Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ eed4ebd2

History | View | Annotate | Download (234 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; //FIXME use s->b4_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 use_weight;
259
    int use_weight_chroma;
260
    int luma_log2_weight_denom;
261
    int chroma_log2_weight_denom;
262
    int luma_weight[2][16];
263
    int luma_offset[2][16];
264
    int chroma_weight[2][16][2];
265
    int chroma_offset[2][16][2];
266
    int implicit_weight[16][16];
267
   
268
    //deblock
269
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0 
270
    int slice_alpha_c0_offset;
271
    int slice_beta_offset;
272
     
273
    int redundant_pic_count;
274
    
275
    int direct_spatial_mv_pred;
276
    int dist_scale_factor[16];
277

    
278
    /**
279
     * num_ref_idx_l0/1_active_minus1 + 1
280
     */
281
    int ref_count[2];// FIXME split for AFF
282
    Picture *short_ref[16];
283
    Picture *long_ref[16];
284
    Picture default_ref_list[2][32];
285
    Picture ref_list[2][32]; //FIXME size?
286
    Picture field_ref_list[2][32]; //FIXME size?
287
    Picture *delayed_pic[16]; //FIXME size?
288
    
289
    /**
290
     * memory management control operations buffer.
291
     */
292
    MMCO mmco[MAX_MMCO_COUNT];
293
    int mmco_index;
294
    
295
    int long_ref_count;  ///< number of actual long term references
296
    int short_ref_count; ///< number of actual short term references
297
    
298
    //data partitioning
299
    GetBitContext intra_gb;
300
    GetBitContext inter_gb;
301
    GetBitContext *intra_gb_ptr;
302
    GetBitContext *inter_gb_ptr;
303
    
304
    DCTELEM mb[16*24] __align8;
305

    
306
    /**
307
     * Cabac
308
     */
309
    CABACContext cabac;
310
    uint8_t      cabac_state[399];
311
    int          cabac_init_idc;
312

    
313
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
314
    uint16_t     *cbp_table;
315
    int top_cbp;
316
    int left_cbp;
317
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
318
    uint8_t     *chroma_pred_mode_table;
319
    int         last_qscale_diff;
320
    int16_t     (*mvd_table[2])[2];
321
    int16_t     mvd_cache[2][5*8][2];
322
    uint8_t     *direct_table;
323
    uint8_t     direct_cache[5*8];
324

    
325
}H264Context;
326

    
327
static VLC coeff_token_vlc[4];
328
static VLC chroma_dc_coeff_token_vlc;
329

    
330
static VLC total_zeros_vlc[15];
331
static VLC chroma_dc_total_zeros_vlc[3];
332

    
333
static VLC run_vlc[6];
334
static VLC run7_vlc;
335

    
336
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
337
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
338
static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
339

    
340
static inline uint32_t pack16to32(int a, int b){
341
#ifdef WORDS_BIGENDIAN
342
   return (b&0xFFFF) + (a<<16);
343
#else
344
   return (a&0xFFFF) + (b<<16);
345
#endif
346
}
347

    
348
/**
349
 * fill a rectangle.
350
 * @param h height of the rectangle, should be a constant
351
 * @param w width of the rectangle, should be a constant
352
 * @param size the size of val (1 or 4), should be a constant
353
 */
354
static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
355
    uint8_t *p= (uint8_t*)vp;
356
    assert(size==1 || size==4);
357
    
358
    w      *= size;
359
    stride *= size;
360
    
361
//FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
362
    if(w==2 && h==2){
363
        *(uint16_t*)(p + 0)=
364
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
365
    }else if(w==2 && h==4){
366
        *(uint16_t*)(p + 0*stride)=
367
        *(uint16_t*)(p + 1*stride)=
368
        *(uint16_t*)(p + 2*stride)=
369
        *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
370
    }else if(w==4 && h==1){
371
        *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
372
    }else if(w==4 && h==2){
373
        *(uint32_t*)(p + 0*stride)=
374
        *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
375
    }else if(w==4 && h==4){
376
        *(uint32_t*)(p + 0*stride)=
377
        *(uint32_t*)(p + 1*stride)=
378
        *(uint32_t*)(p + 2*stride)=
379
        *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
380
    }else if(w==8 && h==1){
381
        *(uint32_t*)(p + 0)=
382
        *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
383
    }else if(w==8 && h==2){
384
        *(uint32_t*)(p + 0 + 0*stride)=
385
        *(uint32_t*)(p + 4 + 0*stride)=
386
        *(uint32_t*)(p + 0 + 1*stride)=
387
        *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
388
    }else if(w==8 && h==4){
389
        *(uint64_t*)(p + 0*stride)=
390
        *(uint64_t*)(p + 1*stride)=
391
        *(uint64_t*)(p + 2*stride)=
392
        *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
393
    }else if(w==16 && h==2){
394
        *(uint64_t*)(p + 0+0*stride)=
395
        *(uint64_t*)(p + 8+0*stride)=
396
        *(uint64_t*)(p + 0+1*stride)=
397
        *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
398
    }else if(w==16 && h==4){
399
        *(uint64_t*)(p + 0+0*stride)=
400
        *(uint64_t*)(p + 8+0*stride)=
401
        *(uint64_t*)(p + 0+1*stride)=
402
        *(uint64_t*)(p + 8+1*stride)=
403
        *(uint64_t*)(p + 0+2*stride)=
404
        *(uint64_t*)(p + 8+2*stride)=
405
        *(uint64_t*)(p + 0+3*stride)=
406
        *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
407
    }else
408
        assert(0);
409
}
410

    
411
static inline void fill_caches(H264Context *h, int mb_type){
412
    MpegEncContext * const s = &h->s;
413
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
414
    int topleft_xy, top_xy, topright_xy, left_xy[2];
415
    int topleft_type, top_type, topright_type, left_type[2];
416
    int left_block[4];
417
    int i;
418

    
419
    //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it 
420
    
421
    if(h->sps.mb_aff){
422
    //FIXME
423
        topleft_xy = 0; /* avoid warning */
424
        top_xy = 0; /* avoid warning */
425
        topright_xy = 0; /* avoid warning */
426
    }else{
427
        topleft_xy = mb_xy-1 - s->mb_stride;
428
        top_xy     = mb_xy   - s->mb_stride;
429
        topright_xy= mb_xy+1 - s->mb_stride;
430
        left_xy[0]   = mb_xy-1;
431
        left_xy[1]   = mb_xy-1;
432
        left_block[0]= 0;
433
        left_block[1]= 1;
434
        left_block[2]= 2;
435
        left_block[3]= 3;
436
    }
437

    
438
    topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
439
    top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
440
    topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
441
    left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
442
    left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
443

    
444
    if(IS_INTRA(mb_type)){
445
        h->topleft_samples_available= 
446
        h->top_samples_available= 
447
        h->left_samples_available= 0xFFFF;
448
        h->topright_samples_available= 0xEEEA;
449

    
450
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
451
            h->topleft_samples_available= 0xB3FF;
452
            h->top_samples_available= 0x33FF;
453
            h->topright_samples_available= 0x26EA;
454
        }
455
        for(i=0; i<2; i++){
456
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
457
                h->topleft_samples_available&= 0xDF5F;
458
                h->left_samples_available&= 0x5F5F;
459
            }
460
        }
461
        
462
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
463
            h->topleft_samples_available&= 0x7FFF;
464
        
465
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
466
            h->topright_samples_available&= 0xFBFF;
467
    
468
        if(IS_INTRA4x4(mb_type)){
469
            if(IS_INTRA4x4(top_type)){
470
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
471
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
472
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
473
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
474
            }else{
475
                int pred;
476
                if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
477
                    pred= 2;
478
                else{
479
                    pred= -1;
480
                }
481
                h->intra4x4_pred_mode_cache[4+8*0]=
482
                h->intra4x4_pred_mode_cache[5+8*0]=
483
                h->intra4x4_pred_mode_cache[6+8*0]=
484
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
485
            }
486
            for(i=0; i<2; i++){
487
                if(IS_INTRA4x4(left_type[i])){
488
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
489
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
490
                }else{
491
                    int pred;
492
                    if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
493
                        pred= 2;
494
                    else{
495
                        pred= -1;
496
                    }
497
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
498
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
499
                }
500
            }
501
        }
502
    }
503
    
504
    
505
/*
506
0 . T T. T T T T 
507
1 L . .L . . . . 
508
2 L . .L . . . . 
509
3 . T TL . . . . 
510
4 L . .L . . . . 
511
5 L . .. . . . . 
512
*/
513
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
514
    if(top_type){
515
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
516
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
517
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
518
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
519
    
520
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
521
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
522
    
523
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
524
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
525
        
526
        h->top_cbp= h->cbp_table[top_xy];
527
    }else{
528
        h->non_zero_count_cache[4+8*0]=      
529
        h->non_zero_count_cache[5+8*0]=
530
        h->non_zero_count_cache[6+8*0]=
531
        h->non_zero_count_cache[7+8*0]=
532
    
533
        h->non_zero_count_cache[1+8*0]=
534
        h->non_zero_count_cache[2+8*0]=
535
    
536
        h->non_zero_count_cache[1+8*3]=
537
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
538
        
539
        if(IS_INTRA(mb_type)) h->top_cbp= 0x1C0;
540
        else                  h->top_cbp= 0;
541
    }
542
    
543
    if(left_type[0]){
544
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
545
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
546
        h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
547
        h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
548
        h->left_cbp= h->cbp_table[left_xy[0]]; //FIXME interlacing
549
    }else{
550
        h->non_zero_count_cache[3+8*1]= 
551
        h->non_zero_count_cache[3+8*2]= 
552
        h->non_zero_count_cache[0+8*1]= 
553
        h->non_zero_count_cache[0+8*4]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
554
        
555
        if(IS_INTRA(mb_type)) h->left_cbp= 0x1C0;//FIXME interlacing
556
        else                  h->left_cbp= 0;
557
    }
558
    
559
    if(left_type[1]){
560
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
561
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
562
        h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
563
        h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
564
    }else{
565
        h->non_zero_count_cache[3+8*3]= 
566
        h->non_zero_count_cache[3+8*4]= 
567
        h->non_zero_count_cache[0+8*2]= 
568
        h->non_zero_count_cache[0+8*5]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
569
    }
570
    
571
#if 1
572
    //FIXME direct mb can skip much of this
573
    if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
574
        int list;
575
        for(list=0; list<2; list++){
576
            if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list) && !IS_DIRECT(mb_type)){
577
                /*if(!h->mv_cache_clean[list]){
578
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
579
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
580
                    h->mv_cache_clean[list]= 1;
581
                }*/
582
                continue;
583
            }
584
            h->mv_cache_clean[list]= 0;
585
            
586
            if(IS_INTER(topleft_type)){
587
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
588
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
589
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
590
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
591
            }else{
592
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
593
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
594
            }
595
            
596
            if(IS_INTER(top_type)){
597
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
598
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
599
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
600
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
601
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
602
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
603
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
604
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
605
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
606
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
607
            }else{
608
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
609
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
610
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
611
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
612
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
613
            }
614

    
615
            if(IS_INTER(topright_type)){
616
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
617
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
618
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
619
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
620
            }else{
621
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
622
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
623
            }
624
            
625
            //FIXME unify cleanup or sth
626
            if(IS_INTER(left_type[0])){
627
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
628
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
629
                *(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]];
630
                *(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]];
631
                h->ref_cache[list][scan8[0] - 1 + 0*8]= 
632
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
633
            }else{
634
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
635
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
636
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
637
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
638
            }
639
            
640
            if(IS_INTER(left_type[1])){
641
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
642
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
643
                *(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]];
644
                *(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]];
645
                h->ref_cache[list][scan8[0] - 1 + 2*8]= 
646
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
647
            }else{
648
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
649
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
650
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
651
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
652
            }
653

    
654
            h->ref_cache[list][scan8[5 ]+1] = 
655
            h->ref_cache[list][scan8[7 ]+1] = 
656
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewher else)
657
            h->ref_cache[list][scan8[4 ]] = 
658
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
659
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
660
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
661
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
662
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
663
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
664

    
665
            if( h->pps.cabac ) {
666
                /* XXX beurk, Load mvd */
667
                if(IS_INTER(topleft_type)){
668
                    const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
669
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
670
                }else{
671
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
672
                }
673

    
674
                if(IS_INTER(top_type)){
675
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
676
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
677
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
678
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
679
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
680
                }else{
681
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]= 
682
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]= 
683
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]= 
684
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
685
                }
686
                if(IS_INTER(left_type[0])){
687
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
688
                    *(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]];
689
                    *(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]];
690
                }else{
691
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
692
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
693
                }
694
                if(IS_INTER(left_type[1])){
695
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
696
                    *(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]];
697
                    *(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]];
698
                }else{
699
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
700
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
701
                }
702
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
703
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
704
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
705
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
706
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
707

    
708
                if(h->slice_type == B_TYPE){
709
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
710

    
711
                    if(IS_DIRECT(top_type)){
712
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
713
                    }else if(IS_8X8(top_type)){
714
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
715
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
716
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
717
                    }else{
718
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
719
                    }
720
                    
721
                    //FIXME interlacing
722
                    if(IS_DIRECT(left_type[0])){
723
                        h->direct_cache[scan8[0] - 1 + 0*8]=
724
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
725
                    }else if(IS_8X8(left_type[0])){
726
                        int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
727
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
728
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
729
                    }else{
730
                        h->direct_cache[scan8[0] - 1 + 0*8]=
731
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
732
                    }
733
                }
734
            }
735
        }
736
    }
737
#endif
738
}
739

    
740
static inline void write_back_intra_pred_mode(H264Context *h){
741
    MpegEncContext * const s = &h->s;
742
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
743

    
744
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
745
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
746
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
747
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
748
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
749
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
750
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
751
}
752

    
753
/**
754
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
755
 */
756
static inline int check_intra4x4_pred_mode(H264Context *h){
757
    MpegEncContext * const s = &h->s;
758
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
759
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
760
    int i;
761
    
762
    if(!(h->top_samples_available&0x8000)){
763
        for(i=0; i<4; i++){
764
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
765
            if(status<0){
766
                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);
767
                return -1;
768
            } else if(status){
769
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
770
            }
771
        }
772
    }
773
    
774
    if(!(h->left_samples_available&0x8000)){
775
        for(i=0; i<4; i++){
776
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
777
            if(status<0){
778
                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);
779
                return -1;
780
            } else if(status){
781
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
782
            }
783
        }
784
    }
785

    
786
    return 0;
787
} //FIXME cleanup like next
788

    
789
/**
790
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
791
 */
792
static inline int check_intra_pred_mode(H264Context *h, int mode){
793
    MpegEncContext * const s = &h->s;
794
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
795
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
796
    
797
    if(mode < 0 || mode > 6) {
798
        av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
799
        return -1;
800
    }
801
    
802
    if(!(h->top_samples_available&0x8000)){
803
        mode= top[ mode ];
804
        if(mode<0){
805
            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);
806
            return -1;
807
        }
808
    }
809
    
810
    if(!(h->left_samples_available&0x8000)){
811
        mode= left[ mode ];
812
        if(mode<0){
813
            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);
814
            return -1;
815
        } 
816
    }
817

    
818
    return mode;
819
}
820

    
821
/**
822
 * gets the predicted intra4x4 prediction mode.
823
 */
824
static inline int pred_intra_mode(H264Context *h, int n){
825
    const int index8= scan8[n];
826
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
827
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
828
    const int min= FFMIN(left, top);
829

    
830
    tprintf("mode:%d %d min:%d\n", left ,top, min);
831

    
832
    if(min<0) return DC_PRED;
833
    else      return min;
834
}
835

    
836
static inline void write_back_non_zero_count(H264Context *h){
837
    MpegEncContext * const s = &h->s;
838
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
839

