Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 693399ba

History | View | Annotate | Download (206 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
    short offset_for_ref_frame[256]; //FIXME dyn aloc?
86
}SPS;
87

    
88
/**
89
 * Picture parameter set
90
 */
91
typedef struct PPS{
92
    int sps_id;
93
    int cabac;                  ///< entropy_coding_mode_flag
94
    int pic_order_present;      ///< pic_order_present_flag
95
    int slice_group_count;      ///< num_slice_groups_minus1 + 1
96
    int mb_slice_group_map_type;
97
    int ref_count[2];           ///< num_ref_idx_l0/1_active_minus1 + 1
98
    int weighted_pred;          ///< weighted_pred_flag
99
    int weighted_bipred_idc;
100
    int init_qp;                ///< pic_init_qp_minus26 + 26
101
    int init_qs;                ///< pic_init_qs_minus26 + 26
102
    int chroma_qp_index_offset;
103
    int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
104
    int constrained_intra_pred; ///< constrained_intra_pred_flag
105
    int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
106
}PPS;
107

    
108
/**
109
 * Memory management control operation opcode.
110
 */
111
typedef enum MMCOOpcode{
112
    MMCO_END=0,
113
    MMCO_SHORT2UNUSED,
114
    MMCO_LONG2UNUSED,
115
    MMCO_SHORT2LONG,
116
    MMCO_SET_MAX_LONG,
117
    MMCO_RESET, 
118
    MMCO_LONG,
119
} MMCOOpcode;
120

    
121
/**
122
 * Memory management control operation.
123
 */
124
typedef struct MMCO{
125
    MMCOOpcode opcode;
126
    int short_frame_num;
127
    int long_index;
128
} MMCO;
129

    
130
/**
131
 * H264Context
132
 */
133
typedef struct H264Context{
134
    MpegEncContext s;
135
    int nal_ref_idc;        
136
    int nal_unit_type;
137
#define NAL_SLICE                1
138
#define NAL_DPA                        2
139
#define NAL_DPB                        3
140
#define NAL_DPC                        4
141
#define NAL_IDR_SLICE                5
142
#define NAL_SEI                        6
143
#define NAL_SPS                        7
144
#define NAL_PPS                        8
145
#define NAL_PICTURE_DELIMITER        9
146
#define NAL_FILTER_DATA                10
147
    uint8_t *rbsp_buffer;
148
    int rbsp_buffer_size;
149

    
150
    int chroma_qp; //QPc
151

    
152
    int prev_mb_skiped; //FIXME remove (IMHO not used)
153

    
154
    //prediction stuff
155
    int chroma_pred_mode;
156
    int intra16x16_pred_mode;
157
    
158
    int8_t intra4x4_pred_mode_cache[5*8];
159
    int8_t (*intra4x4_pred_mode)[8];
160
    void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
161
    void (*pred8x8  [4+3])(uint8_t *src, int stride);
162
    void (*pred16x16[4+3])(uint8_t *src, int stride);
163
    unsigned int topleft_samples_available;
164
    unsigned int top_samples_available;
165
    unsigned int topright_samples_available;
166
    unsigned int left_samples_available;
167
    uint8_t (*top_border)[16+2*8];
168
    uint8_t left_border[17+2*9];
169

    
170
    /**
171
     * non zero coeff count cache.
172
     * is 64 if not available.
173
     */
174
    uint8_t non_zero_count_cache[6*8];
175
    uint8_t (*non_zero_count)[16];
176

    
177
    /**
178
     * Motion vector cache.
179
     */
180
    int16_t mv_cache[2][5*8][2];
181
    int8_t ref_cache[2][5*8];
182
#define LIST_NOT_USED -1 //FIXME rename?
183
#define PART_NOT_AVAILABLE -2
184
    
185
    /**
186
     * is 1 if the specific list MV&references are set to 0,0,-2.
187
     */
188
    int mv_cache_clean[2];
189

    
190
    int block_offset[16+8];
191
    int chroma_subblock_offset[16]; //FIXME remove
192
    
193
    uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
194
    uint16_t *mb2b8_xy;
195
    int b_stride;
196
    int b8_stride;
197

    
198
    int halfpel_flag;
199
    int thirdpel_flag;
200

    
201
    int unknown_svq3_flag;
202
    int next_slice_index;
203

    
204
    SPS sps_buffer[MAX_SPS_COUNT];
205
    SPS sps; ///< current sps
206
    
207
    PPS pps_buffer[MAX_PPS_COUNT];
208
    /**
209
     * current pps
210
     */
211
    PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
212

    
213
    int slice_num;
214
    uint8_t *slice_table_base;
215
    uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
216
    int slice_type;
217
    int slice_type_fixed;
218
    
219
    //interlacing specific flags
220
    int mb_field_decoding_flag;
221
    
222
    int sub_mb_type[4];
223
    
224
    //POC stuff
225
    int poc_lsb;
226
    int poc_msb;
227
    int delta_poc_bottom;
228
    int delta_poc[2];
229
    int frame_num;
230
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
231
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
232
    int frame_num_offset;         ///< for POC type 2
233
    int prev_frame_num_offset;    ///< for POC type 2
234
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
235

    
236
    /**
237
     * frame_num for frames or 2*frame_num for field pics.
238
     */
239
    int curr_pic_num;
240
    
241
    /**
242
     * max_frame_num or 2*max_frame_num for field pics.
243
     */
244
    int max_pic_num;
245

    
246
    //Weighted pred stuff
247
    int luma_log2_weight_denom;
248
    int chroma_log2_weight_denom;
249
    int luma_weight[2][16];
250
    int luma_offset[2][16];
251
    int chroma_weight[2][16][2];
252
    int chroma_offset[2][16][2];
253
   
254
    //deblock
255
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0 
256
    int slice_alpha_c0_offset;
257
    int slice_beta_offset;
258
     
259
    int redundant_pic_count;
260
    
261
    int direct_spatial_mv_pred;
262

    
263
    /**
264
     * num_ref_idx_l0/1_active_minus1 + 1
265
     */
266
    int ref_count[2];// FIXME split for AFF
267
    Picture *short_ref[16];
268
    Picture *long_ref[16];
269
    Picture default_ref_list[2][32];
270
    Picture ref_list[2][32]; //FIXME size?
271
    Picture field_ref_list[2][32]; //FIXME size?
272
    
273
    /**
274
     * memory management control operations buffer.
275
     */
276
    MMCO mmco[MAX_MMCO_COUNT];
277
    int mmco_index;
278
    
279
    int long_ref_count;  ///< number of actual long term references
280
    int short_ref_count; ///< number of actual short term references
281
    
282
    //data partitioning
283
    GetBitContext intra_gb;
284
    GetBitContext inter_gb;
285
    GetBitContext *intra_gb_ptr;
286
    GetBitContext *inter_gb_ptr;
287
    
288
    DCTELEM mb[16*24] __align8;
289

    
290
    /**
291
     * Cabac
292
     */
293
    CABACContext cabac;
294
    uint8_t      cabac_state[399];
295
    int          cabac_init_idc;
296

    
297
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
298
    uint16_t     *cbp_table;
299
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
300
    uint8_t     *chroma_pred_mode_table;
301
    int         last_qscale_diff;
302
    int16_t     (*mvd_table[2])[2];
303
    int16_t     mvd_cache[2][5*8][2];
304

    
305
}H264Context;
306

    
307
static VLC coeff_token_vlc[4];
308
static VLC chroma_dc_coeff_token_vlc;
309

    
310
static VLC total_zeros_vlc[15];
311
static VLC chroma_dc_total_zeros_vlc[3];
312

    
313
static VLC run_vlc[6];
314
static VLC run7_vlc;
315

    
316
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
317
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
318
static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
319

    
320
static inline uint32_t pack16to32(int a, int b){
321
#ifdef WORDS_BIGENDIAN
322
   return (b&0xFFFF) + (a<<16);
323
#else
324
   return (a&0xFFFF) + (b<<16);
325
#endif
326
}
327

    
328
/**
329
 * fill a rectangle.
330
 * @param h height of the recatangle, should be a constant
331
 * @param w width of the recatangle, should be a constant
332
 * @param size the size of val (1 or 4), should be a constant
333
 */
334
static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
335
    uint8_t *p= (uint8_t*)vp;
336
    assert(size==1 || size==4);
337
    
338
    w      *= size;
339
    stride *= size;
340
    
341
//FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
342
    if(w==2 && h==2){
343
        *(uint16_t*)(p + 0)=
344
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
345
    }else if(w==2 && h==4){
346
        *(uint16_t*)(p + 0*stride)=
347
        *(uint16_t*)(p + 1*stride)=
348
        *(uint16_t*)(p + 2*stride)=
349
        *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
350
    }else if(w==4 && h==1){
351
        *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
352
    }else if(w==4 && h==2){
353
        *(uint32_t*)(p + 0*stride)=
354
        *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
355
    }else if(w==4 && h==4){
356
        *(uint32_t*)(p + 0*stride)=
357
        *(uint32_t*)(p + 1*stride)=
358
        *(uint32_t*)(p + 2*stride)=
359
        *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
360
    }else if(w==8 && h==1){
361
        *(uint32_t*)(p + 0)=
362
        *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
363
    }else if(w==8 && h==2){
364
        *(uint32_t*)(p + 0 + 0*stride)=
365
        *(uint32_t*)(p + 4 + 0*stride)=
366
        *(uint32_t*)(p + 0 + 1*stride)=
367
        *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
368
    }else if(w==8 && h==4){
369
        *(uint64_t*)(p + 0*stride)=
370
        *(uint64_t*)(p + 1*stride)=
371
        *(uint64_t*)(p + 2*stride)=
372
        *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
373
    }else if(w==16 && h==2){
374
        *(uint64_t*)(p + 0+0*stride)=
375
        *(uint64_t*)(p + 8+0*stride)=
376
        *(uint64_t*)(p + 0+1*stride)=
377
        *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
378
    }else if(w==16 && h==4){
379
        *(uint64_t*)(p + 0+0*stride)=
380
        *(uint64_t*)(p + 8+0*stride)=
381
        *(uint64_t*)(p + 0+1*stride)=
382
        *(uint64_t*)(p + 8+1*stride)=
383
        *(uint64_t*)(p + 0+2*stride)=
384
        *(uint64_t*)(p + 8+2*stride)=
385
        *(uint64_t*)(p + 0+3*stride)=
386
        *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
387
    }else
388
        assert(0);
389
}
390

    
391
static inline void fill_caches(H264Context *h, int mb_type){
392
    MpegEncContext * const s = &h->s;
393
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
394
    int topleft_xy, top_xy, topright_xy, left_xy[2];
395
    int topleft_type, top_type, topright_type, left_type[2];
396
    int left_block[4];
397
    int i;
398

