
                            mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
4477
                        }else if(IS_SUB_8X4(sub_mb_type)){
4478
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4479
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4480

    
4481
                            mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;
4482
                            mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;
4483
                        }else if(IS_SUB_4X8(sub_mb_type)){
4484
                            mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4485
                            mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4486

    
4487
                            mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;
4488
                            mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;
4489
                        }else{
4490
                            assert(IS_SUB_4X4(sub_mb_type));
4491
                            mv_cache[ 0 ][0]= mx;
4492
                            mv_cache[ 0 ][1]= my;
4493

    
4494
                            mvd_cache[ 0 ][0]= mx - mpx;
4495
                            mvd_cache[ 0 ][1]= my - mpy;
4496
                        }
4497
                    }
4498
                }else{
4499
                    uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4500
                    uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];
4501
                    p[0] = p[1] = p[8] = p[9] = 0;
4502
                    pd[0]= pd[1]= pd[8]= pd[9]= 0;
4503
                }
4504
            }
4505
        }
4506
    } else if( !IS_DIRECT(mb_type) ) {
4507
        int list, mx, my, i, mpx, mpy;
4508
        if(IS_16X16(mb_type)){
4509
            for(list=0; list<2; list++){
4510
                if(IS_DIR(mb_type, 0, list)){
4511
                    if(h->ref_count[list] > 0 ){
4512
                        const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0;
4513
                        fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
4514
                    }
4515
                }
4516
            }
4517
            for(list=0; list<2; list++){
4518
                if(IS_DIR(mb_type, 0, list)){
4519
                    pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);
4520

    
4521
                    mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );
4522
                    my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );
4523
                    tprintf("final mv:%d %d\n", mx, my);
4524

    
4525
                    fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
4526
                    fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
4527
                }
4528
            }
4529
        }
4530
        else if(IS_16X8(mb_type)){
4531
            for(list=0; list<2; list++){
4532
                if(h->ref_count[list]>0){
4533
                    for(i=0; i<2; i++){
4534
                        if(IS_DIR(mb_type, i, list)){
4535
                            const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0;
4536
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
4537
                        }
4538
                    }
4539
                }
4540
            }
4541
            for(list=0; list<2; list++){
4542
                for(i=0; i<2; i++){
4543
                    if(IS_DIR(mb_type, i, list)){