    
840
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
841
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
842
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
843
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
844
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
845
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
846
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
847
    
848
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
849
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
850
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
851

    
852
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
853
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
854
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
855
}
856

    
857
/**
858
 * gets the predicted number of non zero coefficients.
859
 * @param n block index
860
 */
861
static inline int pred_non_zero_count(H264Context *h, int n){
862
    const int index8= scan8[n];
863
    const int left= h->non_zero_count_cache[index8 - 1];
864
    const int top = h->non_zero_count_cache[index8 - 8];
865
    int i= left + top;
866
    
867
    if(i<64) i= (i+1)>>1;
868

    
869
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
870

    
871
    return i&31;
872
}
873

    
874
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
875
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
876

    
877
    if(topright_ref != PART_NOT_AVAILABLE){
878
        *C= h->mv_cache[list][ i - 8 + part_width ];
879
        return topright_ref;
880
    }else{
881
        tprintf("topright MV not available\n");
882

    
883
        *C= h->mv_cache[list][ i - 8 - 1 ];
884
        return h->ref_cache[list][ i - 8 - 1 ];
885
    }
886
}
887

    
888
/**
889
 * gets the predicted MV.
890
 * @param n the block index
891
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
892
 * @param mx the x component of the predicted motion vector
893
 * @param my the y component of the predicted motion vector
894
 */
895
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
896
    const int index8= scan8[n];
897
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
898
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
899
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
900
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
901
    const int16_t * C;
902
    int diagonal_ref, match_count;
903

    
904
    assert(part_width==1 || part_width==2 || part_width==4);
905

    
906
/* mv_cache
907
  B . . A T T T T 
908
  U . . L . . , .
909
  U . . L . . . .
910
  U . . L . . , .
911
  . . . L . . . .
912
*/
913

    
914
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
915
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
916
    if(match_count > 1){ //most common
917
        *mx= mid_pred(A[0], B[0], C[0]);
918
        *my= mid_pred(A[1], B[1], C[1]);
919
    }else if(match_count==1){
920
        if(left_ref==ref){
921
            *mx= A[0];
922
            *my= A[1];        
923
        }else if(top_ref==ref){
924
            *mx= B[0];
925
            *my= B[1];        
926
        }else{
927
            *mx= C[0];
928
            *my= C[1];        
929
        }
930
    }else{
931
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
932
            *mx= A[0];
933
            *my= A[1];        
934
        }else{
935
            *mx= mid_pred(A[0], B[0], C[0]);
936
            *my= mid_pred(A[1], B[1], C[1]);
937
        }
938
    }
939
        
940
    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);
941
}
942

    
943
/**
944
 * gets the directionally predicted 16x8 MV.
945
 * @param n the block index
946
 * @param mx the x component of the predicted motion vector
947
 * @param my the y component of the predicted motion vector
948
 */
949
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
950
    if(n==0){
951
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
952
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
953

    
954
        tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
955
        
956
        if(top_ref == ref){
957
            *mx= B[0];
958
            *my= B[1];
959
            return;
960
        }
961
    }else{
962
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
963
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
964
        
965
        tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
966

    
967
        if(left_ref == ref){
968
            *mx= A[0];
969
            *my= A[1];
970
            return;
971
        }
972
    }
973

    
974
    //RARE
975
    pred_motion(h, n, 4, list, ref, mx, my);
976
}
977

    
978
/**
979
 * gets the directionally predicted 8x16 MV.
980
 * @param n the block index
981
 * @param mx the x component of the predicted motion vector
982
 * @param my the y component of the predicted motion vector
983
 */
984
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
985
    if(n==0){
986
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
987
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
988
        
989
        tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
990

    
991
        if(left_ref == ref){
992
            *mx= A[0];
993
            *my= A[1];
994
            return;
995
        }
996
    }else{
997
        const int16_t * C;
998
        int diagonal_ref;
999

    
1000
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1001
        
1002
        tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
1003

    
1004
        if(diagonal_ref == ref){ 
1005
            *mx= C[0];
1006
            *my= C[1];
1007
            return;
1008
        }
1009
    }
1010

    
1011
    //RARE
1012
    pred_motion(h, n, 2, list, ref, mx, my);
1013
}
1014

    
1015
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1016
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1017
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1018

    
1019
    tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1020

    
1021
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1022
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1023
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1024
       
1025
        *mx = *my = 0;
1026
        return;
1027
    }
1028
        
1029
    pred_motion(h, 0, 4, 0, 0, mx, my);
1030

    
1031
    return;
1032
}
1033

    
1034
static inline void direct_dist_scale_factor(H264Context * const h){
1035
    const int poc = h->s.current_picture_ptr->poc;
1036
    const int poc1 = h->ref_list[1][0].poc;
1037
    int i;
1038
    for(i=0; i<h->ref_count[0]; i++){
1039
        int poc0 = h->ref_list[0][i].poc;
1040
        int td = clip(poc1 - poc0, -128, 127);
1041
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
1042
            h->dist_scale_factor[i] = 256;
1043
        }else{
1044
            int tb = clip(poc - poc0, -128, 127);
1045
            int tx = (16384 + (ABS(td) >> 1)) / td;
1046
            h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1047
        }
1048
    }
1049
}
1050

    
1051
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1052
    MpegEncContext * const s = &h->s;
1053
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1054
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1055
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1056
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1057
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1058
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1059
    const int is_b8x8 = IS_8X8(*mb_type);
1060
    int sub_mb_type;
1061
    int i8, i4;
1062

    
1063
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1064
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1065
         * so we know exactly what block size to use */
1066
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1067
        *mb_type =    MB_TYPE_8x8;
1068
    }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1069
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1070
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1071
    }else{
1072
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1073
        *mb_type =    MB_TYPE_8x8;
1074
    }
1075
    if(!is_b8x8)
1076
        *mb_type |= MB_TYPE_DIRECT2;
1077

    
1078
    if(h->direct_spatial_mv_pred){
1079
        int ref[2];
1080
        int mv[2][2];
1081
        int list;
1082

    
1083
        /* ref = min(neighbors) */
1084
        for(list=0; list<2; list++){
1085
            int refa = h->ref_cache[list][scan8[0] - 1];
1086
            int refb = h->ref_cache[list][scan8[0] - 8];
1087
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1088
            if(refc == -2)
1089
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1090
            ref[list] = refa;
1091
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1092
                ref[list] = refb;
1093
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1094
                ref[list] = refc;
1095
            if(ref[list] < 0)
1096
                ref[list] = -1;
1097
        }
1098

    
1099
        if(ref[0] < 0 && ref[1] < 0){
1100
            ref[0] = ref[1] = 0;
1101
            mv[0][0] = mv[0][1] =
1102
            mv[1][0] = mv[1][1] = 0;
1103
        }else{
1104
            for(list=0; list<2; list++){
1105
                if(ref[list] >= 0)
1106
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1107
                else
1108
                    mv[list][0] = mv[list][1] = 0;
1109
            }
1110
        }
1111

    
1112
        if(ref[1] < 0){
1113
            *mb_type &= ~MB_TYPE_P0L1;
1114
            sub_mb_type &= ~MB_TYPE_P0L1;
1115
        }else if(ref[0] < 0){
1116
            *mb_type &= ~MB_TYPE_P0L0;
1117
            sub_mb_type &= ~MB_TYPE_P0L0;
1118
        }
1119

    
1120
        if(IS_16X16(*mb_type)){
1121
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1122
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1123
            if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1124
                ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1125
                if(ref[0] > 0)
1126
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1127
                else
1128
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1129
                if(ref[1] > 0)
1130
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1131
                else
1132
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1133
            }else{
1134
                fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1135
                fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1136
            }
1137
        }else{
1138
            for(i8=0; i8<4; i8++){
1139
                const int x8 = i8&1;
1140
                const int y8 = i8>>1;
1141
    
1142
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1143
                    continue;
1144
                h->sub_mb_type[i8] = sub_mb_type;
1145
    
1146
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1147
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1148
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1149
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1150
    
1151
                /* col_zero_flag */
1152
                if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1153
                    for(i4=0; i4<4; i4++){
1154
                        const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1155
                        if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1156
                            if(ref[0] == 0)
1157
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1158
                            if(ref[1] == 0)
1159
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1160
                        }
1161
                    }
1162
                }
1163
            }
1164
        }
1165
    }else{ /* direct temporal mv pred */
1166
        /* FIXME assumes that L1ref0 used the same ref lists as current frame */
1167
        if(IS_16X16(*mb_type)){
1168
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1169
            if(IS_INTRA(mb_type_col)){
1170
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1171
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1172
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1173
            }else{
1174
                const int ref0 = l1ref0[0];
1175
                const int dist_scale_factor = h->dist_scale_factor[ref0];
1176
                const int16_t *mv_col = l1mv0[0];
1177
                int mv_l0[2];
1178
                mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1179
                mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1180
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1181
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1182
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]), 4);
1183
            }
1184
        }else{
1185
            for(i8=0; i8<4; i8++){
1186
                const int x8 = i8&1;
1187
                const int y8 = i8>>1;
1188
                int ref0, dist_scale_factor;
1189
    
1190
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1191
                    continue;
1192
                h->sub_mb_type[i8] = sub_mb_type;
1193
                if(IS_INTRA(mb_type_col)){
1194
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1195
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1196
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1197
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1198
                    continue;
1199
                }
1200
    
1201
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1202
                dist_scale_factor = h->dist_scale_factor[ref0];
1203
    
1204
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1205
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1206
                for(i4=0; i4<4; i4++){
1207
                    const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1208
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1209
                    mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1210
                    mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1211
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1212
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1213
                }
1214
            }
1215
        }
1216
    }
1217
}
1218

    
1219
static inline void write_back_motion(H264Context *h, int mb_type){
1220
    MpegEncContext * const s = &h->s;
1221
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1222
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1223
    int list;
1224

    
1225
    for(list=0; list<2; list++){
1226
        int y;
1227
        if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
1228
            if(1){ //FIXME skip or never read if mb_type doesnt use it
1229
                for(y=0; y<4; y++){
1230
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1231
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1232
                }
1233
                if( h->pps.cabac ) {
1234
                    /* FIXME needed ? */
1235
                    for(y=0; y<4; y++){
1236
                        *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1237
                        *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1238
                    }
1239
                }
1240
                for(y=0; y<2; y++){
1241
                    *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1242
                }
1243
            }
1244
            continue;
1245
        }
1246
        
1247
        for(y=0; y<4; y++){
1248
            *(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];
1249
            *(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];
1250
        }
1251
        if( h->pps.cabac ) {
1252
            for(y=0; y<4; y++){
1253
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1254
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1255
            }
1256
        }
1257
        for(y=0; y<2; y++){
1258
            s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1259
            s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1260
        }
1261
    }
1262
    
1263
    if(h->slice_type == B_TYPE && h->pps.cabac){
1264
        if(IS_8X8(mb_type)){
1265
            h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1266
            h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1267
            h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1268
        }
1269
    }
1270
}
1271

    
1272
/**
1273
 * Decodes a network abstraction layer unit.
1274
 * @param consumed is the number of bytes used as input
1275
 * @param length is the length of the array
1276
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1277
 * @returns decoded bytes, might be src+1 if no escapes 
1278
 */
1279
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1280
    int i, si, di;
1281
    uint8_t *dst;
1282

    
1283
//    src[0]&0x80;                //forbidden bit
1284
    h->nal_ref_idc= src[0]>>5;
1285
    h->nal_unit_type= src[0]&0x1F;
1286

    
1287
    src++; length--;
1288
#if 0    
1289
    for(i=0; i<length; i++)
1290
        printf("%2X ", src[i]);
1291
#endif
1292
    for(i=0; i+1<length; i+=2){
1293
        if(src[i]) continue;
1294
        if(i>0 && src[i-1]==0) i--;
1295
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1296
            if(src[i+2]!=3){
1297
                /* startcode, so we must be past the end */
1298
                length=i;
1299
            }
1300
            break;
1301
        }
1302
    }
1303

    
1304
    if(i>=length-1){ //no escaped 0
1305
        *dst_length= length;
1306
        *consumed= length+1; //+1 for the header
1307
        return src; 
1308
    }
1309

    
1310
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1311
    dst= h->rbsp_buffer;
1312

    
1313
//printf("deoding esc\n");
1314
    si=di=0;
1315
    while(si<length){ 
1316
        //remove escapes (very rare 1:2^22)
1317
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1318
            if(src[si+2]==3){ //escape
1319
                dst[di++]= 0;
1320
                dst[di++]= 0;
1321
                si+=3;
1322
                continue;
1323
            }else //next start code
1324
                break;
1325
        }
1326

    
1327
        dst[di++]= src[si++];
1328
    }
1329

    
1330
    *dst_length= di;
1331
    *consumed= si + 1;//+1 for the header
1332
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1333
    return dst;
1334
}
1335

    
1336
#if 0
1337
/**
1338
 * @param src the data which should be escaped
1339
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1340
 * @param length the length of the src data
1341
 * @param dst_length the length of the dst array
1342
 * @returns length of escaped data in bytes or -1 if an error occured
1343
 */
1344
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1345
    int i, escape_count, si, di;
1346
    uint8_t *temp;
1347
    
1348
    assert(length>=0);
1349
    assert(dst_length>0);
1350
    
1351
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1352

1353
    if(length==0) return 1;
1354

1355
    escape_count= 0;
1356
    for(i=0; i<length; i+=2){
1357
        if(src[i]) continue;
1358
        if(i>0 && src[i-1]==0) 
1359
            i--;
1360
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1361
            escape_count++;
1362
            i+=2;
1363
        }
1364
    }
1365
    
1366
    if(escape_count==0){ 
1367
        if(dst+1 != src)
1368
            memcpy(dst+1, src, length);
1369
        return length + 1;
1370
    }
1371
    
1372
    if(length + escape_count + 1> dst_length)
1373
        return -1;
1374

1375
    //this should be damn rare (hopefully)
1376

1377
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1378
    temp= h->rbsp_buffer;
1379
//printf("encoding esc\n");
1380
    
1381
    si= 0;
1382
    di= 0;
1383
    while(si < length){
1384
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1385
            temp[di++]= 0; si++;
1386
            temp[di++]= 0; si++;
1387
            temp[di++]= 3; 
1388
            temp[di++]= src[si++];
1389
        }
1390
        else
1391
            temp[di++]= src[si++];
1392
    }
1393
    memcpy(dst+1, temp, length+escape_count);
1394
    
1395
    assert(di == length+escape_count);
1396
    
1397
    return di + 1;
1398
}
1399

1400
/**
1401
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1402
 */
1403
static void encode_rbsp_trailing(PutBitContext *pb){
1404
    int length;
1405
    put_bits(pb, 1, 1);
1406
    length= (-put_bits_count(pb))&7;
1407
    if(length) put_bits(pb, length, 0);
1408
}
1409
#endif
1410

    
1411
/**
1412
 * identifies the exact end of the bitstream
1413
 * @return the length of the trailing, or 0 if damaged
1414
 */
1415
static int decode_rbsp_trailing(uint8_t *src){
1416
    int v= *src;
1417
    int r;
1418

    
1419
    tprintf("rbsp trailing %X\n", v);
1420

    
1421
    for(r=1; r<9; r++){
1422
        if(v&1) return r;
1423
        v>>=1;
1424
    }
1425
    return 0;
1426
}
1427

    
1428
/**
1429
 * idct tranforms the 16 dc values and dequantize them.
1430
 * @param qp quantization parameter
1431
 */
1432
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1433
    const int qmul= dequant_coeff[qp][0];
1434
#define stride 16
1435
    int i;
1436
    int temp[16]; //FIXME check if this is a good idea
1437
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1438
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1439

    
1440
//memset(block, 64, 2*256);
1441
//return;
1442
    for(i=0; i<4; i++){
1443
        const int offset= y_offset[i];
1444
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1445
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1446
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1447
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1448