    
399
    //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it 
400
    
401
    if(h->sps.mb_aff){
402
    //FIXME
403
        topleft_xy = 0; /* avoid warning */
404
        top_xy = 0; /* avoid warning */
405
        topright_xy = 0; /* avoid warning */
406
    }else{
407
        topleft_xy = mb_xy-1 - s->mb_stride;
408
        top_xy     = mb_xy   - s->mb_stride;
409
        topright_xy= mb_xy+1 - s->mb_stride;
410
        left_xy[0]   = mb_xy-1;
411
        left_xy[1]   = mb_xy-1;
412
        left_block[0]= 0;
413
        left_block[1]= 1;
414
        left_block[2]= 2;
415
        left_block[3]= 3;
416
    }
417

    
418
    topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
419
    top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
420
    topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
421
    left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
422
    left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
423

    
424
    if(IS_INTRA(mb_type)){
425
        h->topleft_samples_available= 
426
        h->top_samples_available= 
427
        h->left_samples_available= 0xFFFF;
428
        h->topright_samples_available= 0xEEEA;
429

    
430
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
431
            h->topleft_samples_available= 0xB3FF;
432
            h->top_samples_available= 0x33FF;
433
            h->topright_samples_available= 0x26EA;
434
        }
435
        for(i=0; i<2; i++){
436
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
437
                h->topleft_samples_available&= 0xDF5F;
438
                h->left_samples_available&= 0x5F5F;
439
            }
440
        }
441
        
442
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
443
            h->topleft_samples_available&= 0x7FFF;
444
        
445
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
446
            h->topright_samples_available&= 0xFBFF;
447
    
448
        if(IS_INTRA4x4(mb_type)){
449
            if(IS_INTRA4x4(top_type)){
450
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
451
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
452
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
453
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
454
            }else{
455
                int pred;
456
                if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
457
                    pred= 2;
458
                else{
459
                    pred= -1;
460
                }
461
                h->intra4x4_pred_mode_cache[4+8*0]=
462
                h->intra4x4_pred_mode_cache[5+8*0]=
463
                h->intra4x4_pred_mode_cache[6+8*0]=
464
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
465
            }
466
            for(i=0; i<2; i++){
467
                if(IS_INTRA4x4(left_type[i])){
468
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
469
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
470
                }else{
471
                    int pred;
472
                    if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
473
                        pred= 2;
474
                    else{
475
                        pred= -1;
476
                    }
477
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
478
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
479
                }
480
            }
481
        }
482
    }
483
    
484
    
485
/*
486
0 . T T. T T T T 
487
1 L . .L . . . . 
488
2 L . .L . . . . 
489
3 . T TL . . . . 
490
4 L . .L . . . . 
491
5 L . .. . . . . 
492
*/
493
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
494
    if(top_type){
495
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
496
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
497
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
498
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
499
    
500
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
501
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
502
    
503
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
504
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
505
    }else{
506
        h->non_zero_count_cache[4+8*0]=      
507
        h->non_zero_count_cache[5+8*0]=
508
        h->non_zero_count_cache[6+8*0]=
509
        h->non_zero_count_cache[7+8*0]=
510
    
511
        h->non_zero_count_cache[1+8*0]=
512
        h->non_zero_count_cache[2+8*0]=
513
    
514
        h->non_zero_count_cache[1+8*3]=
515
        h->non_zero_count_cache[2+8*3]= 64;
516
    }
517
    
518
    if(left_type[0]){
519
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
520
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
521
        h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
522
        h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
523
    }else{
524
        h->non_zero_count_cache[3+8*1]= 
525
        h->non_zero_count_cache[3+8*2]= 
526
        h->non_zero_count_cache[0+8*1]= 
527
        h->non_zero_count_cache[0+8*4]= 64;
528
    }
529
    
530
    if(left_type[1]){
531
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
532
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
533
        h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
534
        h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
535
    }else{
536
        h->non_zero_count_cache[3+8*3]= 
537
        h->non_zero_count_cache[3+8*4]= 
538
        h->non_zero_count_cache[0+8*2]= 
539
        h->non_zero_count_cache[0+8*5]= 64;
540
    }
541
    
542
#if 1
543
    if(IS_INTER(mb_type)){
544
        int list;
545
        for(list=0; list<2; list++){
546
            if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
547
                /*if(!h->mv_cache_clean[list]){
548
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
549
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
550
                    h->mv_cache_clean[list]= 1;
551
                }*/
552
                continue; //FIXME direct mode ...
553
            }
554
            h->mv_cache_clean[list]= 0;
555
            
556
            if(IS_INTER(topleft_type)){
557
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
558
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
559
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
560
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
561
            }else{
562
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
563
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
564
            }
565
            
566
            if(IS_INTER(top_type)){
567
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
568
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
569
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
570
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
571
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
572
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
573
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
574
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
575
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
576
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
577
            }else{
578
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
579
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
580
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
581
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
582
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
583
            }
584

    
585
            if(IS_INTER(topright_type)){
586
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
587
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
588
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
589
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
590
            }else{
591
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
592
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
593
            }
594
            
595
            //FIXME unify cleanup or sth
596
            if(IS_INTER(left_type[0])){
597
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
598
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
599
                *(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]];
600
                *(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]];
601
                h->ref_cache[list][scan8[0] - 1 + 0*8]= 
602
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
603
            }else{
604
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
605
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
606
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
607
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
608
            }
609
            
610
            if(IS_INTER(left_type[1])){
611
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
612
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
613
                *(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]];
614
                *(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]];
615
                h->ref_cache[list][scan8[0] - 1 + 2*8]= 
616
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
617
            }else{
618
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
619
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
620
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
621
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
622
            }
623

    
624
            h->ref_cache[list][scan8[5 ]+1] = 
625
            h->ref_cache[list][scan8[7 ]+1] = 
626
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewher else)
627
            h->ref_cache[list][scan8[4 ]] = 
628
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
629
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
630
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
631
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
632
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
633
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
634

    
635
            if( h->pps.cabac ) {
636
                /* XXX beurk, Load mvd */
637
                if(IS_INTER(topleft_type)){
638
                    const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
639
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
640
                }else{
641
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
642
                }
643

    
644
                if(IS_INTER(top_type)){
645
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
646
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
647
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
648
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
649
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
650
                }else{
651
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]= 
652
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]= 
653
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]= 
654
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
655
                }
656
                if(IS_INTER(left_type[0])){
657
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
658
                    *(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]];
659
                    *(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]];
660
                }else{
661
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
662
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
663
                }
664
                if(IS_INTER(left_type[1])){
665
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
666
                    *(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]];
667
                    *(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]];
668
                }else{
669
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
670
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
671
                }
672
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
673
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
674
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
675
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
676
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
677
            }
678
        }
679
//FIXME
680
    }
681
#endif
682
}
683

    
684
static inline void write_back_intra_pred_mode(H264Context *h){
685
    MpegEncContext * const s = &h->s;
686
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
687

    
688
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
689
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
690
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
691
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
692
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
693
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
694
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
695
}
696

    
697
/**
698
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
699
 */
700
static inline int check_intra4x4_pred_mode(H264Context *h){
701
    MpegEncContext * const s = &h->s;
702
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
703
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
704
    int i;
705
    
706
    if(!(h->top_samples_available&0x8000)){
707
        for(i=0; i<4; i++){
708
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
709
            if(status<0){
710
                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);
711
                return -1;
712
            } else if(status){
713
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
714
            }
715
        }
716
    }
717
    
718
    if(!(h->left_samples_available&0x8000)){
719
        for(i=0; i<4; i++){
720
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
721
            if(status<0){
722
                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);
723
                return -1;
724
            } else if(status){
725
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
726
            }
727
        }
728
    }
729

    
730
    return 0;
731
} //FIXME cleanup like next
732

    
733
/**
734
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
735
 */
736
static inline int check_intra_pred_mode(H264Context *h, int mode){
737
    MpegEncContext * const s = &h->s;
738
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
739
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
740
    
741
    if(!(h->top_samples_available&0x8000)){
742
        mode= top[ mode ];
743
        if(mode<0){
744
            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);
745
            return -1;
746
        }
747
    }
748
    
749
    if(!(h->left_samples_available&0x8000)){
750
        mode= left[ mode ];
751
        if(mode<0){
752
            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);
753
            return -1;
754
        } 
755
    }
756

    
757
    return mode;
758
}
759

    
760
/**
761
 * gets the predicted intra4x4 prediction mode.
762
 */
763
static inline int pred_intra_mode(H264Context *h, int n){
764
    const int index8= scan8[n];
765
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
766
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
767
    const int min= FFMIN(left, top);
768

    
769
    tprintf("mode:%d %d min:%d\n", left ,top, min);
770

    
771
    if(min<0) return DC_PRED;
772
    else      return min;
773
}
774

    
775
static inline void write_back_non_zero_count(H264Context *h){
776
    MpegEncContext * const s = &h->s;
777
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
778

    
779
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
780
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
781
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
782
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
783
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
784
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
785
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
786
    
787
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
788
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
789
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
790

    
791
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
792
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
793
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
794
}
795

    
796
/**
797
 * gets the predicted number of non zero coefficients.
798
 * @param n block index
799
 */
800
static inline int pred_non_zero_count(H264Context *h, int n){
801
    const int index8= scan8[n];
802
    const int left= h->non_zero_count_cache[index8 - 1];
803
    const int top = h->non_zero_count_cache[index8 - 8];
804
    int i= left + top;
805
    
806
    if(i<64) i= (i+1)>>1;
807

    
808
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
809

    
810
    return i&31;
811
}
812

    
813
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
814
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
815

    
816
    if(topright_ref != PART_NOT_AVAILABLE){
817
        *C= h->mv_cache[list][ i - 8 + part_width ];
818
        return topright_ref;
819
    }else{
820
        tprintf("topright MV not available\n");
821

    
822
        *C= h->mv_cache[list][ i - 8 - 1 ];
823
        return h->ref_cache[list][ i - 8 - 1 ];
824
    }
825
}
826

    
827
/**
828
 * gets the predicted MV.
829
 * @param n the block index
830
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
831
 * @param mx the x component of the predicted motion vector
832
 * @param my the y component of the predicted motion vector
833
 */
834
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
835
    const int index8= scan8[n];
836
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
837
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
838
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
839
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
840
    const int16_t * C;
841
    int diagonal_ref, match_count;
842

    
843
    assert(part_width==1 || part_width==2 || part_width==4);
844

    
845
/* mv_cache
846
  B . . A T T T T 
847
  U . . L . . , .
848
  U . . L . . . .
849
  U . . L . . , .
850
  . . . L . . . .
851
*/
852

    
853
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
854
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
855
    if(match_count > 1){ //most common
856
        *mx= mid_pred(A[0], B[0], C[0]);
857
        *my= mid_pred(A[1], B[1], C[1]);
858
    }else if(match_count==1){
859
        if(left_ref==ref){
860
            *mx= A[0];
861
            *my= A[1];        
862
        }else if(top_ref==ref){
863
            *mx= B[0];
864
            *my= B[1];        
865
        }else{
866
            *mx= C[0];
867
            *my= C[1];        
868
        }
869
    }else{
870
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
871
            *mx= A[0];
872
            *my= A[1];        
873
        }else{
874
            *mx= mid_pred(A[0], B[0], C[0]);
875
            *my= mid_pred(A[1], B[1], C[1]);
876
        }
877
    }
878
        