    
1449
        temp[4*i+0]= z0+z3;
1450
        temp[4*i+1]= z1+z2;
1451
        temp[4*i+2]= z1-z2;
1452
        temp[4*i+3]= z0-z3;
1453
    }
1454

    
1455
    for(i=0; i<4; i++){
1456
        const int offset= x_offset[i];
1457
        const int z0= temp[4*0+i] + temp[4*2+i];
1458
        const int z1= temp[4*0+i] - temp[4*2+i];
1459
        const int z2= temp[4*1+i] - temp[4*3+i];
1460
        const int z3= temp[4*1+i] + temp[4*3+i];
1461

    
1462
        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1463
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1464
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1465
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1466
    }
1467
}
1468

    
1469
#if 0
1470
/**
1471
 * dct tranforms the 16 dc values.
1472
 * @param qp quantization parameter ??? FIXME
1473
 */
1474
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1475
//    const int qmul= dequant_coeff[qp][0];
1476
    int i;
1477
    int temp[16]; //FIXME check if this is a good idea
1478
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1479
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1480

1481
    for(i=0; i<4; i++){
1482
        const int offset= y_offset[i];
1483
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1484
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1485
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1486
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1487

1488
        temp[4*i+0]= z0+z3;
1489
        temp[4*i+1]= z1+z2;
1490
        temp[4*i+2]= z1-z2;
1491
        temp[4*i+3]= z0-z3;
1492
    }
1493

1494
    for(i=0; i<4; i++){
1495
        const int offset= x_offset[i];
1496
        const int z0= temp[4*0+i] + temp[4*2+i];
1497
        const int z1= temp[4*0+i] - temp[4*2+i];
1498
        const int z2= temp[4*1+i] - temp[4*3+i];
1499
        const int z3= temp[4*1+i] + temp[4*3+i];
1500

1501
        block[stride*0 +offset]= (z0 + z3)>>1;
1502
        block[stride*2 +offset]= (z1 + z2)>>1;
1503
        block[stride*8 +offset]= (z1 - z2)>>1;
1504
        block[stride*10+offset]= (z0 - z3)>>1;
1505
    }
1506
}
1507
#endif
1508

    
1509
#undef xStride
1510
#undef stride
1511

    
1512
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1513
    const int qmul= dequant_coeff[qp][0];
1514
    const int stride= 16*2;
1515
    const int xStride= 16;
1516
    int a,b,c,d,e;
1517

    
1518
    a= block[stride*0 + xStride*0];
1519
    b= block[stride*0 + xStride*1];
1520
    c= block[stride*1 + xStride*0];
1521
    d= block[stride*1 + xStride*1];
1522

    
1523
    e= a-b;
1524
    a= a+b;
1525
    b= c-d;
1526
    c= c+d;
1527

    
1528
    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1529
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1530
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1531
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1532
}
1533

    
1534
#if 0
1535
static void chroma_dc_dct_c(DCTELEM *block){
1536
    const int stride= 16*2;
1537
    const int xStride= 16;
1538
    int a,b,c,d,e;
1539

1540
    a= block[stride*0 + xStride*0];
1541
    b= block[stride*0 + xStride*1];
1542
    c= block[stride*1 + xStride*0];
1543
    d= block[stride*1 + xStride*1];
1544

1545
    e= a-b;
1546
    a= a+b;
1547
    b= c-d;
1548
    c= c+d;
1549

1550
    block[stride*0 + xStride*0]= (a+c);
1551
    block[stride*0 + xStride*1]= (e+b);
1552
    block[stride*1 + xStride*0]= (a-c);
1553
    block[stride*1 + xStride*1]= (e-b);
1554
}
1555
#endif
1556

    
1557
/**
1558
 * gets the chroma qp.
1559
 */
1560
static inline int get_chroma_qp(H264Context *h, int qscale){
1561
    
1562
    return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1563
}
1564

    
1565

    
1566
#if 0
1567
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1568
    int i;
1569
    //FIXME try int temp instead of block
1570
    
1571
    for(i=0; i<4; i++){
1572
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1573
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1574
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1575
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1576
        const int z0= d0 + d3;
1577
        const int z3= d0 - d3;
1578
        const int z1= d1 + d2;
1579
        const int z2= d1 - d2;
1580
        
1581
        block[0 + 4*i]=   z0 +   z1;
1582
        block[1 + 4*i]= 2*z3 +   z2;
1583
        block[2 + 4*i]=   z0 -   z1;
1584
        block[3 + 4*i]=   z3 - 2*z2;
1585
    }    
1586

1587
    for(i=0; i<4; i++){
1588
        const int z0= block[0*4 + i] + block[3*4 + i];
1589
        const int z3= block[0*4 + i] - block[3*4 + i];
1590
        const int z1= block[1*4 + i] + block[2*4 + i];
1591
        const int z2= block[1*4 + i] - block[2*4 + i];
1592
        
1593
        block[0*4 + i]=   z0 +   z1;
1594
        block[1*4 + i]= 2*z3 +   z2;
1595
        block[2*4 + i]=   z0 -   z1;
1596
        block[3*4 + i]=   z3 - 2*z2;
1597
    }
1598
}
1599
#endif
1600

    
1601
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1602
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1603
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1604
    int i;
1605
    const int * const quant_table= quant_coeff[qscale];
1606
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1607
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1608
    const unsigned int threshold2= (threshold1<<1);
1609
    int last_non_zero;
1610

    
1611
    if(seperate_dc){
1612
        if(qscale<=18){
1613
            //avoid overflows
1614
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1615
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1616
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1617

    
1618
            int level= block[0]*quant_coeff[qscale+18][0];
1619
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1620
                if(level>0){
1621
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1622
                    block[0]= level;
1623
                }else{
1624
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1625
                    block[0]= -level;
1626
                }
1627
//                last_non_zero = i;
1628
            }else{
1629
                block[0]=0;
1630
            }
1631
        }else{
1632
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1633
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1634
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1635

    
1636
            int level= block[0]*quant_table[0];
1637
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1638
                if(level>0){
1639
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1640
                    block[0]= level;
1641
                }else{
1642
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1643
                    block[0]= -level;
1644
                }
1645
//                last_non_zero = i;
1646
            }else{
1647
                block[0]=0;
1648
            }
1649
        }
1650
        last_non_zero= 0;
1651
        i=1;
1652
    }else{
1653
        last_non_zero= -1;
1654
        i=0;
1655
    }
1656

    
1657
    for(; i<16; i++){
1658
        const int j= scantable[i];
1659
        int level= block[j]*quant_table[j];
1660

    
1661
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1662
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1663
        if(((unsigned)(level+threshold1))>threshold2){
1664
            if(level>0){
1665
                level= (bias + level)>>QUANT_SHIFT;
1666
                block[j]= level;
1667
            }else{
1668
                level= (bias - level)>>QUANT_SHIFT;
1669
                block[j]= -level;
1670
            }
1671
            last_non_zero = i;
1672
        }else{
1673
            block[j]=0;
1674
        }
1675
    }
1676

    
1677
    return last_non_zero;
1678
}
1679

    
1680
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1681
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1682
    ((uint32_t*)(src+0*stride))[0]= a;
1683
    ((uint32_t*)(src+1*stride))[0]= a;
1684
    ((uint32_t*)(src+2*stride))[0]= a;
1685
    ((uint32_t*)(src+3*stride))[0]= a;
1686
}
1687

    
1688
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1689
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1690
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1691
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1692
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1693
}
1694

    
1695
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1696
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1697
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1698
    
1699
    ((uint32_t*)(src+0*stride))[0]= 
1700
    ((uint32_t*)(src+1*stride))[0]= 
1701
    ((uint32_t*)(src+2*stride))[0]= 
1702
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1703
}
1704

    
1705
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1706
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1707
    
1708
    ((uint32_t*)(src+0*stride))[0]= 
1709
    ((uint32_t*)(src+1*stride))[0]= 
1710
    ((uint32_t*)(src+2*stride))[0]= 
1711
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1712
}
1713

    
1714
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1715
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1716
    
1717
    ((uint32_t*)(src+0*stride))[0]= 
1718
    ((uint32_t*)(src+1*stride))[0]= 
1719
    ((uint32_t*)(src+2*stride))[0]= 
1720
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1721
}
1722

    
1723
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1724
    ((uint32_t*)(src+0*stride))[0]= 
1725
    ((uint32_t*)(src+1*stride))[0]= 
1726
    ((uint32_t*)(src+2*stride))[0]= 
1727
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1728
}
1729

    
1730

    
1731
#define LOAD_TOP_RIGHT_EDGE\
1732
    const int t4= topright[0];\
1733
    const int t5= topright[1];\
1734
    const int t6= topright[2];\
1735
    const int t7= topright[3];\
1736

    
1737
#define LOAD_LEFT_EDGE\
1738
    const int l0= src[-1+0*stride];\
1739
    const int l1= src[-1+1*stride];\
1740
    const int l2= src[-1+2*stride];\
1741
    const int l3= src[-1+3*stride];\
1742

    
1743
#define LOAD_TOP_EDGE\
1744
    const int t0= src[ 0-1*stride];\
1745
    const int t1= src[ 1-1*stride];\
1746
    const int t2= src[ 2-1*stride];\
1747
    const int t3= src[ 3-1*stride];\
1748

    
1749
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1750
    const int lt= src[-1-1*stride];
1751
    LOAD_TOP_EDGE
1752
    LOAD_LEFT_EDGE
1753

    
1754
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1755
    src[0+2*stride]=
1756
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1757
    src[0+1*stride]=
1758
    src[1+2*stride]=
1759
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1760
    src[0+0*stride]=
1761
    src[1+1*stride]=
1762
    src[2+2*stride]=
1763
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1764
    src[1+0*stride]=
1765
    src[2+1*stride]=
1766
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1767
    src[2+0*stride]=
1768
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1769
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1770
}
1771

    
1772
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1773
    LOAD_TOP_EDGE    
1774
    LOAD_TOP_RIGHT_EDGE    
1775
//    LOAD_LEFT_EDGE    
1776

    
1777
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1778
    src[1+0*stride]=
1779
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1780
    src[2+0*stride]=
1781
    src[1+1*stride]=
1782
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1783
    src[3+0*stride]=
1784
    src[2+1*stride]=
1785
    src[1+2*stride]=
1786
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1787
    src[3+1*stride]=
1788
    src[2+2*stride]=
1789
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1790
    src[3+2*stride]=
1791
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1792
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1793
}
1794

    
1795
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1796
    const int lt= src[-1-1*stride];
1797
    LOAD_TOP_EDGE    
1798
    LOAD_LEFT_EDGE    
1799
    const __attribute__((unused)) int unu= l3;
1800

    
1801
    src[0+0*stride]=
1802
    src[1+2*stride]=(lt + t0 + 1)>>1;
1803
    src[1+0*stride]=
1804
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1805
    src[2+0*stride]=
1806
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1807
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1808
    src[0+1*stride]=
1809
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1810
    src[1+1*stride]=
1811
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1812
    src[2+1*stride]=
1813
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1814
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1815
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1816
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1817
}
1818

    
1819
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1820
    LOAD_TOP_EDGE    
1821
    LOAD_TOP_RIGHT_EDGE    
1822
    const __attribute__((unused)) int unu= t7;
1823

    
1824
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1825
    src[1+0*stride]=
1826
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1827
    src[2+0*stride]=
1828
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1829
    src[3+0*stride]=
1830
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1831
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1832
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1833
    src[1+1*stride]=
1834
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1835
    src[2+1*stride]=
1836
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1837
    src[3+1*stride]=
1838
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1839
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1840
}
1841

    
1842
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1843
    LOAD_LEFT_EDGE    
1844

    
1845
    src[0+0*stride]=(l0 + l1 + 1)>>1;
1846
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1847
    src[2+0*stride]=
1848
    src[0+1*stride]=(l1 + l2 + 1)>>1;
1849
    src[3+0*stride]=
1850
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1851
    src[2+1*stride]=
1852
    src[0+2*stride]=(l2 + l3 + 1)>>1;
1853
    src[3+1*stride]=
1854
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1855
    src[3+2*stride]=
1856
    src[1+3*stride]=
1857
    src[0+3*stride]=
1858
    src[2+2*stride]=
1859
    src[2+3*stride]=
1860
    src[3+3*stride]=l3;
1861
}
1862
    
1863
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1864
    const int lt= src[-1-1*stride];
1865
    LOAD_TOP_EDGE    
1866
    LOAD_LEFT_EDGE    
1867
    const __attribute__((unused)) int unu= t3;
1868

    
1869
    src[0+0*stride]=
1870
    src[2+1*stride]=(lt + l0 + 1)>>1;
1871
    src[1+0*stride]=
1872
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1873
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1874
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1875
    src[0+1*stride]=
1876
    src[2+2*stride]=(l0 + l1 + 1)>>1;
1877
    src[1+1*stride]=
1878
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1879
    src[0+2*stride]=
1880
    src[2+3*stride]=(l1 + l2+ 1)>>1;
1881
    src[1+2*stride]=
1882
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1883
    src[0+3*stride]=(l2 + l3 + 1)>>1;
1884
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1885
}
1886

    
1887
static void pred16x16_vertical_c(uint8_t *src, int stride){
1888
    int i;
1889
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1890
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1891
    const uint32_t c= ((uint32_t*)(src-stride))[2];
1892
    const uint32_t d= ((uint32_t*)(src-stride))[3];
1893
    
1894
    for(i=0; i<16; i++){
1895
        ((uint32_t*)(src+i*stride))[0]= a;
1896
        ((uint32_t*)(src+i*stride))[1]= b;
1897
        ((uint32_t*)(src+i*stride))[2]= c;
1898
        ((uint32_t*)(src+i*stride))[3]= d;
1899
    }
1900
}
1901

    
1902
static void pred16x16_horizontal_c(uint8_t *src, int stride){
1903
    int i;
1904

    
1905
    for(i=0; i<16; i++){
1906
        ((uint32_t*)(src+i*stride))[0]=
1907
        ((uint32_t*)(src+i*stride))[1]=
1908
        ((uint32_t*)(src+i*stride))[2]=
1909
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1910
    }
1911
}
1912

    
1913
static void pred16x16_dc_c(uint8_t *src, int stride){
1914
    int i, dc=0;
1915

    
1916
    for(i=0;i<16; i++){
1917
        dc+= src[-1+i*stride];
1918
    }
1919
    
1920
    for(i=0;i<16; i++){
1921
        dc+= src[i-stride];
1922
    }
1923

    
1924
    dc= 0x01010101*((dc + 16)>>5);
1925

    
1926
    for(i=0; i<16; i++){
1927
        ((uint32_t*)(src+i*stride))[0]=
1928
        ((uint32_t*)(src+i*stride))[1]=
1929
        ((uint32_t*)(src+i*stride))[2]=
1930
        ((uint32_t*)(src+i*stride))[3]= dc;
1931
    }
1932
}
1933

    
1934
static void pred16x16_left_dc_c(uint8_t *src, int stride){
1935
    int i, dc=0;
1936

    
1937
    for(i=0;i<16; i++){
1938
        dc+= src[-1+i*stride];
1939
    }
1940
    
1941
    dc= 0x01010101*((dc + 8)>>4);
1942

    
1943
    for(i=0; i<16; i++){
1944
        ((uint32_t*)(src+i*stride))[0]=
1945
        ((uint32_t*)(src+i*stride))[1]=
1946
        ((uint32_t*)(src+i*stride))[2]=
1947
        ((uint32_t*)(src+i*stride))[3]= dc;
1948
    }
1949
}
1950

    
1951
static void pred16x16_top_dc_c(uint8_t *src, int stride){
1952
    int i, dc=0;
1953

    
1954
    for(i=0;i<16; i++){
1955
        dc+= src[i-stride];
1956
    }
1957
    dc= 0x01010101*((dc + 8)>>4);
1958

    
1959
    for(i=0; i<16; i++){
1960
        ((uint32_t*)(src+i*stride))[0]=
1961
        ((uint32_t*)(src+i*stride))[1]=
1962
        ((uint32_t*)(src+i*stride))[2]=
1963
        ((uint32_t*)(src+i*stride))[3]= dc;
1964
    }
1965
}
1966

    
1967
static void pred16x16_128_dc_c(uint8_t *src, int stride){
1968
    int i;
1969

    
1970
    for(i=0; i<16; i++){
1971
        ((uint32_t*)(src+i*stride))[0]=
1972
        ((uint32_t*)(src+i*stride))[1]=
1973
        ((uint32_t*)(src+i*stride))[2]=
1974
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1975
    }
1976
}
1977

    
1978
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1979
  int i, j, k;
1980
  int a;
1981
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
1982
  const uint8_t * const src0 = src+7-stride;
1983
  const uint8_t *src1 = src+8*stride-1;
1984
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
1985
  int H = src0[1] - src0[-1];
1986
  int V = src1[0] - src2[ 0];
1987
  for(k=2; k<=8; ++k) {
1988
    src1 += stride; src2 -= stride;
1989
    H += k*(src0[k] - src0[-k]);
1990
    V += k*(src1[0] - src2[ 0]);
1991
  }
1992
  if(svq3){
1993
    H = ( 5*(H/4) ) / 16;
1994
    V = ( 5*(V/4) ) / 16;
1995

    
1996
    /* required for 100% accuracy */
1997
    i = H; H = V; V = i;
1998
  }else{
1999
    H = ( 5*H+32 ) >> 6;
2000
    V = ( 5*V+32 ) >> 6;
2001
  }
2002

    
2003
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2004
  for(j=16; j>0; --j) {
2005
    int b = a;
2006
    a += V;
2007
    for(i=-16; i<0; i+=4) {
2008
      src[16+i] = cm[ (b    ) >> 5 ];
2009
      src[17+i] = cm[ (b+  H) >> 5 ];
2010
      src[18+i] = cm[ (b+2*H) >> 5 ];
2011
      src[19+i] = cm[ (b+3*H) >> 5 ];
2012
      b += 4*H;
2013
    }
2014
    src += stride;
2015
  }
2016
}
2017

    
2018
static void pred16x16_plane_c(uint8_t *src, int stride){
2019
    pred16x16_plane_compat_c(src, stride, 0);
2020
}
2021

    
2022
static void pred8x8_vertical_c(uint8_t *src, int stride){
2023
    int i;
2024
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2025
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2026
    
2027
    for(i=0; i<8; i++){
2028
        ((uint32_t*)(src+i*stride))[0]= a;
2029
        ((uint32_t*)(src+i*stride))[1]= b;
2030
    }
2031
}
2032

    
2033
static void pred8x8_horizontal_c(uint8_t *src, int stride){
2034
    int i;
2035

    
2036
    for(i=0; i<8; i++){
2037
        ((uint32_t*)(src+i*stride))[0]=
2038
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2039
    }
2040
}
2041

    
2042
static void pred8x8_128_dc_c(uint8_t *src, int stride){
2043
    int i;
2044

    
2045
    for(i=0; i<4; i++){
2046
        ((uint32_t*)(src+i*stride))[0]= 
2047
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2048
    }
2049
    for(i=4; i<8; i++){
2050
        ((uint32_t*)(src+i*stride))[0]= 
2051
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2052
    }
2053
}
2054

    
2055
static void pred8x8_left_dc_c(uint8_t *src, int stride){
2056
    int i;
2057
    int dc0, dc2;
2058

    
2059
    dc0=dc2=0;
2060
    for(i=0;i<4; i++){
2061
        dc0+= src[-1+i*stride];
2062
        dc2+= src[-1+(i+4)*stride];
2063
    }
2064
    dc0= 0x01010101*((dc0 + 2)>>2);
2065
    dc2= 0x01010101*((dc2 + 2)>>2);
2066

    
2067
    for(i=0; i<4; i++){
2068
        ((uint32_t*)(src+i*stride))[0]=
2069
        ((uint32_t*)(src+i*stride))[1]= dc0;
2070
    }
2071
    for(i=4; i<8; i++){
2072
        ((uint32_t*)(src+i*stride))[0]=
2073
        ((uint32_t*)(src+i*stride))[1]= dc2;
2074
    }
2075
}
2076

    
2077
static void pred8x8_top_dc_c(uint8_t *src, int stride){
2078
    int i;
2079
    int dc0, dc1;
2080

    
2081
    dc0=dc1=0;
2082
    for(i=0;i<4; i++){
2083
        dc0+= src[i-stride];
2084
        dc1+= src[4+i-stride];
2085
    }
2086
    dc0= 0x01010101*((dc0 + 2)>>2);
2087
    dc1= 0x01010101*((dc1 + 2)>>2);
2088

    
2089
    for(i=0; i<4; i++){
2090
        ((uint32_t*)(src+i*stride))[0]= dc0;
2091
        ((uint32_t*)(src+i*stride))[1]= dc1;
2092
    }
2093
    for(i=4; i<8; i++){
2094
        ((uint32_t*)(src+i*stride))[0]= dc0;
2095
        ((uint32_t*)(src+i*stride))[1]= dc1;
2096
    }
2097
}
2098

    
2099

    
2100
static void pred8x8_dc_c(uint8_t *src, int stride){
2101
    int i;
2102
    int dc0, dc1, dc2, dc3;
2103

    
2104
    dc0=dc1=dc2=0;
2105
    for(i=0;i<4; i++){
2106
        dc0+= src[-1+i*stride] + src[i-stride];
2107
        dc1+= src[4+i-stride];
2108
        dc2+= src[-1+(i+4)*stride];
2109
    }
2110
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2111
    dc0= 0x01010101*((dc0 + 4)>>3);
2112
    dc1= 0x01010101*((dc1 + 2)>>2);
2113
    dc2= 0x01010101*((dc2 + 2)>>2);
2114

    
2115
    for(i=0; i<4; i++){
2116
        ((uint32_t*)(src+i*stride))[0]= dc0;
2117
        ((uint32_t*)(src+i*stride))[1]= dc1;
2118
    }
2119
    for(i=4; i<8; i++){
2120
        ((uint32_t*)(src+i*stride))[0]= dc2;
2121
        ((uint32_t*)(src+i*stride))[1]= dc3;
2122
    }
2123
}
2124

    
2125
static void pred8x8_plane_c(uint8_t *src, int stride){
2126
  int j, k;
2127
  int a;
2128
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2129
  const uint8_t * const src0 = src+3-stride;
2130
  const uint8_t *src1 = src+4*stride-1;
2131
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2132
  int H = src0[1] - src0[-1];
2133
  int V = src1[0] - src2[ 0];
2134
  for(k=2; k<=4; ++k) {
2135
    src1 += stride; src2 -= stride;
2136
    H += k*(src0[k] - src0[-k]);
2137
    V += k*(src1[0] - src2[ 0]);
2138
  }
2139
  H = ( 17*H+16 ) >> 5;
2140
  V = ( 17*V+16 ) >> 5;
2141

    
2142
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2143
  for(j=8; j>0; --j) {
2144
    int b = a;
2145
    a += V;
2146
    src[0] = cm[ (b    ) >> 5 ];
2147
    src[1] = cm[ (b+  H) >> 5 ];
2148
    src[2] = cm[ (b+2*H) >> 5 ];
2149
    src[3] = cm[ (b+3*H) >> 5 ];
2150
    src[4] = cm[ (b+4*H) >> 5 ];
2151
    src[5] = cm[ (b+5*H) >> 5 ];
2152
    src[6] = cm[ (b+6*H) >> 5 ];
2153
    src[7] = cm[ (b+7*H) >> 5 ];
2154
    src += stride;
2155
  }
2156
}
2157

    
2158
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2159
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2160
                           int src_x_offset, int src_y_offset,
2161
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2162
    MpegEncContext * const s = &h->s;
2163
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2164
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2165
    const int luma_xy= (mx&3) + ((my&3)<<2);
2166
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2167
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2168
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2169
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2170
    int extra_height= extra_width;
2171
    int emu=0;
2172
    const int full_mx= mx>>2;
2173
    const int full_my= my>>2;
2174
    
2175
    assert(pic->data[0]);
2176
    
2177
    if(mx&7) extra_width -= 3;
2178
    if(my&7) extra_height -= 3;
2179
    
2180
    if(   full_mx < 0-extra_width 
2181
       || full_my < 0-extra_height 
2182
       || full_mx + 16/*FIXME*/ > s->width + extra_width 
2183
       || full_my + 16/*FIXME*/ > s->height + extra_height){
2184
        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);
2185
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2186
        emu=1;
2187
    }
2188
    
2189
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2190
    if(!square){
2191
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2192
    }
2193
    
2194
    if(s->flags&CODEC_FLAG_GRAY) return;
2195
    
2196
    if(emu){
2197
        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);
2198
            src_cb= s->edge_emu_buffer;
2199
    }
2200
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2201

    
2202
    if(emu){
2203
        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);
2204
            src_cr= s->edge_emu_buffer;
2205
    }
2206
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2207
}
2208

    
2209
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2210
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2211
                           int x_offset, int y_offset,
2212
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2213
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2214
                           int list0, int list1){
2215
    MpegEncContext * const s = &h->s;
2216
    qpel_mc_func *qpix_op=  qpix_put;
2217
    h264_chroma_mc_func chroma_op= chroma_put;
2218
    
2219
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2220
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2221
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2222
    x_offset += 8*s->mb_x;
2223
    y_offset += 8*s->mb_y;
2224
    
2225
    if(list0){
2226
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2227
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2228
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2229
                           qpix_op, chroma_op);
2230

    
2231
        qpix_op=  qpix_avg;
2232
        chroma_op= chroma_avg;
2233
    }
2234

    
2235
    if(list1){
2236
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2237
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2238
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2239
                           qpix_op, chroma_op);
2240
    }
2241
}
2242

    
2243
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2244
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2245
                           int x_offset, int y_offset,
2246
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2247
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2248
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2249
                           int list0, int list1){
2250
    MpegEncContext * const s = &h->s;
2251

    
2252
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2253
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2254
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2255
    x_offset += 8*s->mb_x;
2256
    y_offset += 8*s->mb_y;
2257
    
2258
    if(list0 && list1){
2259
        /* don't optimize for luma-only case, since B-frames usually
2260
         * use implicit weights => chroma too. */
2261
        uint8_t *tmp_cb = s->obmc_scratchpad;
2262
        uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2263
        uint8_t *tmp_y  = tmp_cr + 8*s->uvlinesize;
2264
        int refn0 = h->ref_cache[0][ scan8[n] ];
2265
        int refn1 = h->ref_cache[1][ scan8[n] ];
2266

    
2267
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2268
                    dest_y, dest_cb, dest_cr,
2269
                    x_offset, y_offset, qpix_put, chroma_put);
2270
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2271
                    tmp_y, tmp_cb, tmp_cr,
2272
                    x_offset, y_offset, qpix_put, chroma_put);
2273

    
2274
        if(h->use_weight == 2){
2275
            int weight0 = h->implicit_weight[refn0][refn1];
2276
            int weight1 = 64 - weight0;
2277
            luma_weight_avg(  dest_y,  tmp_y,  s->  linesize, 5, weight0, weight1, 0, 0);
2278
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2279
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2280
        }else{
2281
            luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2282
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1], 
2283
                            h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2284
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2285
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0], 
2286
                            h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2287
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2288
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1], 
2289
                            h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2290
        }
2291
    }else{
2292
        int list = list1 ? 1 : 0;
2293
        int refn = h->ref_cache[list][ scan8[n] ];
2294
        Picture *ref= &h->ref_list[list][refn];
2295
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2296
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
2297
                    qpix_put, chroma_put);
2298

    
2299
        luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2300
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
2301
        if(h->use_weight_chroma){
2302
            chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2303
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2304
            chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2305
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2306
        }
2307
    }
2308
}
2309

    
2310
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2311
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2312
                           int x_offset, int y_offset,
2313
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2314
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2315
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg, 
2316
                           int list0, int list1){
2317
    if((h->use_weight==2 && list0 && list1
2318
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2319
       || h->use_weight==1)
2320
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2321
                         x_offset, y_offset, qpix_put, chroma_put,
2322
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2323
    else
2324
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2325
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2326
}
2327

    
2328
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2329
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2330
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2331
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2332
    MpegEncContext * const s = &h->s;
2333
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2334
    const int mb_type= s->current_picture.mb_type[mb_xy];
2335
    
2336
    assert(IS_INTER(mb_type));
2337
    
2338
    if(IS_16X16(mb_type)){
2339
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2340
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2341
                &weight_op[0], &weight_avg[0],
2342
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2343
    }else if(IS_16X8(mb_type)){
2344
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2345
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2346
                &weight_op[1], &weight_avg[1],
2347
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2348
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2349
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2350
                &weight_op[1], &weight_avg[1],
2351
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2352
    }else if(IS_8X16(mb_type)){
2353
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2354
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2355
                &weight_op[2], &weight_avg[2],
2356
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2357
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2358
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2359
                &weight_op[2], &weight_avg[2],
2360
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2361
    }else{
2362
        int i;
2363
        
2364
        assert(IS_8X8(mb_type));
2365

    
2366
        for(i=0; i<4; i++){
2367
            const int sub_mb_type= h->sub_mb_type[i];
2368
            const int n= 4*i;
2369
            int x_offset= (i&1)<<2;
2370
            int y_offset= (i&2)<<1;
2371

    
2372
            if(IS_SUB_8X8(sub_mb_type)){
2373
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2374
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2375
                    &weight_op[3], &weight_avg[3],
2376
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2377
            }else if(IS_SUB_8X4(sub_mb_type)){
2378
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2379
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2380
                    &weight_op[4], &weight_avg[4],
2381
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2382
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2383
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2384
                    &weight_op[4], &weight_avg[4],
2385
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2386
            }else if(IS_SUB_4X8(sub_mb_type)){
2387
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2388
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2389
                    &weight_op[5], &weight_avg[5],
2390
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2391
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2392
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2393
                    &weight_op[5], &weight_avg[5],
2394
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2395
            }else{
2396
                int j;
2397
                assert(IS_SUB_4X4(sub_mb_type));
2398
                for(j=0; j<4; j++){
2399
                    int sub_x_offset= x_offset + 2*(j&1);
2400
                    int sub_y_offset= y_offset +   (j&2);
2401
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2402
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2403
                        &weight_op[6], &weight_avg[6],
2404
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2405
                }
2406
            }
2407
        }
2408
    }
2409
}
2410

    
2411
static void decode_init_vlc(H264Context *h){
2412
    static int done = 0;
2413

    
2414
    if (!done) {
2415
        int i;
2416
        done = 1;
2417

    
2418
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2419
                 &chroma_dc_coeff_token_len [0], 1, 1,
2420
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2421

    
2422
        for(i=0; i<4; i++){
2423
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2424
                     &coeff_token_len [i][0], 1, 1,
2425
                     &coeff_token_bits[i][0], 1, 1, 1);
2426
        }
2427

    
2428
        for(i=0; i<3; i++){
2429
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2430
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2431
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2432
        }
2433
        for(i=0; i<15; i++){
2434
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2435
                     &total_zeros_len [i][0], 1, 1,
2436
                     &total_zeros_bits[i][0], 1, 1, 1);
2437
        }
2438

    
2439
        for(i=0; i<6; i++){
2440
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2441
                     &run_len [i][0], 1, 1,
2442
                     &run_bits[i][0], 1, 1, 1);
2443
        }
2444
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2445
                 &run_len [6][0], 1, 1,
2446
                 &run_bits[6][0], 1, 1, 1);
2447
    }
2448
}
2449