879
    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);
880
}
881

    
882
/**
883
 * gets the directionally predicted 16x8 MV.
884
 * @param n the block index
885
 * @param mx the x component of the predicted motion vector
886
 * @param my the y component of the predicted motion vector
887
 */
888
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
889
    if(n==0){
890
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
891
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
892

    
893
        tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
894
        
895
        if(top_ref == ref){
896
            *mx= B[0];
897
            *my= B[1];
898
            return;
899
        }
900
    }else{
901
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
902
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
903
        
904
        tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
905

    
906
        if(left_ref == ref){
907
            *mx= A[0];
908
            *my= A[1];
909
            return;
910
        }
911
    }
912

    
913
    //RARE
914
    pred_motion(h, n, 4, list, ref, mx, my);
915
}
916

    
917
/**
918
 * gets the directionally predicted 8x16 MV.
919
 * @param n the block index
920
 * @param mx the x component of the predicted motion vector
921
 * @param my the y component of the predicted motion vector
922
 */
923
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
924
    if(n==0){
925
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
926
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
927
        
928
        tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
929

    
930
        if(left_ref == ref){
931
            *mx= A[0];
932
            *my= A[1];
933
            return;
934
        }
935
    }else{
936
        const int16_t * C;
937
        int diagonal_ref;
938

    
939
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
940
        
941
        tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
942

    
943
        if(diagonal_ref == ref){ 
944
            *mx= C[0];
945
            *my= C[1];
946
            return;
947
        }
948
    }
949

    
950
    //RARE
951
    pred_motion(h, n, 2, list, ref, mx, my);
952
}
953

    
954
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
955
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
956
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
957

    
958
    tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
959

    
960
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
961
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
962
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
963
       
964
        *mx = *my = 0;
965
        return;
966
    }
967
        
968
    pred_motion(h, 0, 4, 0, 0, mx, my);
969

    
970
    return;
971
}
972

    
973
static inline void write_back_motion(H264Context *h, int mb_type){
974
    MpegEncContext * const s = &h->s;
975
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
976
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
977
    int list;
978

    
979
    for(list=0; list<2; list++){
980
        int y;
981
        if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
982
            if(1){ //FIXME skip or never read if mb_type doesnt use it
983
                for(y=0; y<4; y++){
984
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
985
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
986
                }
987
                if( h->pps.cabac ) {
988
                    /* FIXME needed ? */
989
                    for(y=0; y<4; y++){
990
                        *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
991
                        *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
992
                    }
993
                }
994
                for(y=0; y<2; y++){
995
                    *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
996
                }
997
            }
998
            continue; //FIXME direct mode ...
999
        }
1000
        
1001
        for(y=0; y<4; y++){
1002
            *(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];
1003
            *(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];
1004
        }
1005
        if( h->pps.cabac ) {
1006
            for(y=0; y<4; y++){
1007
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1008
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1009
            }
1010
        }
1011
        for(y=0; y<2; y++){
1012
            s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1013
            s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1014
        }
1015
    }
1016
}
1017

    
1018
/**
1019
 * Decodes a network abstraction layer unit.
1020
 * @param consumed is the number of bytes used as input
1021
 * @param length is the length of the array
1022
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1023
 * @returns decoded bytes, might be src+1 if no escapes 
1024
 */
1025
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1026
    int i, si, di;
1027
    uint8_t *dst;
1028

    
1029
//    src[0]&0x80;                //forbidden bit
1030
    h->nal_ref_idc= src[0]>>5;
1031
    h->nal_unit_type= src[0]&0x1F;
1032

    
1033
    src++; length--;
1034
#if 0    
1035
    for(i=0; i<length; i++)
1036
        printf("%2X ", src[i]);
1037
#endif
1038
    for(i=0; i+1<length; i+=2){
1039
        if(src[i]) continue;
1040
        if(i>0 && src[i-1]==0) i--;
1041
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1042
            if(src[i+2]!=3){
1043
                /* startcode, so we must be past the end */
1044
                length=i;
1045
            }
1046
            break;
1047
        }
1048
    }
1049

    
1050
    if(i>=length-1){ //no escaped 0
1051
        *dst_length= length;
1052
        *consumed= length+1; //+1 for the header
1053
        return src; 
1054
    }
1055

    
1056
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1057
    dst= h->rbsp_buffer;
1058

    
1059
//printf("deoding esc\n");
1060
    si=di=0;
1061
    while(si<length){ 
1062
        //remove escapes (very rare 1:2^22)
1063
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1064
            if(src[si+2]==3){ //escape
1065
                dst[di++]= 0;
1066
                dst[di++]= 0;
1067
                si+=3;
1068
            }else //next start code
1069
                break;
1070
        }
1071

    
1072
        dst[di++]= src[si++];
1073
    }
1074

    
1075
    *dst_length= di;
1076
    *consumed= si + 1;//+1 for the header
1077
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1078
    return dst;
1079
}
1080

    
1081
#if 0
1082
/**
1083
 * @param src the data which should be escaped
1084
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1085
 * @param length the length of the src data
1086
 * @param dst_length the length of the dst array
1087
 * @returns length of escaped data in bytes or -1 if an error occured
1088
 */
1089
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1090
    int i, escape_count, si, di;
1091
    uint8_t *temp;
1092
    
1093
    assert(length>=0);
1094
    assert(dst_length>0);
1095
    
1096
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1097

1098
    if(length==0) return 1;
1099

1100
    escape_count= 0;
1101
    for(i=0; i<length; i+=2){
1102
        if(src[i]) continue;
1103
        if(i>0 && src[i-1]==0) 
1104
            i--;
1105
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1106
            escape_count++;
1107
            i+=2;
1108
        }
1109
    }
1110
    
1111
    if(escape_count==0){ 
1112
        if(dst+1 != src)
1113
            memcpy(dst+1, src, length);
1114
        return length + 1;
1115
    }
1116
    
1117
    if(length + escape_count + 1> dst_length)
1118
        return -1;
1119

1120
    //this should be damn rare (hopefully)
1121

1122
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1123
    temp= h->rbsp_buffer;
1124
//printf("encoding esc\n");
1125
    
1126
    si= 0;
1127
    di= 0;
1128
    while(si < length){
1129
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1130
            temp[di++]= 0; si++;
1131
            temp[di++]= 0; si++;
1132
            temp[di++]= 3; 
1133
            temp[di++]= src[si++];
1134
        }
1135
        else
1136
            temp[di++]= src[si++];
1137
    }
1138
    memcpy(dst+1, temp, length+escape_count);
1139
    
1140
    assert(di == length+escape_count);
1141
    
1142
    return di + 1;
1143
}
1144

1145
/**
1146
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1147
 */
1148
static void encode_rbsp_trailing(PutBitContext *pb){
1149
    int length;
1150
    put_bits(pb, 1, 1);
1151
    length= (-put_bits_count(pb))&7;
1152
    if(length) put_bits(pb, length, 0);
1153
}
1154
#endif
1155

    
1156
/**
1157
 * identifies the exact end of the bitstream
1158
 * @return the length of the trailing, or 0 if damaged
1159
 */
1160
static int decode_rbsp_trailing(uint8_t *src){
1161
    int v= *src;
1162
    int r;
1163

    
1164
    tprintf("rbsp trailing %X\n", v);
1165

    
1166
    for(r=1; r<9; r++){
1167
        if(v&1) return r;
1168
        v>>=1;
1169
    }
1170
    return 0;
1171
}
1172

    
1173
/**
1174
 * idct tranforms the 16 dc values and dequantize them.
1175
 * @param qp quantization parameter
1176
 */
1177
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1178
    const int qmul= dequant_coeff[qp][0];
1179
#define stride 16
1180
    int i;
1181
    int temp[16]; //FIXME check if this is a good idea
1182
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1183
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1184

    
1185
//memset(block, 64, 2*256);
1186
//return;
1187
    for(i=0; i<4; i++){
1188
        const int offset= y_offset[i];
1189
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1190
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1191
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1192
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1193

    
1194
        temp[4*i+0]= z0+z3;
1195
        temp[4*i+1]= z1+z2;
1196
        temp[4*i+2]= z1-z2;
1197
        temp[4*i+3]= z0-z3;
1198
    }
1199

    
1200
    for(i=0; i<4; i++){
1201
        const int offset= x_offset[i];
1202
        const int z0= temp[4*0+i] + temp[4*2+i];
1203
        const int z1= temp[4*0+i] - temp[4*2+i];
1204
        const int z2= temp[4*1+i] - temp[4*3+i];
1205
        const int z3= temp[4*1+i] + temp[4*3+i];
1206

    
1207
        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1208
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1209
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1210
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1211
    }
1212
}
1213

    
1214
#if 0
1215
/**
1216
 * dct tranforms the 16 dc values.
1217
 * @param qp quantization parameter ??? FIXME
1218
 */
1219
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1220
//    const int qmul= dequant_coeff[qp][0];
1221
    int i;
1222
    int temp[16]; //FIXME check if this is a good idea
1223
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1224
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1225

1226
    for(i=0; i<4; i++){
1227
        const int offset= y_offset[i];
1228
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1229
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1230
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1231
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1232

1233
        temp[4*i+0]= z0+z3;
1234
        temp[4*i+1]= z1+z2;
1235
        temp[4*i+2]= z1-z2;
1236
        temp[4*i+3]= z0-z3;
1237
    }
1238

1239
    for(i=0; i<4; i++){
1240
        const int offset= x_offset[i];
1241
        const int z0= temp[4*0+i] + temp[4*2+i];
1242
        const int z1= temp[4*0+i] - temp[4*2+i];
1243
        const int z2= temp[4*1+i] - temp[4*3+i];
1244
        const int z3= temp[4*1+i] + temp[4*3+i];
1245

1246
        block[stride*0 +offset]= (z0 + z3)>>1;
1247
        block[stride*2 +offset]= (z1 + z2)>>1;
1248
        block[stride*8 +offset]= (z1 - z2)>>1;
1249
        block[stride*10+offset]= (z0 - z3)>>1;
1250
    }
1251
}
1252
#endif
1253

    
1254
#undef xStride
1255
#undef stride
1256

    
1257
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1258
    const int qmul= dequant_coeff[qp][0];
1259
    const int stride= 16*2;
1260
    const int xStride= 16;
1261
    int a,b,c,d,e;
1262

    
1263
    a= block[stride*0 + xStride*0];
1264
    b= block[stride*0 + xStride*1];
1265
    c= block[stride*1 + xStride*0];
1266
    d= block[stride*1 + xStride*1];
1267

    
1268
    e= a-b;
1269
    a= a+b;
1270
    b= c-d;
1271
    c= c+d;
1272

    
1273
    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1274
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1275
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1276
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1277
}
1278

    
1279
#if 0
1280
static void chroma_dc_dct_c(DCTELEM *block){
1281
    const int stride= 16*2;
1282
    const int xStride= 16;
1283
    int a,b,c,d,e;
1284