    
2450
/**
2451
 * Sets the intra prediction function pointers.
2452
 */
2453
static void init_pred_ptrs(H264Context *h){
2454
//    MpegEncContext * const s = &h->s;
2455

    
2456
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2457
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2458
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2459
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2460
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2461
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2462
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2463
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2464
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2465
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2466
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2467
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2468

    
2469
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2470
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2471
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2472
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2473
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2474
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2475
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2476

    
2477
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2478
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2479
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2480
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2481
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2482
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2483
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2484
}
2485

    
2486
static void free_tables(H264Context *h){
2487
    av_freep(&h->intra4x4_pred_mode);
2488
    av_freep(&h->chroma_pred_mode_table);
2489
    av_freep(&h->cbp_table);
2490
    av_freep(&h->mvd_table[0]);
2491
    av_freep(&h->mvd_table[1]);
2492
    av_freep(&h->direct_table);
2493
    av_freep(&h->non_zero_count);
2494
    av_freep(&h->slice_table_base);
2495
    av_freep(&h->top_border);
2496
    h->slice_table= NULL;
2497

    
2498
    av_freep(&h->mb2b_xy);
2499
    av_freep(&h->mb2b8_xy);
2500

    
2501
    av_freep(&h->s.obmc_scratchpad);
2502
}
2503

    
2504
/**
2505
 * allocates tables.
2506
 * needs widzh/height
2507
 */
2508
static int alloc_tables(H264Context *h){
2509
    MpegEncContext * const s = &h->s;
2510
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2511
    int x,y;
2512

    
2513
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2514

    
2515
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2516
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2517
    CHECKED_ALLOCZ(h->top_border       , s->mb_width * (16+8+8) * sizeof(uint8_t))
2518
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2519

    
2520
    if( h->pps.cabac ) {
2521
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2522
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2523
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2524
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2525
    }
2526

    
2527
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2528
    h->slice_table= h->slice_table_base + s->mb_stride + 1;
2529

    
2530
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint16_t));
2531
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2532
    for(y=0; y<s->mb_height; y++){
2533
        for(x=0; x<s->mb_width; x++){
2534
            const int mb_xy= x + y*s->mb_stride;
2535
            const int b_xy = 4*x + 4*y*h->b_stride;
2536
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2537
        
2538
            h->mb2b_xy [mb_xy]= b_xy;
2539
            h->mb2b8_xy[mb_xy]= b8_xy;
2540
        }
2541
    }
2542

    
2543
    s->obmc_scratchpad = NULL;
2544

    
2545
    return 0;
2546
fail:
2547
    free_tables(h);
2548
    return -1;
2549
}
2550

    
2551
static void common_init(H264Context *h){
2552
    MpegEncContext * const s = &h->s;
2553

    
2554
    s->width = s->avctx->width;
2555
    s->height = s->avctx->height;
2556
    s->codec_id= s->avctx->codec->id;
2557
    
2558
    init_pred_ptrs(h);
2559

    
2560
    s->unrestricted_mv=1;
2561
    s->decode=1; //FIXME
2562
}
2563

    
2564
static int decode_init(AVCodecContext *avctx){
2565
    H264Context *h= avctx->priv_data;
2566
    MpegEncContext * const s = &h->s;
2567

    
2568
    MPV_decode_defaults(s);
2569
    
2570
    s->avctx = avctx;
2571
    common_init(h);
2572

    
2573
    s->out_format = FMT_H264;
2574
    s->workaround_bugs= avctx->workaround_bugs;
2575

    
2576
    // set defaults
2577
//    s->decode_mb= ff_h263_decode_mb;
2578
    s->low_delay= 1;
2579
    avctx->pix_fmt= PIX_FMT_YUV420P;
2580

    
2581
    decode_init_vlc(h);
2582
    
2583
    if(avctx->codec_tag != 0x31637661 && avctx->codec_tag != 0x31435641) // avc1
2584
        h->is_avc = 0;
2585
    else {
2586
        if((avctx->extradata_size == 0) || (avctx->extradata == NULL)) {
2587
            av_log(avctx, AV_LOG_ERROR, "AVC codec requires avcC data\n");
2588
            return -1;
2589
        }
2590
        h->is_avc = 1;
2591
        h->got_avcC = 0;
2592
    }
2593

    
2594
    return 0;
2595
}
2596

    
2597
static void frame_start(H264Context *h){
2598
    MpegEncContext * const s = &h->s;
2599
    int i;
2600

    
2601
    MPV_frame_start(s, s->avctx);
2602
    ff_er_frame_start(s);
2603
    h->mmco_index=0;
2604

    
2605
    assert(s->linesize && s->uvlinesize);
2606

    
2607
    for(i=0; i<16; i++){
2608
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2609
        h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2610
    }
2611
    for(i=0; i<4; i++){
2612
        h->block_offset[16+i]=
2613
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2614
    }
2615

    
2616
    /* can't be in alloc_tables because linesize isn't known there.
2617
     * FIXME: redo bipred weight to not require extra buffer? */
2618
    if(!s->obmc_scratchpad)
2619
        s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
2620

    
2621
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2622
}
2623

    
2624
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2625
    MpegEncContext * const s = &h->s;
2626
    int i;
2627
    
2628
    src_y  -=   linesize;
2629
    src_cb -= uvlinesize;
2630
    src_cr -= uvlinesize;
2631

    
2632
    h->left_border[0]= h->top_border[s->mb_x][15];
2633
    for(i=1; i<17; i++){
2634
        h->left_border[i]= src_y[15+i*  linesize];
2635
    }
2636
    
2637
    *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2638
    *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2639

    
2640
    if(!(s->flags&CODEC_FLAG_GRAY)){
2641
        h->left_border[17  ]= h->top_border[s->mb_x][16+7];
2642
        h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2643
        for(i=1; i<9; i++){
2644
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
2645
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2646
        }
2647
        *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2648
        *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2649
    }
2650
}
2651

    
2652
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){
2653
    MpegEncContext * const s = &h->s;
2654
    int temp8, i;
2655
    uint64_t temp64;
2656
    int deblock_left = (s->mb_x > 0);
2657
    int deblock_top  = (s->mb_y > 0);
2658

    
2659
    src_y  -=   linesize + 1;
2660
    src_cb -= uvlinesize + 1;
2661
    src_cr -= uvlinesize + 1;
2662

    
2663
#define XCHG(a,b,t,xchg)\
2664
t= a;\
2665
if(xchg)\
2666
    a= b;\
2667
b= t;
2668

    
2669
    if(deblock_left){
2670
        for(i = !deblock_top; i<17; i++){
2671
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2672
        }
2673
    }
2674

    
2675
    if(deblock_top){
2676
        XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2677
        XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2678
    }
2679

    
2680
    if(!(s->flags&CODEC_FLAG_GRAY)){
2681
        if(deblock_left){
2682
            for(i = !deblock_top; i<9; i++){
2683
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
2684
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2685
            }
2686
        }
2687
        if(deblock_top){
2688
            XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2689
            XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2690
        }
2691
    }
2692
}
2693

    
2694
static void hl_decode_mb(H264Context *h){
2695
    MpegEncContext * const s = &h->s;
2696
    const int mb_x= s->mb_x;
2697
    const int mb_y= s->mb_y;
2698
    const int mb_xy= mb_x + mb_y*s->mb_stride;
2699
    const int mb_type= s->current_picture.mb_type[mb_xy];
2700
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2701
    int linesize, uvlinesize /*dct_offset*/;
2702
    int i;
2703

    
2704
    if(!s->decode)
2705
        return;
2706

    
2707
    if(s->mb_skiped){
2708
    }
2709

    
2710
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2711
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2712
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2713

    
2714
    if (h->mb_field_decoding_flag) {
2715
        linesize = s->linesize * 2;
2716
        uvlinesize = s->uvlinesize * 2;
2717
        if(mb_y&1){ //FIXME move out of this func?
2718
            dest_y -= s->linesize*15;
2719
            dest_cb-= s->linesize*7;
2720
            dest_cr-= s->linesize*7;
2721
        }
2722
    } else {
2723
        linesize = s->linesize;
2724
        uvlinesize = s->uvlinesize;
2725
//        dct_offset = s->linesize * 16;
2726
    }
2727

    
2728
    if(IS_INTRA(mb_type)){
2729
        if(h->deblocking_filter)
2730
            xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2731

    
2732
        if(!(s->flags&CODEC_FLAG_GRAY)){
2733
            h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2734
            h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2735
        }
2736

    
2737
        if(IS_INTRA4x4(mb_type)){
2738
            if(!s->encoding){
2739
                for(i=0; i<16; i++){
2740
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2741
                    uint8_t *topright;
2742
                    const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2743
                    int tr;
2744

    
2745
                    if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2746
                        const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2747
                        assert(mb_y || linesize <= h->block_offset[i]);
2748
                        if(!topright_avail){
2749
                            tr= ptr[3 - linesize]*0x01010101;
2750
                            topright= (uint8_t*) &tr;
2751
                        }else if(i==5 && h->deblocking_filter){
2752
                            tr= *(uint32_t*)h->top_border[mb_x+1];
2753
                            topright= (uint8_t*) &tr;
2754
                        }else
2755
                            topright= ptr + 4 - linesize;
2756
                    }else
2757
                        topright= NULL;
2758

    
2759
                    h->pred4x4[ dir ](ptr, topright, linesize);
2760
                    if(h->non_zero_count_cache[ scan8[i] ]){
2761
                        if(s->codec_id == CODEC_ID_H264)
2762
                            s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2763
                        else
2764
                            svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2765
                    }
2766
                }
2767
            }
2768
        }else{
2769
            h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2770
            if(s->codec_id == CODEC_ID_H264)
2771
                h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2772
            else
2773
                svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2774
        }
2775
        if(h->deblocking_filter)
2776
            xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2777
    }else if(s->codec_id == CODEC_ID_H264){
2778
        hl_motion(h, dest_y, dest_cb, dest_cr,
2779
                  s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
2780
                  s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
2781
                  s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2782
    }
2783

    
2784

    
2785
    if(!IS_INTRA4x4(mb_type)){
2786
        if(s->codec_id == CODEC_ID_H264){
2787
            for(i=0; i<16; i++){
2788
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2789
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2790
                    s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2791
                }
2792
            }
2793
        }else{
2794
            for(i=0; i<16; i++){
2795
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2796
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2797
                    svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2798
                }
2799
            }
2800
        }
2801
    }
2802

    
2803
    if(!(s->flags&CODEC_FLAG_GRAY)){
2804
        chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2805
        chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2806
        if(s->codec_id == CODEC_ID_H264){
2807
            for(i=16; i<16+4; i++){
2808
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2809
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2810
                    s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2811
                }
2812
            }
2813
            for(i=20; i<20+4; i++){
2814
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2815
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2816
                    s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2817
                }
2818
            }
2819
        }else{
2820
            for(i=16; i<16+4; i++){
2821
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2822
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2823
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2824
                }
2825
            }
2826
            for(i=20; i<20+4; i++){
2827
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2828
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2829
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2830
                }
2831
            }
2832
        }
2833
    }
2834
    if(h->deblocking_filter) {
2835
        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2836
        filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2837
    }
2838
}
2839

    
2840
/**
2841
 * fills the default_ref_list.
2842
 */
2843
static int fill_default_ref_list(H264Context *h){
2844
    MpegEncContext * const s = &h->s;
2845
    int i;
2846
    Picture sorted_short_ref[16];
2847
    
2848
    if(h->slice_type==B_TYPE){
2849
        int out_i;
2850
        int limit= -1;
2851

    
2852
        for(out_i=0; out_i<h->short_ref_count; out_i++){
2853
            int best_i=-1;
2854
            int best_poc=INT_MAX;
2855

    
2856
            for(i=0; i<h->short_ref_count; i++){
2857
                const int poc= h->short_ref[i]->poc;
2858
                if(poc > limit && poc < best_poc){
2859
                    best_poc= poc;
2860
                    best_i= i;
2861
                }
2862
            }
2863
            
2864
            assert(best_i != -1);
2865
            
2866
            limit= best_poc;
2867
            sorted_short_ref[out_i]= *h->short_ref[best_i];
2868
        }
2869
    }
2870

    
2871
    if(s->picture_structure == PICT_FRAME){
2872
        if(h->slice_type==B_TYPE){
2873
            const int current_poc= s->current_picture_ptr->poc;
2874
            int list;
2875

    
2876
            for(list=0; list<2; list++){
2877
                int index=0;
2878

    
2879
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2880
                    const int i2= list ? i : h->short_ref_count - i - 1;
2881
                    const int poc= sorted_short_ref[i2].poc;
2882
                    
2883
                    if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2884

    
2885
                    if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2886
                        h->default_ref_list[list][index  ]= sorted_short_ref[i2];
2887
                        h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2888
                    }
2889
                }
2890

    
2891
                for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2892
                    if(h->long_ref[i]->reference != 3) continue;
2893

    
2894
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
2895
                    h->default_ref_list[ list ][index++].pic_id= i;;
2896
                }
2897
                
2898
                if(h->long_ref_count > 1 && h->short_ref_count==0){
2899
                    Picture temp= h->default_ref_list[1][0];
2900
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
2901
                    h->default_ref_list[1][0] = temp;
2902
                }
2903

    
2904
                if(index < h->ref_count[ list ])
2905
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2906
            }
2907
        }else{
2908
            int index=0;
2909
            for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2910
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2911
                h->default_ref_list[0][index  ]= *h->short_ref[i];
2912
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2913
            }
2914
            for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2915
                if(h->long_ref[i]->reference != 3) continue;
2916
                h->default_ref_list[0][index  ]= *h->long_ref[i];
2917
                h->default_ref_list[0][index++].pic_id= i;;
2918
            }
2919
            if(index < h->ref_count[0])
2920
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2921
        }
2922
    }else{ //FIELD
2923
        if(h->slice_type==B_TYPE){
2924
        }else{
2925
            //FIXME second field balh
2926
        }
2927
    }
2928
    return 0;
2929
}
2930

    
2931
static int decode_ref_pic_list_reordering(H264Context *h){
2932
    MpegEncContext * const s = &h->s;
2933
    int list;
2934
    
2935
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2936
    
2937
    for(list=0; list<2; list++){
2938
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2939

    
2940
        if(get_bits1(&s->gb)){
2941
            int pred= h->curr_pic_num;
2942
            int index;
2943

    
2944
            for(index=0; ; index++){
2945
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2946
                int pic_id;
2947
                int i;
2948
                
2949
                if(reordering_of_pic_nums_idc==3) 
2950
                    break;
2951
                
2952
                if(index >= h->ref_count[list]){
2953
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2954
                    return -1;
2955
                }
2956
                
2957
                if(reordering_of_pic_nums_idc<3){
2958
                    if(reordering_of_pic_nums_idc<2){
2959
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2960

    
2961
                        if(abs_diff_pic_num >= h->max_pic_num){
2962
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2963
                            return -1;
2964
                        }
2965

    
2966
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2967
                        else                                pred+= abs_diff_pic_num;
2968
                        pred &= h->max_pic_num - 1;
2969
                    
2970
                        for(i= h->ref_count[list]-1; i>=index; i--){
2971
                            if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2972
                                break;
2973
                        }
2974
                    }else{
2975
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2976

    
2977
                        for(i= h->ref_count[list]-1; i>=index; i--){
2978
                            if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2979
                                break;
2980
                        }
2981
                    }
2982

    
2983
                    if(i < index){
2984
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2985
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2986
                    }else if(i > index){
2987
                        Picture tmp= h->ref_list[list][i];
2988
                        for(; i>index; i--){
2989
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2990
                        }
2991
                        h->ref_list[list][index]= tmp;
2992
                    }
2993
                }else{
2994
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2995
                    return -1;
2996
                }
2997
            }
2998
        }
2999

    
3000
        if(h->slice_type!=B_TYPE) break;
3001
    }
3002
    
3003
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3004
        direct_dist_scale_factor(h);
3005
    return 0;    
3006
}
3007

    
3008
static int pred_weight_table(H264Context *h){
3009
    MpegEncContext * const s = &h->s;
3010
    int list, i;
3011
    int luma_def, chroma_def;
3012
    
3013
    h->use_weight= 0;
3014
    h->use_weight_chroma= 0;
3015
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3016
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3017
    luma_def = 1<<h->luma_log2_weight_denom;
3018
    chroma_def = 1<<h->chroma_log2_weight_denom;
3019

    
3020
    for(list=0; list<2; list++){
3021
        for(i=0; i<h->ref_count[list]; i++){
3022
            int luma_weight_flag, chroma_weight_flag;
3023
            
3024
            luma_weight_flag= get_bits1(&s->gb);
3025
            if(luma_weight_flag){
3026
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3027
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3028
                if(   h->luma_weight[list][i] != luma_def
3029
                   || h->luma_offset[list][i] != 0)
3030
                    h->use_weight= 1;
3031
            }else{
3032
                h->luma_weight[list][i]= luma_def;
3033
                h->luma_offset[list][i]= 0;
3034
            }
3035

    
3036
            chroma_weight_flag= get_bits1(&s->gb);
3037
            if(chroma_weight_flag){
3038
                int j;
3039
                for(j=0; j<2; j++){
3040
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3041
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3042
                    if(   h->chroma_weight[list][i][j] != chroma_def
3043
                       || h->chroma_offset[list][i][j] != 0)
3044
                        h->use_weight_chroma= 1;
3045
                }
3046
            }else{
3047
                int j;
3048
                for(j=0; j<2; j++){
3049
                    h->chroma_weight[list][i][j]= chroma_def;
3050
                    h->chroma_offset[list][i][j]= 0;
3051
                }
3052
            }
3053
        }
3054
        if(h->slice_type != B_TYPE) break;
3055
    }
3056
    h->use_weight= h->use_weight || h->use_weight_chroma;
3057
    return 0;
3058
}
3059

    
3060
static void implicit_weight_table(H264Context *h){
3061
    MpegEncContext * const s = &h->s;
3062
    int list, i;
3063
    int ref0, ref1;
3064
    int cur_poc = s->current_picture_ptr->poc;
3065

    
3066
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3067
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3068
        h->use_weight= 0;
3069
        h->use_weight_chroma= 0;
3070
        return;
3071
    }
3072

    
3073
    h->use_weight= 2;
3074
    h->use_weight_chroma= 2;
3075
    h->luma_log2_weight_denom= 5;
3076
    h->chroma_log2_weight_denom= 5;
3077

    
3078
    /* FIXME: MBAFF */
3079
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3080
        int poc0 = h->ref_list[0][ref0].poc;
3081
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3082
            int poc1 = h->ref_list[0][ref1].poc;
3083
            int td = clip(poc1 - poc0, -128, 127);
3084
            if(td){
3085
                int tb = clip(cur_poc - poc0, -128, 127);
3086
                int tx = (16384 + (ABS(td) >> 1)) / td;
3087
                int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3088
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3089
                    h->implicit_weight[ref0][ref1] = 32;
3090
                else
3091
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3092
            }else
3093
                h->implicit_weight[ref0][ref1] = 32;
3094
        }
3095
    }
3096
}
3097

    
3098
/**
3099
 * instantaneous decoder refresh.
3100
 */
3101
static void idr(H264Context *h){
3102
    int i,j;
3103

    
3104
#define CHECK_DELAY(pic) \
3105
    for(j = 0; h->delayed_pic[j]; j++) \
3106
        if(pic == h->delayed_pic[j]){ \
3107
            pic->reference=1; \
3108
            break; \
3109
        }
3110

    
3111
    for(i=0; i<h->long_ref_count; i++){
3112
        h->long_ref[i]->reference=0;
3113
        CHECK_DELAY(h->long_ref[i]);
3114
        h->long_ref[i]= NULL;
3115
    }
3116
    h->long_ref_count=0;
3117

    
3118
    for(i=0; i<h->short_ref_count; i++){
3119
        h->short_ref[i]->reference=0;
3120
        CHECK_DELAY(h->short_ref[i]);
3121
        h->short_ref[i]= NULL;
3122
    }
3123
    h->short_ref_count=0;
3124
}
3125
#undef CHECK_DELAY
3126

    
3127
/**
3128
 *
3129
 * @return the removed picture or NULL if an error occures
3130
 */
3131
static Picture * remove_short(H264Context *h, int frame_num){
3132
    MpegEncContext * const s = &h->s;
3133
    int i;
3134
    
3135
    if(s->avctx->debug&FF_DEBUG_MMCO)
3136
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3137
    
3138
    for(i=0; i<h->short_ref_count; i++){
3139
        Picture *pic= h->short_ref[i];
3140
        if(s->avctx->debug&FF_DEBUG_MMCO)
3141
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3142
        if(pic->frame_num == frame_num){
3143
            h->short_ref[i]= NULL;
3144
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3145
            h->short_ref_count--;
3146
            return pic;
3147
        }
3148
    }
3149
    return NULL;
3150
}
3151

    
3152
/**
3153
 *
3154
 * @return the removed picture or NULL if an error occures
3155
 */
3156
static Picture * remove_long(H264Context *h, int i){
3157
    Picture *pic;
3158

    
3159
    if(i >= h->long_ref_count) return NULL;
3160
    pic= h->long_ref[i];
3161
    if(pic==NULL) return NULL;
3162
    
3163
    h->long_ref[i]= NULL;
3164
    memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
3165
    h->long_ref_count--;
3166

    
3167
    return pic;
3168
}
3169

    
3170
/**
3171
 * Executes the reference picture marking (memory management control operations).
3172
 */
3173
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3174
    MpegEncContext * const s = &h->s;
3175
    int i;
3176
    int current_is_long=0;
3177
    Picture *pic;
3178
    
3179
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3180
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3181
        
3182
    for(i=0; i<mmco_count; i++){
3183
        if(s->avctx->debug&FF_DEBUG_MMCO)
3184
            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);
3185

    
3186
        switch(mmco[i].opcode){
3187
        case MMCO_SHORT2UNUSED:
3188
            pic= remove_short(h, mmco[i].short_frame_num);
3189
            if(pic==NULL) return -1;
3190
            pic->reference= 0;
3191
            break;
3192
        case MMCO_SHORT2LONG:
3193
            pic= remove_long(h, mmco[i].long_index);
3194
            if(pic) pic->reference=0;
3195
            
3196
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3197
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
3198
            break;
3199
        case MMCO_LONG2UNUSED:
3200
            pic= remove_long(h, mmco[i].long_index);
3201
            if(pic==NULL) return -1;
3202
            pic->reference= 0;
3203
            break;
3204
        case MMCO_LONG:
3205
            pic= remove_long(h, mmco[i].long_index);
3206
            if(pic) pic->reference=0;
3207
            
3208
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3209
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
3210
            h->long_ref_count++;
3211
            
3212
            current_is_long=1;
3213
            break;
3214
        case MMCO_SET_MAX_LONG:
3215
            assert(mmco[i].long_index <= 16);
3216
            while(mmco[i].long_index < h->long_ref_count){
3217
                pic= remove_long(h, mmco[i].long_index);
3218
                pic->reference=0;
3219
            }
3220
            while(mmco[i].long_index > h->long_ref_count){
3221
                h->long_ref[ h->long_ref_count++ ]= NULL;
3222
            }
3223
            break;
3224
        case MMCO_RESET:
3225
            while(h->short_ref_count){
3226
                pic= remove_short(h, h->short_ref[0]->frame_num);
3227
                pic->reference=0;
3228
            }
3229
            while(h->long_ref_count){
3230
                pic= remove_long(h, h->long_ref_count-1);
3231
                pic->reference=0;
3232
            }
3233
            break;
3234
        default: assert(0);
3235
        }
3236
    }
3237
    
3238
    if(!current_is_long){
3239
        pic= remove_short(h, s->current_picture_ptr->frame_num);
3240
        if(pic){
3241
            pic->reference=0;
3242
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3243
        }
3244
        
3245
        if(h->short_ref_count)
3246
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3247

    
3248
        h->short_ref[0]= s->current_picture_ptr;
3249
        h->short_ref[0]->long_ref=0;
3250
        h->short_ref_count++;
3251
    }
3252
    
3253
    return 0; 
3254
}
3255

    
3256
static int decode_ref_pic_marking(H264Context *h){
3257
    MpegEncContext * const s = &h->s;
3258
    int i;
3259
    
3260
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3261
        s->broken_link= get_bits1(&s->gb) -1;
3262
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3263
        if(h->mmco[0].long_index == -1)
3264
            h->mmco_index= 0;
3265
        else{
3266
            h->mmco[0].opcode= MMCO_LONG;
3267
            h->mmco_index= 1;
3268
        } 
3269
    }else{
3270
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3271
            for(i= 0; i<MAX_MMCO_COUNT; i++) { 
3272
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3273

    
3274
                h->mmco[i].opcode= opcode;
3275
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3276
                    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
3277
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3278
                        fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3279
                        return -1;
3280
                    }*/
3281
                }
3282
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3283
                    h->mmco[i].long_index= get_ue_golomb(&s->gb);
3284
                    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){
3285
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3286
                        return -1;
3287
                    }
3288
                }
3289
                    
3290
                if(opcode > MMCO_LONG){
3291
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3292
                    return -1;
3293
                }
3294
                if(opcode == MMCO_END)
3295
                    break;
3296
            }
3297
            h->mmco_index= i;
3298
        }else{
3299
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3300

    
3301
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3302
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3303
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3304
                h->mmco_index= 1;
3305
            }else
3306
                h->mmco_index= 0;
3307
        }
3308
    }
3309
    
3310
    return 0; 
3311
}
3312

    
3313
static int init_poc(H264Context *h){
3314
    MpegEncContext * const s = &h->s;
3315
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3316
    int field_poc[2];
3317

    
3318
    if(h->nal_unit_type == NAL_IDR_SLICE){
3319
        h->frame_num_offset= 0;
3320
    }else{
3321
        if(h->frame_num < h->prev_frame_num)
3322
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3323
        else
3324
            h->frame_num_offset= h->prev_frame_num_offset;
3325
    }
3326

    
3327
    if(h->sps.poc_type==0){
3328
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3329

    
3330
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3331
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3332
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3333
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3334
        else
3335
            h->poc_msb = h->prev_poc_msb;
3336
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3337
        field_poc[0] = 
3338
        field_poc[1] = h->poc_msb + h->poc_lsb;
3339
        if(s->picture_structure == PICT_FRAME) 
3340
            field_poc[1] += h->delta_poc_bottom;
3341
    }else if(h->sps.poc_type==1){
3342
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3343
        int i;
3344

    
3345
        if(h->sps.poc_cycle_length != 0)
3346
            abs_frame_num = h->frame_num_offset + h->frame_num;
3347
        else
3348
            abs_frame_num = 0;
3349

    
3350
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
3351
            abs_frame_num--;
3352
            
3353
        expected_delta_per_poc_cycle = 0;
3354
        for(i=0; i < h->sps.poc_cycle_length; i++)
3355
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3356

    
3357
        if(abs_frame_num > 0){
3358
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3359
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3360

    
3361
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3362
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
3363
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3364
        } else
3365
            expectedpoc = 0;
3366

    
3367
        if(h->nal_ref_idc == 0) 
3368
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3369
        
3370
        field_poc[0] = expectedpoc + h->delta_poc[0];
3371
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3372

    
3373
        if(s->picture_structure == PICT_FRAME)
3374
            field_poc[1] += h->delta_poc[1];
3375
    }else{
3376
        int poc;
3377
        if(h->nal_unit_type == NAL_IDR_SLICE){
3378
            poc= 0;
3379
        }else{
3380
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3381
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3382
        }
3383
        field_poc[0]= poc;
3384
        field_poc[1]= poc;
3385
    }
3386
    
3387
    if(s->picture_structure != PICT_BOTTOM_FIELD)
3388
        s->current_picture_ptr->field_poc[0]= field_poc[0];
3389
    if(s->picture_structure != PICT_TOP_FIELD)
3390
        s->current_picture_ptr->field_poc[1]= field_poc[1];
3391
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3392
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3393

    
3394
    return 0;
3395
}
3396

    
3397
/**
3398
 * decodes a slice header.
3399
 * this will allso call MPV_common_init() and frame_start() as needed
3400
 */
3401
static int decode_slice_header(H264Context *h){
3402
    MpegEncContext * const s = &h->s;
3403
    int first_mb_in_slice, pps_id;
3404
    int num_ref_idx_active_override_flag;
3405
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3406

    
3407
    s->current_picture.reference= h->nal_ref_idc != 0;
3408

    
3409
    first_mb_in_slice= get_ue_golomb(&s->gb);
3410

    
3411
    h->slice_type= get_ue_golomb(&s->gb);
3412
    if(h->slice_type > 9){
3413
        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);
3414
        return -1;
3415
    }
3416
    if(h->slice_type > 4){
3417
        h->slice_type -= 5;
3418
        h->slice_type_fixed=1;
3419
    }else
3420
        h->slice_type_fixed=0;
3421
    
3422
    h->slice_type= slice_type_map[ h->slice_type ];
3423
    
3424
    s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
3425
        
3426
    pps_id= get_ue_golomb(&s->gb);
3427
    if(pps_id>255){
3428
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3429
        return -1;
3430
    }
3431
    h->pps= h->pps_buffer[pps_id];
3432
    if(h->pps.slice_group_count == 0){
3433
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3434
        return -1;
3435
    }
3436

    
3437
    h->sps= h->sps_buffer[ h->pps.sps_id ];
3438
    if(h->sps.log2_max_frame_num == 0){
3439
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3440
        return -1;
3441
    }
3442
    
3443
    s->mb_width= h->sps.mb_width;
3444
    s->mb_height= h->sps.mb_height;
3445
    
3446
    h->b_stride=  s->mb_width*4 + 1;
3447
    h->b8_stride= s->mb_width*2 + 1;
3448

    
3449
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3450
    s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
3451
    
3452
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3453
    if(h->sps.frame_mbs_only_flag)
3454
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
3455
    else
3456
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
3457
    
3458
    if (s->context_initialized 
3459
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3460
        free_tables(h);
3461
        MPV_common_end(s);
3462
    }
3463
    if (!s->context_initialized) {
3464
        if (MPV_common_init(s) < 0)
3465
            return -1;
3466

    
3467
        alloc_tables(h);
3468

    
3469
        s->avctx->width = s->width;
3470
        s->avctx->height = s->height;
3471
        s->avctx->sample_aspect_ratio= h->sps.sar;
3472
        if(!s->avctx->sample_aspect_ratio.den)
3473
            s->avctx->sample_aspect_ratio.den = 1;
3474

    
3475
        if(h->sps.timing_info_present_flag && h->sps.fixed_frame_rate_flag){
3476
            s->avctx->frame_rate = h->sps.time_scale;
3477
            s->avctx->frame_rate_base = h->sps.num_units_in_tick;
3478
        }
3479
    }
3480

    
3481
    if(h->slice_num == 0){
3482
        frame_start(h);
3483
    }
3484

    
3485
    s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3486
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3487

    
3488
    if(h->sps.frame_mbs_only_flag){
3489
        s->picture_structure= PICT_FRAME;
3490
    }else{
3491
        if(get_bits1(&s->gb)) //field_pic_flag
3492
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3493
        else
3494
            s->picture_structure= PICT_FRAME;
3495
    }
3496

    
3497
    if(s->picture_structure==PICT_FRAME){
3498
        h->curr_pic_num=   h->frame_num;
3499
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3500
    }else{
3501
        h->curr_pic_num= 2*h->frame_num;
3502
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3503
    }
3504
        