1285
    a= block[stride*0 + xStride*0];
1286
    b= block[stride*0 + xStride*1];
1287
    c= block[stride*1 + xStride*0];
1288
    d= block[stride*1 + xStride*1];
1289

1290
    e= a-b;
1291
    a= a+b;
1292
    b= c-d;
1293
    c= c+d;
1294

1295
    block[stride*0 + xStride*0]= (a+c);
1296
    block[stride*0 + xStride*1]= (e+b);
1297
    block[stride*1 + xStride*0]= (a-c);
1298
    block[stride*1 + xStride*1]= (e-b);
1299
}
1300
#endif
1301

    
1302
/**
1303
 * gets the chroma qp.
1304
 */
1305
static inline int get_chroma_qp(H264Context *h, int qscale){
1306
    
1307
    return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1308
}
1309

    
1310

    
1311
/**
1312
 *
1313
 */
1314
static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1315
    int i;
1316
    uint8_t *cm = cropTbl + MAX_NEG_CROP;
1317

    
1318
    block[0] += 32;
1319

    
1320
    for(i=0; i<4; i++){
1321
        const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
1322
        const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
1323
        const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
1324
        const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
1325

    
1326
        block[0 + 4*i]= z0 + z3;
1327
        block[1 + 4*i]= z1 + z2;
1328
        block[2 + 4*i]= z1 - z2;
1329
        block[3 + 4*i]= z0 - z3;
1330
    }
1331

    
1332
    for(i=0; i<4; i++){
1333
        const int z0=  block[i + 4*0]     +  block[i + 4*2];
1334
        const int z1=  block[i + 4*0]     -  block[i + 4*2];
1335
        const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
1336
        const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
1337

    
1338
        dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1339
        dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1340
        dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1341
        dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1342
    }
1343
}
1344

    
1345
#if 0
1346
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1347
    int i;
1348
    //FIXME try int temp instead of block
1349
    
1350
    for(i=0; i<4; i++){
1351
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1352
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1353
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1354
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1355
        const int z0= d0 + d3;
1356
        const int z3= d0 - d3;
1357
        const int z1= d1 + d2;
1358
        const int z2= d1 - d2;
1359
        
1360
        block[0 + 4*i]=   z0 +   z1;
1361
        block[1 + 4*i]= 2*z3 +   z2;
1362
        block[2 + 4*i]=   z0 -   z1;
1363
        block[3 + 4*i]=   z3 - 2*z2;
1364
    }    
1365

1366
    for(i=0; i<4; i++){
1367
        const int z0= block[0*4 + i] + block[3*4 + i];
1368
        const int z3= block[0*4 + i] - block[3*4 + i];
1369
        const int z1= block[1*4 + i] + block[2*4 + i];
1370
        const int z2= block[1*4 + i] - block[2*4 + i];
1371
        
1372
        block[0*4 + i]=   z0 +   z1;
1373
        block[1*4 + i]= 2*z3 +   z2;
1374
        block[2*4 + i]=   z0 -   z1;
1375
        block[3*4 + i]=   z3 - 2*z2;
1376
    }
1377
}
1378
#endif
1379

    
1380
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1381
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1382
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1383
    int i;
1384
    const int * const quant_table= quant_coeff[qscale];
1385
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1386
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1387
    const unsigned int threshold2= (threshold1<<1);
1388
    int last_non_zero;
1389

    
1390
    if(seperate_dc){
1391
        if(qscale<=18){
1392
            //avoid overflows
1393
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1394
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1395
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1396

    
1397
            int level= block[0]*quant_coeff[qscale+18][0];
1398
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1399
                if(level>0){
1400
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1401
                    block[0]= level;
1402
                }else{
1403
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1404
                    block[0]= -level;
1405
                }
1406
//                last_non_zero = i;
1407
            }else{
1408
                block[0]=0;
1409
            }
1410
        }else{
1411
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1412
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1413
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1414

    
1415
            int level= block[0]*quant_table[0];
1416
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1417
                if(level>0){
1418
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1419
                    block[0]= level;
1420
                }else{
1421
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1422
                    block[0]= -level;
1423
                }
1424
//                last_non_zero = i;
1425
            }else{
1426
                block[0]=0;
1427
            }
1428
        }
1429
        last_non_zero= 0;
1430
        i=1;
1431
    }else{
1432
        last_non_zero= -1;
1433
        i=0;
1434
    }
1435

    
1436
    for(; i<16; i++){
1437
        const int j= scantable[i];
1438
        int level= block[j]*quant_table[j];
1439

    
1440
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1441
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1442
        if(((unsigned)(level+threshold1))>threshold2){
1443
            if(level>0){
1444
                level= (bias + level)>>QUANT_SHIFT;
1445
                block[j]= level;
1446
            }else{
1447
                level= (bias - level)>>QUANT_SHIFT;
1448
                block[j]= -level;
1449
            }
1450
            last_non_zero = i;
1451
        }else{
1452
            block[j]=0;
1453
        }
1454
    }
1455

    
1456
    return last_non_zero;
1457
}
1458

    
1459
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1460
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1461
    ((uint32_t*)(src+0*stride))[0]= a;
1462
    ((uint32_t*)(src+1*stride))[0]= a;
1463
    ((uint32_t*)(src+2*stride))[0]= a;
1464
    ((uint32_t*)(src+3*stride))[0]= a;
1465
}
1466

    
1467
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1468
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1469
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1470
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1471
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1472
}
1473

    
1474
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1475
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1476
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1477
    
1478
    ((uint32_t*)(src+0*stride))[0]= 
1479
    ((uint32_t*)(src+1*stride))[0]= 
1480
    ((uint32_t*)(src+2*stride))[0]= 
1481
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1482
}
1483

    
1484
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1485
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1486
    
1487
    ((uint32_t*)(src+0*stride))[0]= 
1488
    ((uint32_t*)(src+1*stride))[0]= 
1489
    ((uint32_t*)(src+2*stride))[0]= 
1490
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1491
}
1492

    
1493
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1494
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1495
    
1496
    ((uint32_t*)(src+0*stride))[0]= 
1497
    ((uint32_t*)(src+1*stride))[0]= 
1498
    ((uint32_t*)(src+2*stride))[0]= 
1499
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1500
}
1501

    
1502
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1503
    ((uint32_t*)(src+0*stride))[0]= 
1504
    ((uint32_t*)(src+1*stride))[0]= 
1505
    ((uint32_t*)(src+2*stride))[0]= 
1506
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1507
}
1508

    
1509

    
1510
#define LOAD_TOP_RIGHT_EDGE\
1511
    const int t4= topright[0];\
1512
    const int t5= topright[1];\
1513
    const int t6= topright[2];\
1514
    const int t7= topright[3];\
1515

    
1516
#define LOAD_LEFT_EDGE\
1517
    const int l0= src[-1+0*stride];\
1518
    const int l1= src[-1+1*stride];\
1519
    const int l2= src[-1+2*stride];\
1520
    const int l3= src[-1+3*stride];\
1521

    
1522
#define LOAD_TOP_EDGE\
1523
    const int t0= src[ 0-1*stride];\
1524
    const int t1= src[ 1-1*stride];\
1525
    const int t2= src[ 2-1*stride];\
1526
    const int t3= src[ 3-1*stride];\
1527

    
1528
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1529
    const int lt= src[-1-1*stride];
1530
    LOAD_TOP_EDGE
1531
    LOAD_LEFT_EDGE
1532

    
1533
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1534
    src[0+2*stride]=
1535
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1536
    src[0+1*stride]=
1537
    src[1+2*stride]=
1538
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1539
    src[0+0*stride]=
1540
    src[1+1*stride]=
1541
    src[2+2*stride]=
1542
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1543
    src[1+0*stride]=
1544
    src[2+1*stride]=
1545
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1546
    src[2+0*stride]=
1547
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1548
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1549
}
1550

    
1551
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1552
    LOAD_TOP_EDGE    
1553
    LOAD_TOP_RIGHT_EDGE    
1554
//    LOAD_LEFT_EDGE    
1555

    
1556
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1557
    src[1+0*stride]=
1558
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1559
    src[2+0*stride]=
1560
    src[1+1*stride]=
1561
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1562
    src[3+0*stride]=
1563
    src[2+1*stride]=
1564
    src[1+2*stride]=
1565
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1566
    src[3+1*stride]=
1567
    src[2+2*stride]=
1568
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1569
    src[3+2*stride]=
1570
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1571
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1572
}
1573

    
1574
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1575
    const int lt= src[-1-1*stride];
1576
    LOAD_TOP_EDGE    
1577
    LOAD_LEFT_EDGE    
1578
    const __attribute__((unused)) int unu= l3;
1579

    
1580
    src[0+0*stride]=
1581
    src[1+2*stride]=(lt + t0 + 1)>>1;
1582
    src[1+0*stride]=
1583
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1584
    src[2+0*stride]=
1585
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1586
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1587
    src[0+1*stride]=
1588
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1589
    src[1+1*stride]=
1590
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1591
    src[2+1*stride]=
1592
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1593
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1594
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1595
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1596
}
1597

    
1598
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1599
    LOAD_TOP_EDGE    
1600
    LOAD_TOP_RIGHT_EDGE    
1601
    const __attribute__((unused)) int unu= t7;
1602

    
1603
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1604
    src[1+0*stride]=
1605
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1606
    src[2+0*stride]=
1607
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1608
    src[3+0*stride]=
1609
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1610
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1611
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1612
    src[1+1*stride]=
1613
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1614
    src[2+1*stride]=
1615
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1616
    src[3+1*stride]=
1617
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1618
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1619
}
1620

    
1621
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1622
    LOAD_LEFT_EDGE    
1623

    
1624
    src[0+0*stride]=(l0 + l1 + 1)>>1;
1625
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1626
    src[2+0*stride]=
1627
    src[0+1*stride]=(l1 + l2 + 1)>>1;
1628
    src[3+0*stride]=
1629
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1630
    src[2+1*stride]=
1631
    src[0+2*stride]=(l2 + l3 + 1)>>1;
1632
    src[3+1*stride]=
1633
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1634
    src[3+2*stride]=
1635
    src[1+3*stride]=
1636
    src[0+3*stride]=
1637
    src[2+2*stride]=
1638
    src[2+3*stride]=
1639
    src[3+3*stride]=l3;
1640
}
1641
    
1642
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1643
    const int lt= src[-1-1*stride];
1644
    LOAD_TOP_EDGE    
1645
    LOAD_LEFT_EDGE    
1646
    const __attribute__((unused)) int unu= t3;
1647

    
1648
    src[0+0*stride]=
1649
    src[2+1*stride]=(lt + l0 + 1)>>1;
1650
    src[1+0*stride]=
1651
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1652
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1653
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1654
    src[0+1*stride]=
1655
    src[2+2*stride]=(l0 + l1 + 1)>>1;
1656
    src[1+1*stride]=
1657
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1658
    src[0+2*stride]=
1659
    src[2+3*stride]=(l1 + l2+ 1)>>1;
1660
    src[1+2*stride]=
1661
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1662
    src[0+3*stride]=(l2 + l3 + 1)>>1;
1663
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1664
}
1665

    
1666
static void pred16x16_vertical_c(uint8_t *src, int stride){
1667
    int i;
1668
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1669
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1670
    const uint32_t c= ((uint32_t*)(src-stride))[2];
1671
    const uint32_t d= ((uint32_t*)(src-stride))[3];
1672
    
1673
    for(i=0; i<16; i++){
1674
        ((uint32_t*)(src+i*stride))[0]= a;
1675
        ((uint32_t*)(src+i*stride))[1]= b;
1676
        ((uint32_t*)(src+i*stride))[2]= c;
1677
        ((uint32_t*)(src+i*stride))[3]= d;
1678
    }
1679
}
1680

    
1681
static void pred16x16_horizontal_c(uint8_t *src, int stride){
1682
    int i;
1683

    
1684
    for(i=0; i<16; i++){
1685
        ((uint32_t*)(src+i*stride))[0]=
1686
        ((uint32_t*)(src+i*stride))[1]=
1687
        ((uint32_t*)(src+i*stride))[2]=
1688
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1689
    }
1690
}
1691

    
1692
static void pred16x16_dc_c(uint8_t *src, int stride){
1693
    int i, dc=0;
1694

    
1695
    for(i=0;i<16; i++){
1696
        dc+= src[-1+i*stride];
1697
    }
1698
    
1699
    for(i=0;i<16; i++){
1700
        dc+= src[i-stride];
1701
    }
1702

    
1703
    dc= 0x01010101*((dc + 16)>>5);
1704

    
1705
    for(i=0; i<16; i++){
1706
        ((uint32_t*)(src+i*stride))[0]=
1707
        ((uint32_t*)(src+i*stride))[1]=
1708
        ((uint32_t*)(src+i*stride))[2]=
1709
        ((uint32_t*)(src+i*stride))[3]= dc;
1710
    }
1711
}
1712

    
1713
static void pred16x16_left_dc_c(uint8_t *src, int stride){
1714
    int i, dc=0;
1715

    
1716
    for(i=0;i<16; i++){
1717
        dc+= src[-1+i*stride];
1718
    }
1719
    
1720
    dc= 0x01010101*((dc + 8)>>4);
1721

    
1722
    for(i=0; i<16; i++){
1723
        ((uint32_t*)(src+i*stride))[0]=
1724
        ((uint32_t*)(src+i*stride))[1]=
1725
        ((uint32_t*)(src+i*stride))[2]=
1726
        ((uint32_t*)(src+i*stride))[3]= dc;
1727
    }
1728
}
1729

    
1730
static void pred16x16_top_dc_c(uint8_t *src, int stride){
1731
    int i, dc=0;
1732

    
1733
    for(i=0;i<16; i++){
1734
        dc+= src[i-stride];
1735
    }
1736
    dc= 0x01010101*((dc + 8)>>4);
1737

    
1738
    for(i=0; i<16; i++){
1739
        ((uint32_t*)(src+i*stride))[0]=
1740
        ((uint32_t*)(src+i*stride))[1]=
1741
        ((uint32_t*)(src+i*stride))[2]=
1742
        ((uint32_t*)(src+i*stride))[3]= dc;
1743
    }
1744
}
1745

    
1746
static void pred16x16_128_dc_c(uint8_t *src, int stride){
1747
    int i;
1748

    
1749
    for(i=0; i<16; i++){
1750
        ((uint32_t*)(src+i*stride))[0]=
1751
        ((uint32_t*)(src+i*stride))[1]=
1752
        ((uint32_t*)(src+i*stride))[2]=
1753
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1754
    }
1755
}
1756

    
1757
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1758
  int i, j, k;
1759
  int a;
1760
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
1761
  const uint8_t * const src0 = src+7-stride;
1762
  const uint8_t *src1 = src+8*stride-1;
1763
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
1764
  int H = src0[1] - src0[-1];
1765
  int V = src1[0] - src2[ 0];
1766
  for(k=2; k<=8; ++k) {
1767
    src1 += stride; src2 -= stride;
1768
    H += k*(src0[k] - src0[-k]);
1769
    V += k*(src1[0] - src2[ 0]);
1770
  }
1771
  if(svq3){
1772
    H = ( 5*(H/4) ) / 16;
1773
    V = ( 5*(V/4) ) / 16;
1774

    
1775
    /* required for 100% accuracy */
1776
    i = H; H = V; V = i;
1777
  }else{
1778
    H = ( 5*H+32 ) >> 6;
1779
    V = ( 5*V+32 ) >> 6;
1780
  }
1781

    
1782
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1783
  for(j=16; j>0; --j) {
1784
    int b = a;
1785
    a += V;
1786
    for(i=-16; i<0; i+=4) {
1787
      src[16+i] = cm[ (b    ) >> 5 ];
1788
      src[17+i] = cm[ (b+  H) >> 5 ];
1789
      src[18+i] = cm[ (b+2*H) >> 5 ];
1790
      src[19+i] = cm[ (b+3*H) >> 5 ];
1791
      b += 4*H;
1792
    }
1793
    src += stride;
1794
  }
1795
}
1796

    
1797
static void pred16x16_plane_c(uint8_t *src, int stride){
1798
    pred16x16_plane_compat_c(src, stride, 0);
1799
}
1800

    
1801
static void pred8x8_vertical_c(uint8_t *src, int stride){
1802
    int i;
1803
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1804
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1805
    
1806
    for(i=0; i<8; i++){
1807
        ((uint32_t*)(src+i*stride))[0]= a;
1808
        ((uint32_t*)(src+i*stride))[1]= b;
1809
    }
1810
}
1811

    
1812
static void pred8x8_horizontal_c(uint8_t *src, int stride){
1813
    int i;
1814

    
1815
    for(i=0; i<8; i++){
1816
        ((uint32_t*)(src+i*stride))[0]=
1817
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1818
    }
1819
}
1820

    
1821
static void pred8x8_128_dc_c(uint8_t *src, int stride){
1822
    int i;
1823

    
1824
    for(i=0; i<4; i++){
1825
        ((uint32_t*)(src+i*stride))[0]= 
1826
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1827
    }
1828
    for(i=4; i<8; i++){
1829
        ((uint32_t*)(src+i*stride))[0]= 
1830
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1831
    }
1832
}
1833

    
1834
static void pred8x8_left_dc_c(uint8_t *src, int stride){
1835
    int i;
1836
    int dc0, dc2;
1837

    
1838
    dc0=dc2=0;
1839
    for(i=0;i<4; i++){
1840
        dc0+= src[-1+i*stride];
1841
        dc2+= src[-1+(i+4)*stride];
1842
    }
1843
    dc0= 0x01010101*((dc0 + 2)>>2);
1844
    dc2= 0x01010101*((dc2 + 2)>>2);
1845

    
1846
    for(i=0; i<4; i++){
1847
        ((uint32_t*)(src+i*stride))[0]=
1848
        ((uint32_t*)(src+i*stride))[1]= dc0;
1849
    }
1850
    for(i=4; i<8; i++){
1851
        ((uint32_t*)(src+i*stride))[0]=
1852
        ((uint32_t*)(src+i*stride))[1]= dc2;
1853
    }
1854
}
1855

    
1856
static void pred8x8_top_dc_c(uint8_t *src, int stride){
1857
    int i;
1858
    int dc0, dc1;
1859

    
1860
    dc0=dc1=0;
1861
    for(i=0;i<4; i++){
1862
        dc0+= src[i-stride];
1863
        dc1+= src[4+i-stride];
1864
    }
1865
    dc0= 0x01010101*((dc0 + 2)>>2);
1866
    dc1= 0x01010101*((dc1 + 2)>>2);
1867

    
1868
    for(i=0; i<4; i++){
1869
        ((uint32_t*)(src+i*stride))[0]= dc0;
1870
        ((uint32_t*)(src+i*stride))[1]= dc1;
1871
    }
1872
    for(i=4; i<8; i++){
1873
        ((uint32_t*)(src+i*stride))[0]= dc0;
1874
        ((uint32_t*)(src+i*stride))[1]= dc1;
1875
    }
1876
}
1877

    
1878

    
1879
static void pred8x8_dc_c(uint8_t *src, int stride){
1880
    int i;
1881
    int dc0, dc1, dc2, dc3;
1882

    
1883
    dc0=dc1=dc2=0;
1884
    for(i=0;i<4; i++){
1885
        dc0+= src[-1+i*stride] + src[i-stride];
1886
        dc1+= src[4+i-stride];
1887
        dc2+= src[-1+(i+4)*stride];
1888
    }
1889
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1890
    dc0= 0x01010101*((dc0 + 4)>>3);
1891
    dc1= 0x01010101*((dc1 + 2)>>2);
1892
    dc2= 0x01010101*((dc2 + 2)>>2);
1893

    
1894
    for(i=0; i<4; i++){
1895
        ((uint32_t*)(src+i*stride))[0]= dc0;
1896
        ((uint32_t*)(src+i*stride))[1]= dc1;
1897
    }
1898
    for(i=4; i<8; i++){
1899
        ((uint32_t*)(src+i*stride))[0]= dc2;
1900
        ((uint32_t*)(src+i*stride))[1]= dc3;
1901
    }
1902
}
1903

    
1904
static void pred8x8_plane_c(uint8_t *src, int stride){
1905
  int j, k;
1906
  int a;
1907
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
1908
  const uint8_t * const src0 = src+3-stride;
1909
  const uint8_t *src1 = src+4*stride-1;
1910
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
1911
  int H = src0[1] - src0[-1];
1912
  int V = src1[0] - src2[ 0];
1913
  for(k=2; k<=4; ++k) {
1914
    src1 += stride; src2 -= stride;
1915
    H += k*(src0[k] - src0[-k]);
1916
    V += k*(src1[0] - src2[ 0]);
1917
  }
1918
  H = ( 17*H+16 ) >> 5;
1919
  V = ( 17*V+16 ) >> 5;
1920

    
1921
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1922
  for(j=8; j>0; --j) {
1923
    int b = a;
1924
    a += V;
1925
    src[0] = cm[ (b    ) >> 5 ];
1926
    src[1] = cm[ (b+  H) >> 5 ];
1927
    src[2] = cm[ (b+2*H) >> 5 ];
1928
    src[3] = cm[ (b+3*H) >> 5 ];
1929
    src[4] = cm[ (b+4*H) >> 5 ];
1930
    src[5] = cm[ (b+5*H) >> 5 ];
1931
    src[6] = cm[ (b+6*H) >> 5 ];
1932
    src[7] = cm[ (b+7*H) >> 5 ];
1933
    src += stride;
1934
  }
1935
}
1936