3505
    if(h->nal_unit_type == NAL_IDR_SLICE){
3506
        get_ue_golomb(&s->gb); /* idr_pic_id */
3507
    }
3508
   
3509
    if(h->sps.poc_type==0){
3510
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3511
        
3512
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3513
            h->delta_poc_bottom= get_se_golomb(&s->gb);
3514
        }
3515
    }
3516
    
3517
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3518
        h->delta_poc[0]= get_se_golomb(&s->gb);
3519
        
3520
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3521
            h->delta_poc[1]= get_se_golomb(&s->gb);
3522
    }
3523
    
3524
    init_poc(h);
3525
    
3526
    if(h->pps.redundant_pic_cnt_present){
3527
        h->redundant_pic_count= get_ue_golomb(&s->gb);
3528
    }
3529

    
3530
    //set defaults, might be overriden a few line later
3531
    h->ref_count[0]= h->pps.ref_count[0];
3532
    h->ref_count[1]= h->pps.ref_count[1];
3533

    
3534
    if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3535
        if(h->slice_type == B_TYPE){
3536
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
3537
        }
3538
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
3539
    
3540
        if(num_ref_idx_active_override_flag){
3541
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3542
            if(h->slice_type==B_TYPE)
3543
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3544

    
3545
            if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3546
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3547
                return -1;
3548
            }
3549
        }
3550
    }
3551

    
3552
    if(h->slice_num == 0){
3553
        fill_default_ref_list(h);
3554
    }
3555

    
3556
    decode_ref_pic_list_reordering(h);
3557

    
3558
    if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) 
3559
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3560
        pred_weight_table(h);
3561
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
3562
        implicit_weight_table(h);
3563
    else
3564
        h->use_weight = 0;
3565
    
3566
    if(s->current_picture.reference)
3567
        decode_ref_pic_marking(h);
3568

    
3569
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3570
        h->cabac_init_idc = get_ue_golomb(&s->gb);
3571

    
3572
    h->last_qscale_diff = 0;
3573
    s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3574
    if(s->qscale<0 || s->qscale>51){
3575
        av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3576
        return -1;
3577
    }
3578
    h->chroma_qp = get_chroma_qp(h, s->qscale);
3579
    //FIXME qscale / qp ... stuff
3580
    if(h->slice_type == SP_TYPE){
3581
        get_bits1(&s->gb); /* sp_for_switch_flag */
3582
    }
3583
    if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3584
        get_se_golomb(&s->gb); /* slice_qs_delta */
3585
    }
3586

    
3587
    h->deblocking_filter = 1;
3588
    h->slice_alpha_c0_offset = 0;
3589
    h->slice_beta_offset = 0;
3590
    if( h->pps.deblocking_filter_parameters_present ) {
3591
        h->deblocking_filter= get_ue_golomb(&s->gb);
3592
        if(h->deblocking_filter < 2) 
3593
            h->deblocking_filter^= 1; // 1<->0
3594

    
3595
        if( h->deblocking_filter ) {
3596
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3597
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3598
        }
3599
    }
3600

    
3601
#if 0 //FMO
3602
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3603
        slice_group_change_cycle= get_bits(&s->gb, ?);
3604
#endif
3605

    
3606
    h->slice_num++;
3607

    
3608
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3609
        av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d weight:%d%s\n", 
3610
               h->slice_num, first_mb_in_slice, 
3611
               av_get_pict_type_char(h->slice_type),
3612
               pps_id, h->frame_num,
3613
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3614
               h->ref_count[0], h->ref_count[1],
3615
               s->qscale,
3616
               h->deblocking_filter,
3617
               h->use_weight,
3618
               h->use_weight==1 && h->use_weight_chroma ? "c" : ""
3619
               );
3620
    }
3621

    
3622
    return 0;
3623
}
3624

    
3625
/**
3626
 *
3627
 */
3628
static inline int get_level_prefix(GetBitContext *gb){
3629
    unsigned int buf;
3630
    int log;
3631
    
3632
    OPEN_READER(re, gb);
3633
    UPDATE_CACHE(re, gb);
3634
    buf=GET_CACHE(re, gb);
3635
    
3636
    log= 32 - av_log2(buf);
3637
#ifdef TRACE
3638
    print_bin(buf>>(32-log), log);
3639
    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__);
3640
#endif
3641

    
3642
    LAST_SKIP_BITS(re, gb, log);
3643
    CLOSE_READER(re, gb);
3644

    
3645
    return log-1;
3646
}
3647

    
3648
/**
3649
 * decodes a residual block.
3650
 * @param n block index
3651
 * @param scantable scantable
3652
 * @param max_coeff number of coefficients in the block
3653
 * @return <0 if an error occured
3654
 */
3655
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3656
    MpegEncContext * const s = &h->s;
3657
    const uint16_t *qmul= dequant_coeff[qp];
3658
    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};
3659
    int level[16], run[16];
3660
    int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3661

    
3662
    //FIXME put trailing_onex into the context
3663

    
3664
    if(n == CHROMA_DC_BLOCK_INDEX){
3665
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3666
        total_coeff= coeff_token>>2;
3667
    }else{    
3668
        if(n == LUMA_DC_BLOCK_INDEX){
3669
            total_coeff= pred_non_zero_count(h, 0);
3670
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3671
            total_coeff= coeff_token>>2;
3672
        }else{
3673
            total_coeff= pred_non_zero_count(h, n);
3674
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3675
            total_coeff= coeff_token>>2;
3676
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3677
        }
3678
    }
3679

    
3680
    //FIXME set last_non_zero?
3681

    
3682
    if(total_coeff==0)
3683
        return 0;
3684
        
3685
    trailing_ones= coeff_token&3;
3686
    tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3687
    assert(total_coeff<=16);
3688
    
3689
    for(i=0; i<trailing_ones; i++){
3690
        level[i]= 1 - 2*get_bits1(gb);
3691
    }
3692

    
3693
    suffix_length= total_coeff > 10 && trailing_ones < 3;
3694

    
3695
    for(; i<total_coeff; i++){
3696
        const int prefix= get_level_prefix(gb);
3697
        int level_code, mask;
3698

    
3699
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3700
            if(suffix_length)
3701
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3702
            else
3703
                level_code= (prefix<<suffix_length); //part
3704
        }else if(prefix==14){
3705
            if(suffix_length)
3706
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3707
            else
3708
                level_code= prefix + get_bits(gb, 4); //part
3709
        }else if(prefix==15){
3710
            level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3711
            if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3712
        }else{
3713
            av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3714
            return -1;
3715
        }
3716

    
3717
        if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3718

    
3719
        mask= -(level_code&1);
3720
        level[i]= (((2+level_code)>>1) ^ mask) - mask;
3721

    
3722
        if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3723

    
3724
#if 1
3725
        if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3726
#else        
3727
        if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3728
        /* ? == prefix > 2 or sth */
3729
#endif
3730
        tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3731
    }
3732

    
3733
    if(total_coeff == max_coeff)
3734
        zeros_left=0;
3735
    else{
3736
        if(n == CHROMA_DC_BLOCK_INDEX)
3737
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3738
        else
3739
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3740
    }
3741
    
3742
    for(i=0; i<total_coeff-1; i++){
3743
        if(zeros_left <=0)
3744
            break;
3745
        else if(zeros_left < 7){
3746
            run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3747
        }else{
3748
            run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3749
        }
3750
        zeros_left -= run[i];
3751
    }
3752

    
3753
    if(zeros_left<0){
3754
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3755
        return -1;
3756
    }
3757
    
3758
    for(; i<total_coeff-1; i++){
3759
        run[i]= 0;
3760
    }
3761

    
3762
    run[i]= zeros_left;
3763

    
3764
    coeff_num=-1;
3765
    if(n > 24){
3766
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3767
            int j;
3768

    
3769
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3770
            j= scantable[ coeff_num ];
3771

    
3772
            block[j]= level[i];
3773
        }
3774
    }else{
3775
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into  rundecode?
3776
            int j;
3777

    
3778
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3779
            j= scantable[ coeff_num ];
3780

    
3781
            block[j]= level[i] * qmul[j];
3782
//            printf("%d %d  ", block[j], qmul[j]);
3783
        }
3784
    }
3785
    return 0;
3786
}
3787

    
3788
/**
3789
 * decodes a P_SKIP or B_SKIP macroblock
3790
 */
3791
static void decode_mb_skip(H264Context *h){
3792
    MpegEncContext * const s = &h->s;
3793
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3794
    int mb_type;
3795
    
3796
    memset(h->non_zero_count[mb_xy], 0, 16);
3797
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3798

    
3799
    if( h->slice_type == B_TYPE )
3800
    {
3801
        // just for fill_caches. pred_direct_motion will set the real mb_type
3802
        mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
3803
        //FIXME mbaff
3804

    
3805
        fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3806
        pred_direct_motion(h, &mb_type);
3807
        if(h->pps.cabac){
3808
            fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
3809
            fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
3810
        }
3811
    }
3812
    else
3813
    {
3814
        int mx, my;
3815
        mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
3816

    
3817
        if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3818
            h->mb_field_decoding_flag= get_bits1(&s->gb);
3819
        }
3820
        if(h->mb_field_decoding_flag)
3821
            mb_type|= MB_TYPE_INTERLACED;
3822
        
3823
        fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3824
        pred_pskip_motion(h, &mx, &my);
3825
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3826
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3827
        if(h->pps.cabac)
3828
            fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
3829
    }
3830

    
3831
    write_back_motion(h, mb_type);
3832
    s->current_picture.mb_type[mb_xy]= mb_type|MB_TYPE_SKIP;
3833
    s->current_picture.qscale_table[mb_xy]= s->qscale;
3834
    h->slice_table[ mb_xy ]= h->slice_num;
3835
    h->prev_mb_skiped= 1;
3836
}
3837

    
3838
/**
3839
 * decodes a macroblock
3840
 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3841
 */
3842
static int decode_mb_cavlc(H264Context *h){
3843
    MpegEncContext * const s = &h->s;
3844
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3845
    int mb_type, partition_count, cbp;
3846

    
3847
    s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?    
3848

    
3849
    tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3850
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3851
                down the code */
3852
    if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3853
        if(s->mb_skip_run==-1)
3854
            s->mb_skip_run= get_ue_golomb(&s->gb);
3855
        
3856
        if (s->mb_skip_run--) {
3857
            decode_mb_skip(h);
3858
            return 0;
3859
        }
3860
    }
3861
    if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3862
        if((s->mb_y&1)==0)
3863
            h->mb_field_decoding_flag = get_bits1(&s->gb);
3864
    }else
3865
        h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3866
    
3867
    h->prev_mb_skiped= 0;
3868
    
3869
    mb_type= get_ue_golomb(&s->gb);
3870
    if(h->slice_type == B_TYPE){
3871
        if(mb_type < 23){
3872
            partition_count= b_mb_type_info[mb_type].partition_count;
3873
            mb_type=         b_mb_type_info[mb_type].type;
3874
        }else{
3875
            mb_type -= 23;
3876
            goto decode_intra_mb;
3877
        }
3878
    }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3879
        if(mb_type < 5){
3880
            partition_count= p_mb_type_info[mb_type].partition_count;
3881
            mb_type=         p_mb_type_info[mb_type].type;
3882
        }else{
3883
            mb_type -= 5;
3884
            goto decode_intra_mb;
3885
        }
3886
    }else{
3887
       assert(h->slice_type == I_TYPE);
3888
decode_intra_mb:
3889
        if(mb_type > 25){
3890
            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);
3891
            return -1;
3892
        }
3893
        partition_count=0;
3894
        cbp= i_mb_type_info[mb_type].cbp;
3895
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3896
        mb_type= i_mb_type_info[mb_type].type;
3897
    }
3898

    
3899
    if(h->mb_field_decoding_flag)
3900
        mb_type |= MB_TYPE_INTERLACED;
3901

    
3902
    s->current_picture.mb_type[mb_xy]= mb_type;
3903
    h->slice_table[ mb_xy ]= h->slice_num;
3904
    
3905
    if(IS_INTRA_PCM(mb_type)){
3906
        const uint8_t *ptr;
3907
        int x, y;
3908
        
3909
        // we assume these blocks are very rare so we dont optimize it
3910
        align_get_bits(&s->gb);
3911
        
3912
        ptr= s->gb.buffer + get_bits_count(&s->gb);
3913
    
3914
        for(y=0; y<16; y++){
3915
            const int index= 4*(y&3) + 64*(y>>2);
3916
            for(x=0; x<16; x++){
3917
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3918
            }
3919
        }
3920
        for(y=0; y<8; y++){
3921
            const int index= 256 + 4*(y&3) + 32*(y>>2);
3922
            for(x=0; x<8; x++){
3923
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3924
            }
3925
        }
3926
        for(y=0; y<8; y++){
3927
            const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3928
            for(x=0; x<8; x++){
3929
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3930
            }
3931
        }
3932
    
3933
        skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3934
        
3935
        //FIXME deblock filter, non_zero_count_cache init ...
3936
        memset(h->non_zero_count[mb_xy], 16, 16);
3937
        s->current_picture.qscale_table[mb_xy]= s->qscale;
3938
        
3939
        return 0;
3940
    }
3941
        
3942
    fill_caches(h, mb_type);
3943

    
3944
    //mb_pred
3945
    if(IS_INTRA(mb_type)){
3946
//            init_top_left_availability(h);
3947
            if(IS_INTRA4x4(mb_type)){
3948
                int i;
3949

    
3950
//                fill_intra4x4_pred_table(h);
3951
                for(i=0; i<16; i++){
3952
                    const int mode_coded= !get_bits1(&s->gb);
3953
                    const int predicted_mode=  pred_intra_mode(h, i);
3954
                    int mode;
3955

    
3956
                    if(mode_coded){
3957
                        const int rem_mode= get_bits(&s->gb, 3);
3958
                        if(rem_mode<predicted_mode)
3959
                            mode= rem_mode;
3960
                        else
3961
                            mode= rem_mode + 1;
3962
                    }else{
3963
                        mode= predicted_mode;
3964
                    }
3965
                    
3966
                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3967
                }
3968
                write_back_intra_pred_mode(h);
3969
                if( check_intra4x4_pred_mode(h) < 0)
3970
                    return -1;
3971
            }else{
3972
                h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3973
                if(h->intra16x16_pred_mode < 0)
3974
                    return -1;
3975
            }
3976
            h->chroma_pred_mode= get_ue_golomb(&s->gb);
3977

    
3978
            h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3979
            if(h->chroma_pred_mode < 0)
3980
                return -1;
3981
    }else if(partition_count==4){
3982
        int i, j, sub_partition_count[4], list, ref[2][4];
3983
        
3984
        if(h->slice_type == B_TYPE){
3985
            for(i=0; i<4; i++){
3986
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3987
                if(h->sub_mb_type[i] >=13){
3988
                    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);
3989
                    return -1;
3990
                }
3991
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3992
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3993
            }
3994
            if(   IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
3995
               || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3]))
3996
                pred_direct_motion(h, &mb_type);
3997
        }else{
3998
            assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3999
            for(i=0; i<4; i++){
4000
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4001
                if(h->sub_mb_type[i] >=4){
4002
                    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);
4003
                    return -1;
4004
                }
4005
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4006
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4007
            }
4008
        }
4009
        
4010
        for(list=0; list<2; list++){
4011
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4012
            if(ref_count == 0) continue;
4013
            for(i=0; i<4; i++){
4014
                if(IS_DIRECT(h->sub_mb_type[i])) continue;
4015
                if(IS_DIR(h->sub_mb_type[i], 0, list)){
4016
                    ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
4017
                }else{
4018
                 //FIXME
4019
                    ref[list][i] = -1;
4020
                }
4021
            }
4022
        }
4023
        
4024
        for(list=0; list<2; list++){
4025
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4026
            if(ref_count == 0) continue;
4027

    
4028
            for(i=0; i<4; i++){
4029
                if(IS_DIRECT(h->sub_mb_type[i])) continue;
4030
                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4031
                h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4032

    
4033
                if(IS_DIR(h->sub_mb_type[i], 0, list)){
4034
                    const int sub_mb_type= h->sub_mb_type[i];
4035
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4036
                    for(j=0; j<sub_partition_count[i]; j++){
4037
                        int mx, my;
4038
                        const int index= 4*i + block_width*j;
4039
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4040
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
4041
                        mx += get_se_golomb(&s->gb);
4042
                        my += get_se_golomb(&s->gb);
4043
                        tprintf("final mv:%d %d\n", mx, my);
4044

    
4045
                        if(IS_SUB_8X8(sub_mb_type)){
4046
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= 
4047
                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4048
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= 
4049
                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4050
                        }else if(IS_SUB_8X4(sub_mb_type)){
4051
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4052
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4053
                        }else if(IS_SUB_4X8(sub_mb_type)){
4054
                            mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4055
                            mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4056
                        }else{
4057
                            assert(IS_SUB_4X4(sub_mb_type));
4058
                            mv_cache[ 0 ][0]= mx;
4059
                            mv_cache[ 0 ][1]= my;
4060
                        }
4061
                    }
4062
                }else{
4063
                    uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4064
                    p[0] = p[1]=
4065
                    p[8] = p[9]= 0;
4066
                }
4067
            }
4068
        }
4069
    }else if(IS_DIRECT(mb_type)){
4070
        pred_direct_motion(h, &mb_type);
4071
        s->current_picture.mb_type[mb_xy]= mb_type;
4072
    }else{
4073
        int list, mx, my, i;
4074
         //FIXME we should set ref_idx_l? to 0 if we use that later ...
4075
        if(IS_16X16(mb_type)){
4076
            for(list=0; list<2; list++){
4077
                if(h->ref_count[list]>0){
4078
                    if(IS_DIR(mb_type, 0, list)){
4079
                        const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4080
                        fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
4081
                    }
4082
                }
4083
            }
4084
            for(list=0; list<2; list++){
4085
                if(IS_DIR(mb_type, 0, list)){
4086
                    pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
4087
                    mx += get_se_golomb(&s->gb);
4088
                    my += get_se_golomb(&s->gb);
4089
                    tprintf("final mv:%d %d\n", mx, my);
4090

    
4091
                    fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
4092
                }
4093
            }
4094
        }
4095
        else if(IS_16X8(mb_type)){
4096
            for(list=0; list<2; list++){
4097
                if(h->ref_count[list]>0){
4098
                    for(i=0; i<2; i++){
4099
                        if(IS_DIR(mb_type, i, list)){
4100
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4101
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
4102
                        }else // needed only for mixed refs (e.g. B_L0_L1_16x8)
4103
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
4104
                    }
4105
                }
4106
            }
4107
            for(list=0; list<2; list++){
4108
                for(i=0; i<2; i++){
4109
                    if(IS_DIR(mb_type, i, list)){
4110
                        pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
4111
                        mx += get_se_golomb(&s->gb);
4112
                        my += get_se_golomb(&s->gb);
4113
                        tprintf("final mv:%d %d\n", mx, my);
4114

    
4115
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
4116
                    }else
4117
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
4118
                }
4119
            }
4120
        }else{
4121
            assert(IS_8X16(mb_type));
4122
            for(list=0; list<2; list++){
4123
                if(h->ref_count[list]>0){
4124
                    for(i=0; i<2; i++){
4125
                        if(IS_DIR(mb_type, i, list)){ //FIXME optimize
4126
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4127
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
4128
                        }else // needed only for mixed refs
4129
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
4130
                    }
4131
                }
4132
            }
4133
            for(list=0; list<2; list++){
4134
                for(i=0; i<2; i++){
4135
                    if(IS_DIR(mb_type, i, list)){
4136
                        pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
4137
                        mx += get_se_golomb(&s->gb);
4138
                        my += get_se_golomb(&s->gb);
4139
                        tprintf("final mv:%d %d\n", mx, my);
4140

    
4141
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
4142
                    }else
4143
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
4144
                }
4145
            }
4146
        }
4147
    }
4148
    
4149
    if(IS_INTER(mb_type))
4150
        write_back_motion(h, mb_type);
4151
    
4152
    if(!IS_INTRA16x16(mb_type)){
4153
        cbp= get_ue_golomb(&s->gb);
4154
        if(cbp > 47){
4155
            av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
4156
            return -1;
4157
        }
4158
        
4159
        if(IS_INTRA4x4(mb_type))
4160
            cbp= golomb_to_intra4x4_cbp[cbp];
4161
        else
4162
            cbp= golomb_to_inter_cbp[cbp];
4163
    }
4164

    
4165
    if(cbp || IS_INTRA16x16(mb_type)){
4166
        int i8x8, i4x4, chroma_idx;
4167
        int chroma_qp, dquant;
4168
        GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
4169
        const uint8_t *scan, *dc_scan;
4170
        
4171
//        fill_non_zero_count_cache(h);
4172

    
4173
        if(IS_INTERLACED(mb_type)){
4174
            scan= field_scan;
4175
            dc_scan= luma_dc_field_scan;
4176
        }else{
4177
            scan= zigzag_scan;
4178
            dc_scan= luma_dc_zigzag_scan;
4179
        }
4180

    
4181
        dquant= get_se_golomb(&s->gb);
4182

    
4183
        if( dquant > 25 || dquant < -26 ){
4184
            av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
4185
            return -1;
4186
        }
4187
        
4188
        s->qscale += dquant;
4189
        if(((unsigned)s->qscale) > 51){
4190
            if(s->qscale<0) s->qscale+= 52;
4191
            else            s->qscale-= 52;
4192
        }
4193
        
4194
        h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
4195
        if(IS_INTRA16x16(mb_type)){
4196
            if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
4197
                return -1; //FIXME continue if partotioned and other retirn -1 too
4198
            }
4199

    
4200
            assert((cbp&15) == 0 || (cbp&15) == 15);
4201

    
4202
            if(cbp&15){
4203
                for(i8x8=0; i8x8<4; i8x8++){
4204
                    for(i4x4=0; i4x4<4; i4x4++){
4205
                        const int index= i4x4 + 4*i8x8;
4206
                        if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
4207
                            return -1;
4208
                        }
4209
                    }
4210
                }
4211
            }else{
4212
                fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
4213
            }
4214
        }else{
4215
            for(i8x8=0; i8x8<4; i8x8++){
4216
                if(cbp & (1<<i8x8)){
4217
                    for(i4x4=0; i4x4<4; i4x4++){
4218
                        const int index= i4x4 + 4*i8x8;
4219
                        
4220
                        if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
4221
                            return -1;
4222
                        }
4223
                    }
4224
                }else{
4225
                    uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4226
                    nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
4227
                }
4228
            }
4229
        }
4230
        
4231
        if(cbp&0x30){
4232
            for(chroma_idx=0; chroma_idx<2; chroma_idx++)
4233
                if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
4234
                    return -1;
4235
                }
4236
        }
4237

    
4238
        if(cbp&0x20){
4239
            for(chroma_idx=0; chroma_idx<2; chroma_idx++){
4240
                for(i4x4=0; i4x4<4; i4x4++){
4241
                    const int index= 16 + 4*chroma_idx + i4x4;
4242
                    if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
4243
                        return -1;
4244
                    }
4245
                }
4246
            }
4247
        }else{
4248
            uint8_t * const nnz= &h->non_zero_count_cache[0];
4249
            nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4250
            nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4251
        }
4252
    }else{
4253
        uint8_t * const nnz= &h->non_zero_count_cache[0];
4254
        fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
4255
        nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4256
        nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4257
    }
4258
    s->current_picture.qscale_table[mb_xy]= s->qscale;
4259
    write_back_non_zero_count(h);
4260

    
4261
    return 0;
4262
}
4263

    
4264
static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {
4265
    uint8_t *state= &h->cabac_state[ctx_base];
4266
    int mb_type;
4267
    
4268
    if(intra_slice){
4269
        MpegEncContext * const s = &h->s;
4270
        const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4271
        int ctx=0;
4272
        if( s->mb_x > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-1] ) )
4273
            ctx++;
4274
        if( s->mb_y > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
4275
            ctx++;
4276
        if( get_cabac( &h->cabac, &state[ctx] ) == 0 )
4277
            return 0;   /* I4x4 */
4278
        state += 2;
4279
    }else{
4280
        if( get_cabac( &h->cabac, &state[0] ) == 0 )
4281
            return 0;   /* I4x4 */
4282
    }
4283

    
4284
    if( get_cabac_terminate( &h->cabac ) )
4285
        return 25;  /* PCM */
4286

    
4287
    mb_type = 1; /* I16x16 */
4288
    if( get_cabac( &h->cabac, &state[1] ) )
4289
        mb_type += 12;  /* cbp_luma != 0 */
4290

    
4291
    if( get_cabac( &h->cabac, &state[2] ) ) {
4292
        if( get_cabac( &h->cabac, &state[2+intra_slice] ) )
4293
            mb_type += 4 * 2;   /* cbp_chroma == 2 */
4294
        else
4295
            mb_type += 4 * 1;   /* cbp_chroma == 1 */
4296
    }
4297
    if( get_cabac( &h->cabac, &state[3+intra_slice] ) )
4298
        mb_type += 2;
4299
    if( get_cabac( &h->cabac, &state[3+2*intra_slice] ) )
4300
        mb_type += 1;
4301
    return mb_type;
4302
}
4303

    
4304
static int decode_cabac_mb_type( H264Context *h ) {
4305
    MpegEncContext * const s = &h->s;
4306

    
4307
    if( h->slice_type == I_TYPE ) {
4308
        return decode_cabac_intra_mb_type(h, 3, 1);
4309
    } else if( h->slice_type == P_TYPE ) {
4310
        if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
4311
            /* P-type */
4312
            if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
4313
                if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
4314
                    return 0; /* P_L0_D16x16; */
4315
                else
4316
                    return 3; /* P_8x8; */
4317
            } else {
4318
                if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
4319
                    return 2; /* P_L0_D8x16; */
4320
                else
4321
                    return 1; /* P_L0_D16x8; */
4322
            }
4323
        } else {
4324
            return decode_cabac_intra_mb_type(h, 17, 0) + 5;
4325
        }
4326
    } else if( h->slice_type == B_TYPE ) {
4327
        const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4328
        int ctx = 0;
4329
        int bits;
4330

    
4331
        if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mb_xy-1] )
4332
                      && !IS_DIRECT( s->current_picture.mb_type[mb_xy-1] ) )
4333
            ctx++;
4334
        if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mb_xy-s->mb_stride] )
4335
                      && !IS_DIRECT( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
4336
            ctx++;
4337

    
4338
        if( !get_cabac( &h->cabac, &h->cabac_state[27+ctx] ) )
4339
            return 0; /* B_Direct_16x16 */
4340

    
4341
        if( !get_cabac( &h->cabac, &h->cabac_state[27+3] ) ) {
4342
            return 1 + get_cabac( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */
4343
        }
4344

    
4345
        bits = get_cabac( &h->cabac, &h->cabac_state[27+4] ) << 3;
4346
        bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 2;
4347
        bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 1;
4348
        bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] );
4349
        if( bits < 8 )
4350
            return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
4351
        else if( bits == 13 ) {
4352
            return decode_cabac_intra_mb_type(h, 32, 0) + 23;
4353
        } else if( bits == 14 )
4354
            return 11; /* B_L1_L0_8x16 */
4355
        else if( bits == 15 )
4356
            return 22; /* B_8x8 */
4357

    
4358
        bits= ( bits<<1 ) | get_cabac( &h->cabac, &h->cabac_state[27+5] );
4359
        return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */
4360
    } else {
4361
        /* TODO SI/SP frames? */
4362
        return -1;
4363
    }
4364
}
4365

    
4366
static int decode_cabac_mb_skip( H264Context *h) {
4367
    MpegEncContext * const s = &h->s;
4368
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4369
    const int mba_xy = mb_xy - 1;
4370
    const int mbb_xy = mb_xy - s->mb_stride;
4371
    int ctx = 0;
4372

    
4373
    if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mba_xy] ) )
4374
        ctx++;
4375
    if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ) )
4376
        ctx++;
4377

    
4378
    if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
4379
        return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
4380
    else /* B-frame */
4381
        return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
4382
}
4383

    
4384
static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
4385
    int mode = 0;
4386

    
4387
    if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
4388
        return pred_mode;
4389

    
4390
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4391
        mode += 1;
4392
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4393
        mode += 2;
4394
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4395
        mode += 4;
4396
    if( mode >= pred_mode )
4397
        return mode + 1;
4398
    else
4399
        return mode;
4400
}
4401

    
4402
static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
4403
    MpegEncContext * const s = &h->s;
4404
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4405
    const int mba_xy = mb_xy - 1;
4406
    const int mbb_xy = mb_xy - s->mb_stride;
4407

    
4408
    int ctx = 0;
4409

    
4410
    /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
4411
    if( s->mb_x > 0 && h->chroma_pred_mode_table[mba_xy] != 0 )
4412
        ctx++;
4413

    
4414
    if( s->mb_y > 0 && h->chroma_pred_mode_table[mbb_xy] != 0 )
4415
        ctx++;
4416

    
4417
    if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
4418
        return 0;
4419

    
4420
    if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4421
        return 1;
4422
    if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4423
        return 2;
4424
    else
4425
        return 3;
4426
}
4427

    
4428
static const uint8_t block_idx_x[16] = {
4429
    0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
4430
};
4431
static const uint8_t block_idx_y[16] = {
4432
    0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
4433
};
4434
static const uint8_t block_idx_xy[4][4] = {
4435
    { 0, 2, 8,  10},
4436
    { 1, 3, 9,  11},
4437
    { 4, 6, 12, 14},
4438
    { 5, 7, 13, 15}
4439
};
4440

    
4441
static int decode_cabac_mb_cbp_luma( H264Context *h) {
4442
    MpegEncContext * const s = &h->s;
4443
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4444

    
4445
    int cbp = 0;
4446
    int i8x8;
4447

    
4448
    h->cbp_table[mb_xy] = 0;  /* FIXME aaahahahah beurk */
4449

    
4450
    for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
4451
        int mba_xy = -1;
4452
        int mbb_xy = -1;
4453
        int x, y;
4454
        int ctx = 0;
4455

    
4456
        x = block_idx_x[4*i8x8];
4457
        y = block_idx_y[4*i8x8];
4458

    
4459
        if( x > 0 )
4460
            mba_xy = mb_xy;
4461
        else if( s->mb_x > 0 )
4462
            mba_xy = mb_xy - 1;
4463

    
4464
        if( y > 0 )
4465
            mbb_xy = mb_xy;
4466
        else if( s->mb_y > 0 )
4467
            mbb_xy = mb_xy - s->mb_stride;
4468

    
4469
        /* No need to test for skip as we put 0 for skip block */
4470
        if( mba_xy >= 0 ) {
4471
            int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4472
            if( ((h->cbp_table[mba_xy] >> i8x8a)&0x01) == 0 )
4473
                ctx++;
4474
        }
4475

    
4476
        if( mbb_xy >= 0 ) {
4477
            int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4478
            if( ((h->cbp_table[mbb_xy] >> i8x8b)&0x01) == 0 )
4479
                ctx += 2;
4480
        }
4481

    
4482
        if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
4483
            cbp |= 1 << i8x8;
4484
            h->cbp_table[mb_xy] = cbp;  /* FIXME aaahahahah beurk */
4485
        }
4486
    }
4487
    return cbp;
4488
}
4489