    
1937
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1938
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1939
                           int src_x_offset, int src_y_offset,
1940
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1941
    MpegEncContext * const s = &h->s;
1942
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1943
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1944
    const int luma_xy= (mx&3) + ((my&3)<<2);
1945
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1946
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1947
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1948
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1949
    int extra_height= extra_width;
1950
    int emu=0;
1951
    const int full_mx= mx>>2;
1952
    const int full_my= my>>2;
1953
    
1954
    assert(pic->data[0]);
1955
    
1956
    if(mx&7) extra_width -= 3;
1957
    if(my&7) extra_height -= 3;
1958
    
1959
    if(   full_mx < 0-extra_width 
1960
       || full_my < 0-extra_height 
1961
       || full_mx + 16/*FIXME*/ > s->width + extra_width 
1962
       || full_my + 16/*FIXME*/ > s->height + extra_height){
1963
        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);
1964
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1965
        emu=1;
1966
    }
1967
    
1968
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1969
    if(!square){
1970
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1971
    }
1972
    
1973
    if(s->flags&CODEC_FLAG_GRAY) return;
1974
    
1975
    if(emu){
1976
        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);
1977
            src_cb= s->edge_emu_buffer;
1978
    }
1979
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1980

    
1981
    if(emu){
1982
        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);
1983
            src_cr= s->edge_emu_buffer;
1984
    }
1985
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1986
}
1987

    
1988
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1989
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1990
                           int x_offset, int y_offset,
1991
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1992
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1993
                           int list0, int list1){
1994
    MpegEncContext * const s = &h->s;
1995
    qpel_mc_func *qpix_op=  qpix_put;
1996
    h264_chroma_mc_func chroma_op= chroma_put;
1997
    
1998
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
1999
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2000
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2001
    x_offset += 8*s->mb_x;
2002
    y_offset += 8*s->mb_y;
2003
    
2004
    if(list0){
2005
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2006
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2007
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2008
                           qpix_op, chroma_op);
2009

    
2010
        qpix_op=  qpix_avg;
2011
        chroma_op= chroma_avg;
2012
    }
2013

    
2014
    if(list1){
2015
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2016
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2017
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2018
                           qpix_op, chroma_op);
2019
    }
2020
}
2021

    
2022
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2023
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2024
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
2025
    MpegEncContext * const s = &h->s;
2026
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2027
    const int mb_type= s->current_picture.mb_type[mb_xy];
2028
    
2029
    assert(IS_INTER(mb_type));
2030
    
2031
    if(IS_16X16(mb_type)){
2032
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2033
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2034
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2035
    }else if(IS_16X8(mb_type)){
2036
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2037
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2038
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2039
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2040
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2041
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2042
    }else if(IS_8X16(mb_type)){
2043
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2044
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2045
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2046
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2047
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2048
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2049
    }else{
2050
        int i;
2051
        
2052
        assert(IS_8X8(mb_type));
2053

    
2054
        for(i=0; i<4; i++){
2055
            const int sub_mb_type= h->sub_mb_type[i];
2056
            const int n= 4*i;
2057
            int x_offset= (i&1)<<2;
2058
            int y_offset= (i&2)<<1;
2059

    
2060
            if(IS_SUB_8X8(sub_mb_type)){
2061
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2062
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2063
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2064
            }else if(IS_SUB_8X4(sub_mb_type)){
2065
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2066
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2067
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2068
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2069
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2070
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2071
            }else if(IS_SUB_4X8(sub_mb_type)){
2072
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2073
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2074
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2075
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2076
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2077
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2078
            }else{
2079
                int j;
2080
                assert(IS_SUB_4X4(sub_mb_type));
2081
                for(j=0; j<4; j++){
2082
                    int sub_x_offset= x_offset + 2*(j&1);
2083
                    int sub_y_offset= y_offset +   (j&2);
2084
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2085
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2086
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2087
                }
2088
            }
2089
        }
2090
    }
2091
}
2092

    
2093
static void decode_init_vlc(H264Context *h){
2094
    static int done = 0;
2095

    
2096
    if (!done) {
2097
        int i;
2098
        done = 1;
2099

    
2100
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2101
                 &chroma_dc_coeff_token_len [0], 1, 1,
2102
                 &chroma_dc_coeff_token_bits[0], 1, 1);
2103

    
2104
        for(i=0; i<4; i++){
2105
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2106
                     &coeff_token_len [i][0], 1, 1,
2107
                     &coeff_token_bits[i][0], 1, 1);
2108
        }
2109

    
2110
        for(i=0; i<3; i++){
2111
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2112
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2113
                     &chroma_dc_total_zeros_bits[i][0], 1, 1);
2114
        }
2115
        for(i=0; i<15; i++){
2116
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2117
                     &total_zeros_len [i][0], 1, 1,
2118
                     &total_zeros_bits[i][0], 1, 1);
2119
        }
2120

    
2121
        for(i=0; i<6; i++){
2122
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2123
                     &run_len [i][0], 1, 1,
2124
                     &run_bits[i][0], 1, 1);
2125
        }
2126
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2127
                 &run_len [6][0], 1, 1,
2128
                 &run_bits[6][0], 1, 1);
2129
    }
2130
}
2131

    
2132
/**
2133
 * Sets the intra prediction function pointers.
2134
 */
2135
static void init_pred_ptrs(H264Context *h){
2136
//    MpegEncContext * const s = &h->s;
2137

    
2138
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2139
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2140
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2141
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2142
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2143
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2144
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2145
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2146
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2147
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2148
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2149
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2150

    
2151
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2152
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2153
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2154
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2155
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2156
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2157
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2158

    
2159
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2160
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2161
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2162
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2163
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2164
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2165
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2166
}
2167

    
2168
static void free_tables(H264Context *h){
2169
    av_freep(&h->intra4x4_pred_mode);
2170
    av_freep(&h->chroma_pred_mode_table);
2171
    av_freep(&h->cbp_table);
2172
    av_freep(&h->mvd_table[0]);
2173
    av_freep(&h->mvd_table[1]);
2174
    av_freep(&h->non_zero_count);
2175
    av_freep(&h->slice_table_base);
2176
    av_freep(&h->top_border);
2177
    h->slice_table= NULL;
2178

    
2179
    av_freep(&h->mb2b_xy);
2180
    av_freep(&h->mb2b8_xy);
2181
}
2182

    
2183
/**
2184
 * allocates tables.
2185
 * needs widzh/height
2186
 */
2187
static int alloc_tables(H264Context *h){
2188
    MpegEncContext * const s = &h->s;
2189
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2190
    int x,y;
2191

    
2192
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2193

    
2194
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2195
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2196
    CHECKED_ALLOCZ(h->top_border       , s->mb_width * (16+8+8) * sizeof(uint8_t))
2197

    
2198
    if( h->pps.cabac ) {
2199
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2200
        CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2201
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2202
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2203
    }
2204

    
2205
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2206
    h->slice_table= h->slice_table_base + s->mb_stride + 1;
2207

    
2208
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint16_t));
2209
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2210
    for(y=0; y<s->mb_height; y++){
2211
        for(x=0; x<s->mb_width; x++){
2212
            const int mb_xy= x + y*s->mb_stride;
2213
            const int b_xy = 4*x + 4*y*h->b_stride;
2214
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2215
        
2216
            h->mb2b_xy [mb_xy]= b_xy;
2217
            h->mb2b8_xy[mb_xy]= b8_xy;
2218
        }
2219
    }
2220
    
2221
    return 0;
2222
fail:
2223
    free_tables(h);
2224
    return -1;
2225
}
2226

    
2227
static void common_init(H264Context *h){
2228
    MpegEncContext * const s = &h->s;
2229

    
2230
    s->width = s->avctx->width;
2231
    s->height = s->avctx->height;
2232
    s->codec_id= s->avctx->codec->id;
2233
    
2234
    init_pred_ptrs(h);
2235

    
2236
    s->unrestricted_mv=1;
2237
    s->decode=1; //FIXME
2238
}
2239

    
2240
static int decode_init(AVCodecContext *avctx){
2241
    H264Context *h= avctx->priv_data;
2242
    MpegEncContext * const s = &h->s;
2243

    
2244
    MPV_decode_defaults(s);
2245
    
2246
    s->avctx = avctx;
2247
    common_init(h);
2248

    
2249
    s->out_format = FMT_H264;
2250
    s->workaround_bugs= avctx->workaround_bugs;
2251

    
2252
    // set defaults
2253
//    s->decode_mb= ff_h263_decode_mb;
2254
    s->low_delay= 1;
2255
    avctx->pix_fmt= PIX_FMT_YUV420P;
2256

    
2257
    decode_init_vlc(h);
2258
    
2259
    return 0;
2260
}
2261

    
2262
static void frame_start(H264Context *h){
2263
    MpegEncContext * const s = &h->s;
2264
    int i;
2265

    
2266
    MPV_frame_start(s, s->avctx);
2267
    ff_er_frame_start(s);
2268
    h->mmco_index=0;
2269

    
2270
    assert(s->linesize && s->uvlinesize);
2271

    
2272
    for(i=0; i<16; i++){
2273
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2274
        h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2275
    }
2276
    for(i=0; i<4; i++){
2277
        h->block_offset[16+i]=
2278
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2279
    }
2280

    
2281
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2282
}
2283

    
2284
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2285
    MpegEncContext * const s = &h->s;
2286
    int i;
2287
    
2288
    src_y  -=   linesize;
2289
    src_cb -= uvlinesize;
2290
    src_cr -= uvlinesize;
2291

    
2292
    h->left_border[0]= h->top_border[s->mb_x][15];
2293
    for(i=1; i<17; i++){
2294
        h->left_border[i]= src_y[15+i*  linesize];
2295
    }
2296
    
2297
    *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2298
    *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2299

    
2300
    if(!(s->flags&CODEC_FLAG_GRAY)){
2301
        h->left_border[17  ]= h->top_border[s->mb_x][16+7];
2302
        h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2303
        for(i=1; i<9; i++){
2304
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
2305
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2306
        }
2307
        *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2308
        *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2309
    }
2310
}
2311

    
2312
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){
2313
    MpegEncContext * const s = &h->s;
2314
    int temp8, i;
2315
    uint64_t temp64;
2316

    
2317
    src_y  -=   linesize + 1;
2318
    src_cb -= uvlinesize + 1;
2319
    src_cr -= uvlinesize + 1;
2320

    
2321
#define XCHG(a,b,t,xchg)\
2322
t= a;\
2323
if(xchg)\
2324
    a= b;\
2325
b= t;
2326
    
2327
    for(i=0; i<17; i++){
2328
        XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2329
    }
2330
    
2331
    XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2332
    XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2333

    
2334
    if(!(s->flags&CODEC_FLAG_GRAY)){
2335
        for(i=0; i<9; i++){
2336
            XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
2337
            XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2338
        }
2339
        XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2340
        XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2341
    }
2342
}
2343

    
2344
static void hl_decode_mb(H264Context *h){
2345
    MpegEncContext * const s = &h->s;
2346
    const int mb_x= s->mb_x;
2347
    const int mb_y= s->mb_y;
2348
    const int mb_xy= mb_x + mb_y*s->mb_stride;
2349
    const int mb_type= s->current_picture.mb_type[mb_xy];
2350
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2351
    int linesize, uvlinesize /*dct_offset*/;
2352
    int i;
2353

    
2354
    if(!s->decode)
2355
        return;
2356

    
2357
    if(s->mb_skiped){
2358
    }
2359

    
2360
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2361
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2362
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2363

    
2364
    if (h->mb_field_decoding_flag) {
2365
        linesize = s->linesize * 2;
2366
        uvlinesize = s->uvlinesize * 2;
2367
        if(mb_y&1){ //FIXME move out of this func?
2368
            dest_y -= s->linesize*15;
2369
            dest_cb-= s->linesize*7;
2370
            dest_cr-= s->linesize*7;
2371
        }
2372
    } else {
2373
        linesize = s->linesize;
2374
        uvlinesize = s->uvlinesize;
2375
//        dct_offset = s->linesize * 16;
2376
    }
2377

    
2378
    if(IS_INTRA(mb_type)){
2379
        if(h->deblocking_filter)
2380
            xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2381

    
2382
        if(!(s->flags&CODEC_FLAG_GRAY)){
2383
            h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2384
            h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2385
        }
2386

    
2387
        if(IS_INTRA4x4(mb_type)){
2388
            if(!s->encoding){
2389
                for(i=0; i<16; i++){
2390
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2391
                    uint8_t *topright= ptr + 4 - linesize;
2392
                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2393
                    const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2394
                    int tr;
2395

    
2396
                    if(!topright_avail){
2397
                        tr= ptr[3 - linesize]*0x01010101;
2398
                        topright= (uint8_t*) &tr;
2399
                    }else if(i==5 && h->deblocking_filter){
2400
                        tr= *(uint32_t*)h->top_border[mb_x+1];
2401
                        topright= (uint8_t*) &tr;
2402
                    }
2403

    
2404
                    h->pred4x4[ dir ](ptr, topright, linesize);
2405
                    if(h->non_zero_count_cache[ scan8[i] ]){
2406
                        if(s->codec_id == CODEC_ID_H264)
2407
                            h264_add_idct_c(ptr, h->mb + i*16, linesize);
2408
                        else
2409
                            svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2410
                    }
2411
                }
2412
            }
2413
        }else{
2414
            h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2415
            if(s->codec_id == CODEC_ID_H264)
2416
                h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2417
            else
2418
                svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2419
        }
2420
        if(h->deblocking_filter)
2421
            xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2422
    }else if(s->codec_id == CODEC_ID_H264){
2423
        hl_motion(h, dest_y, dest_cb, dest_cr,
2424
                  s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
2425
                  s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2426
    }
2427

    
2428

    
2429
    if(!IS_INTRA4x4(mb_type)){
2430
        if(s->codec_id == CODEC_ID_H264){
2431
            for(i=0; i<16; i++){
2432
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2433
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2434
                    h264_add_idct_c(ptr, h->mb + i*16, linesize);
2435
                }
2436
            }
2437
        }else{
2438
            for(i=0; i<16; i++){
2439
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2440
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2441
                    svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2442
                }
2443
            }
2444
        }
2445
    }
2446

    
2447
    if(!(s->flags&CODEC_FLAG_GRAY)){
2448
        chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2449
        chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2450
        if(s->codec_id == CODEC_ID_H264){
2451
            for(i=16; i<16+4; i++){
2452
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2453
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2454
                    h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2455
                }
2456
            }
2457
            for(i=20; i<20+4; i++){
2458
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2459
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2460
                    h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2461
                }
2462
            }
2463
        }else{
2464
            for(i=16; i<16+4; i++){
2465
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2466
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2467
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2468
                }
2469
            }
2470
            for(i=20; i<20+4; i++){
2471
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2472
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2473
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2474
                }
2475
            }
2476
        }
2477
    }
2478
    if(h->deblocking_filter) {
2479
        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2480
        filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2481
    }
2482
}
2483

    
2484
/**
2485
 * fills the default_ref_list.
2486
 */
2487
static int fill_default_ref_list(H264Context *h){
2488
    MpegEncContext * const s = &h->s;
2489
    int i;
2490
    Picture sorted_short_ref[16];
2491
    
2492
    if(h->slice_type==B_TYPE){
2493
        int out_i;
2494
        int limit= -1;
2495

    
2496
        for(out_i=0; out_i<h->short_ref_count; out_i++){
2497
            int best_i=-1;
2498
            int best_poc=-1;
2499

    
2500
            for(i=0; i<h->short_ref_count; i++){
2501
                const int poc= h->short_ref[i]->poc;
2502
                if(poc > limit && poc < best_poc){
2503
                    best_poc= poc;
2504
                    best_i= i;
2505
                }
2506
            }
2507
            
2508
            assert(best_i != -1);
2509
            
2510
            limit= best_poc;
2511
            sorted_short_ref[out_i]= *h->short_ref[best_i];
2512
        }
2513
    }
2514

    
2515
    if(s->picture_structure == PICT_FRAME){
2516
        if(h->slice_type==B_TYPE){
2517
            const int current_poc= s->current_picture_ptr->poc;
2518
            int list;
2519

    
2520
            for(list=0; list<2; list++){
2521
                int index=0;
2522

    
2523
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2524
                    const int i2= list ? h->short_ref_count - i - 1 : i;
2525
                    const int poc= sorted_short_ref[i2].poc;
2526
                    
2527
                    if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2528

    
2529
                    if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2530
                        h->default_ref_list[list][index  ]= sorted_short_ref[i2];
2531
                        h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2532
                    }
2533
                }
2534

    
2535
                for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2536
                    if(h->long_ref[i]->reference != 3) continue;
2537

    
2538
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
2539
                    h->default_ref_list[ list ][index++].pic_id= i;;
2540
                }
2541
                
2542
                if(h->long_ref_count > 1 && h->short_ref_count==0){
2543
                    Picture temp= h->default_ref_list[1][0];
2544
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
2545
                    h->default_ref_list[1][0] = temp;
2546
                }
2547

    
2548
                if(index < h->ref_count[ list ])
2549
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2550
            }
2551
        }else{
2552
            int index=0;
2553
            for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2554
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2555
                h->default_ref_list[0][index  ]= *h->short_ref[i];
2556
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2557
            }
2558
            for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2559
                if(h->long_ref[i]->reference != 3) continue;
2560
                h->default_ref_list[0][index  ]= *h->long_ref[i];
2561
                h->default_ref_list[0][index++].pic_id= i;;
2562
            }
2563
            if(index < h->ref_count[0])
2564
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2565
        }
2566
    }else{ //FIELD
2567
        if(h->slice_type==B_TYPE){
2568
        }else{
2569
            //FIXME second field balh
2570
        }
2571
    }
2572
    return 0;
2573
}
2574

    
2575
static int decode_ref_pic_list_reordering(H264Context *h){
2576
    MpegEncContext * const s = &h->s;
2577
    int list;
2578
    
2579
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2580
    
2581
    for(list=0; list<2; list++){
2582
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2583

    
2584
        if(get_bits1(&s->gb)){
2585
            int pred= h->curr_pic_num;
2586
            int index;
2587

    
2588
            for(index=0; ; index++){
2589
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2590
                int pic_id;
2591
                int i;
2592
                
2593
                
2594
                if(index >= h->ref_count[list]){
2595
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2596
                    return -1;
2597
                }
2598
                
2599
                if(reordering_of_pic_nums_idc<3){
2600
                    if(reordering_of_pic_nums_idc<2){
2601
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2602

    
2603
                        if(abs_diff_pic_num >= h->max_pic_num){
2604
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2605
                            return -1;
2606
                        }
2607

    
2608
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2609
                        else                                pred+= abs_diff_pic_num;
2610
                        pred &= h->max_pic_num - 1;
2611
                    
2612
                        for(i= h->ref_count[list]-1; i>=index; i--){
2613
                            if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2614
                                break;
2615
                        }
2616
                    }else{
2617
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2618

    
2619
                        for(i= h->ref_count[list]-1; i>=index; i--){
2620
                            if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2621
                                break;
2622
                        }
2623
                    }
2624

    
2625
                    if(i < index){
2626
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2627
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2628
                    }else if(i > index){
2629
                        Picture tmp= h->ref_list[list][i];
2630
                        for(; i>index; i--){
2631
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2632
                        }
2633
                        h->ref_list[list][index]= tmp;
2634
                    }
2635
                }else if(reordering_of_pic_nums_idc==3) 
2636
                    break;
2637
                else{
2638
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2639
                    return -1;
2640
                }
2641
            }
2642
        }
2643

    
2644
        if(h->slice_type!=B_TYPE) break;
2645
    }
2646
    return 0;    
2647
}
2648

    
2649
static int pred_weight_table(H264Context *h){
2650
    MpegEncContext * const s = &h->s;
2651
    int list, i;
2652
    
2653
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2654
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2655

    
2656
    for(list=0; list<2; list++){
2657
        for(i=0; i<h->ref_count[list]; i++){
2658
            int luma_weight_flag, chroma_weight_flag;
2659
            
2660
            luma_weight_flag= get_bits1(&s->gb);
2661
            if(luma_weight_flag){
2662
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
2663
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
2664
            }
2665

    
2666
            chroma_weight_flag= get_bits1(&s->gb);
2667
            if(chroma_weight_flag){
2668
                int j;
2669
                for(j=0; j<2; j++){
2670
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2671
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2672
                }
2673
            }
2674
        }
2675
        if(h->slice_type != B_TYPE) break;
2676
    }
2677
    return 0;
2678
}
2679

    
2680
/**
2681
 * instantaneos decoder refresh.
2682
 */
2683
static void idr(H264Context *h){
2684
    int i;
2685

    
2686
    for(i=0; i<h->long_ref_count; i++){
2687
        h->long_ref[i]->reference=0;
2688
        h->long_ref[i]= NULL;
2689
    }
2690
    h->long_ref_count=0;
2691

    
2692
    for(i=0; i<h->short_ref_count; i++){
2693
        h->short_ref[i]->reference=0;
2694
        h->short_ref[i]= NULL;
2695
    }
2696
    h->short_ref_count=0;
2697
}
2698

    
2699
/**
2700
 *
2701
 * @return the removed picture or NULL if an error occures
2702
 */
2703
static Picture * remove_short(H264Context *h, int frame_num){
2704
    MpegEncContext * const s = &h->s;
2705
    int i;
2706
    
2707
    if(s->avctx->debug&FF_DEBUG_MMCO)
2708
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
2709
    
2710
    for(i=0; i<h->short_ref_count; i++){
2711
        Picture *pic= h->short_ref[i];
2712
        if(s->avctx->debug&FF_DEBUG_MMCO)
2713
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
2714
        if(pic->frame_num == frame_num){
2715
            h->short_ref[i]= NULL;
2716
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2717
            h->short_ref_count--;
2718
            return pic;
2719
        }
2720
    }
2721
    return NULL;
2722
}
2723

    
2724
/**
2725
 *
2726
 * @return the removed picture or NULL if an error occures
2727
 */
2728
static Picture * remove_long(H264Context *h, int i){
2729
    Picture *pic;
2730

    
2731
    if(i >= h->long_ref_count) return NULL;
2732
    pic= h->long_ref[i];
2733
    if(pic==NULL) return NULL;
2734
    
2735
    h->long_ref[i]= NULL;
2736
    memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2737
    h->long_ref_count--;
2738

    
2739
    return pic;
2740
}
2741

    
2742
/**
2743
 * Executes the reference picture marking (memory management control operations).
2744
 */
2745
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2746
    MpegEncContext * const s = &h->s;
2747
    int i;
2748
    int current_is_long=0;
2749
    Picture *pic;
2750
    
2751
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2752
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
2753
        
2754
    for(i=0; i<mmco_count; i++){
2755
        if(s->avctx->debug&FF_DEBUG_MMCO)
2756
            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);
2757

    
2758
        switch(mmco[i].opcode){
2759
        case MMCO_SHORT2UNUSED:
2760
            pic= remove_short(h, mmco[i].short_frame_num);
2761
            if(pic==NULL) return -1;
2762
            pic->reference= 0;
2763
            break;
2764
        case MMCO_SHORT2LONG:
2765
            pic= remove_long(h, mmco[i].long_index);
2766
            if(pic) pic->reference=0;
2767
            
2768
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2769
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
2770
            break;
2771
        case MMCO_LONG2UNUSED:
2772
            pic= remove_long(h, mmco[i].long_index);
2773
            if(pic==NULL) return -1;
2774
            pic->reference= 0;
2775
            break;
2776
        case MMCO_LONG:
2777
            pic= remove_long(h, mmco[i].long_index);
2778
            if(pic) pic->reference=0;
2779
            
2780
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2781
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
2782
            h->long_ref_count++;
2783
            
2784
            current_is_long=1;
2785
            break;
2786
        case MMCO_SET_MAX_LONG:
2787
            assert(mmco[i].long_index <= 16);
2788
            while(mmco[i].long_index < h->long_ref_count){
2789
                pic= remove_long(h, mmco[i].long_index);
2790
                pic->reference=0;
2791
            }
2792
            while(mmco[i].long_index > h->long_ref_count){
2793
                h->long_ref[ h->long_ref_count++ ]= NULL;
2794
            }
2795
            break;
2796
        case MMCO_RESET:
2797
            while(h->short_ref_count){
2798
                pic= remove_short(h, h->short_ref[0]->frame_num);
2799
                pic->reference=0;
2800
            }
2801
            while(h->long_ref_count){
2802
                pic= remove_long(h, h->long_ref_count-1);
2803
                pic->reference=0;
2804
            }
2805
            break;
2806
        default: assert(0);
2807
        }
2808
    }
2809
    
2810
    if(!current_is_long){
2811
        pic= remove_short(h, s->current_picture_ptr->frame_num);
2812
        if(pic){
2813
            pic->reference=0;
2814
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
2815
        }
2816
        
2817
        if(h->short_ref_count)
2818
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2819

    
2820
        h->short_ref[0]= s->current_picture_ptr;
2821
        h->short_ref[0]->long_ref=0;
2822
        h->short_ref_count++;
2823
    }
2824
    
2825
    return 0; 
2826
}
2827

    
2828
static int decode_ref_pic_marking(H264Context *h){
2829
    MpegEncContext * const s = &h->s;
2830
    int i;
2831
    
2832
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2833
        s->broken_link= get_bits1(&s->gb) -1;
2834
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2835
        if(h->mmco[0].long_index == -1)
2836
            h->mmco_index= 0;
2837
        else{
2838
            h->mmco[0].opcode= MMCO_LONG;
2839
            h->mmco_index= 1;
2840
        } 
2841
    }else{
2842
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2843
            for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) { 
2844
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2845

    
2846
                h->mmco[i].opcode= opcode;
2847
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2848
                    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
2849
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2850
                        fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2851
                        return -1;
2852
                    }*/
2853
                }
2854
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2855
                    h->mmco[i].long_index= get_ue_golomb(&s->gb);
2856
                    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){
2857
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
2858
                        return -1;
2859
                    }
2860
                }
2861
                    
2862
                if(opcode > MMCO_LONG){
2863
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
2864
                    return -1;
2865
                }
2866
            }
2867
            h->mmco_index= i;
2868
        }else{
2869
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2870

    
2871
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2872
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2873
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2874
                h->mmco_index= 1;
2875
            }else
2876
                h->mmco_index= 0;
2877
        }
2878
    }
2879
    
2880
    return 0; 
2881
}
2882

    
2883
static int init_poc(H264Context *h){
2884
    MpegEncContext * const s = &h->s;
2885
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2886
    int field_poc[2];
2887

    
2888
    if(h->nal_unit_type == NAL_IDR_SLICE){
2889
        h->frame_num_offset= 0;
2890
    }else{
2891
        if(h->frame_num < h->prev_frame_num)
2892
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2893
        else
2894
            h->frame_num_offset= h->prev_frame_num_offset;
2895
    }
2896

    
2897
    if(h->sps.poc_type==0){
2898
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2899

    
2900
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2901
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2902
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2903
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2904
        else
2905
            h->poc_msb = h->prev_poc_msb;
2906
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2907
        field_poc[0] = 
2908
        field_poc[1] = h->poc_msb + h->poc_lsb;
2909
        if(s->picture_structure == PICT_FRAME) 
2910
            field_poc[1] += h->delta_poc_bottom;
2911
    }else if(h->sps.poc_type==1){
2912
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2913
        int i;
2914

    
2915
        if(h->sps.poc_cycle_length != 0)
2916
            abs_frame_num = h->frame_num_offset + h->frame_num;
2917
        else
2918
            abs_frame_num = 0;
2919

    
2920
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
2921
            abs_frame_num--;
2922
            
2923
        expected_delta_per_poc_cycle = 0;
2924
        for(i=0; i < h->sps.poc_cycle_length; i++)
2925
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2926

    
2927
        if(abs_frame_num > 0){
2928
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2929
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2930

    
2931
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2932
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
2933
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2934
        } else
2935
            expectedpoc = 0;
2936

    
2937
        if(h->nal_ref_idc == 0) 
2938
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2939
        
2940
        field_poc[0] = expectedpoc + h->delta_poc[0];
2941
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2942

    
2943
        if(s->picture_structure == PICT_FRAME)
2944
            field_poc[1] += h->delta_poc[1];
2945
    }else{
2946
        int poc;
2947
        if(h->nal_unit_type == NAL_IDR_SLICE){
2948
            poc= 0;
2949
        }else{
2950
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2951
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2952
        }
2953
        field_poc[0]= poc;
2954
        field_poc[1]= poc;
2955
    }
2956
    
2957
    if(s->picture_structure != PICT_BOTTOM_FIELD)
2958
        s->current_picture_ptr->field_poc[0]= field_poc[0];
2959
    if(s->picture_structure != PICT_TOP_FIELD)
2960
        s->current_picture_ptr->field_poc[1]= field_poc[1];
2961
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2962
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2963

    
2964
    return 0;
2965
}
2966

    
2967
/**
2968
 * decodes a slice header.
2969
 * this will allso call MPV_common_init() and frame_start() as needed
2970
 */
2971
static int decode_slice_header(H264Context *h){
2972
    MpegEncContext * const s = &h->s;
2973
    int first_mb_in_slice, pps_id;
2974
    int num_ref_idx_active_override_flag;
2975
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2976

    
2977
    s->current_picture.reference= h->nal_ref_idc != 0;
2978

    
2979
    first_mb_in_slice= get_ue_golomb(&s->gb);
2980

    
2981
    h->slice_type= get_ue_golomb(&s->gb);
2982
    if(h->slice_type > 9){
2983
        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);
2984
    }
2985
    if(h->slice_type > 4){
2986
        h->slice_type -= 5;
2987
        h->slice_type_fixed=1;
2988
    }else
2989
        h->slice_type_fixed=0;
2990
    
2991
    h->slice_type= slice_type_map[ h->slice_type ];
2992
    
2993
    s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
2994
        
2995
    pps_id= get_ue_golomb(&s->gb);
2996
    if(pps_id>255){
2997
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2998
        return -1;
2999
    }
3000
    h->pps= h->pps_buffer[pps_id];
3001
    if(h->pps.slice_group_count == 0){
3002
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3003
        return -1;
3004
    }
3005

    
3006
    h->sps= h->sps_buffer[ h->pps.sps_id ];
3007
    if(h->sps.log2_max_frame_num == 0){
3008
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3009
        return -1;
3010
    }
3011
    
3012
    s->mb_width= h->sps.mb_width;
3013
    s->mb_height= h->sps.mb_height;
3014
    
3015
    h->b_stride=  s->mb_width*4;
3016
    h->b8_stride= s->mb_width*2;
3017

    
3018
    s->mb_x = first_mb_in_slice % s->mb_width;
3019
    s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
3020
    
3021
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3022
    if(h->sps.frame_mbs_only_flag)
3023
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
3024
    else
3025
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
3026
    
3027
    if (s->context_initialized 
3028
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3029
        free_tables(h);
3030
        MPV_common_end(s);
3031
    }
3032
    if (!s->context_initialized) {
3033
        if (MPV_common_init(s) < 0)
3034
            return -1;
3035

    
3036
        alloc_tables(h);
3037

    
3038
        s->avctx->width = s->width;
3039
        s->avctx->height = s->height;
3040
        s->avctx->sample_aspect_ratio= h->sps.sar;
3041
    }
3042

    
3043
    if(first_mb_in_slice == 0){
3044
        frame_start(h);
3045
    }
3046

    
3047
    s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3048
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3049

    
3050
    if(h->sps.frame_mbs_only_flag){
3051
        s->picture_structure= PICT_FRAME;
3052
    }else{
3053
        if(get_bits1(&s->gb)) //field_pic_flag
3054
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3055
        else
3056
            s->picture_structure= PICT_FRAME;
3057
    }
3058

    
3059
    if(s->picture_structure==PICT_FRAME){
3060
        h->curr_pic_num=   h->frame_num;
3061
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3062
    }else{
3063
        h->curr_pic_num= 2*h->frame_num;
3064
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3065
    }
3066
        
3067
    if(h->nal_unit_type == NAL_IDR_SLICE){
3068
        get_ue_golomb(&s->gb); /* idr_pic_id */
3069
    }
3070
   
3071
    if(h->sps.poc_type==0){
3072
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3073
        
3074
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3075
            h->delta_poc_bottom= get_se_golomb(&s->gb);
3076
        }
3077
    }
3078
    
3079
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3080
        h->delta_poc[0]= get_se_golomb(&s->gb);
3081
        
3082
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3083
            h->delta_poc[1]= get_se_golomb(&s->gb);
3084
    }
3085
    
3086
    init_poc(h);
3087
    
3088
    if(h->pps.redundant_pic_cnt_present){
3089
        h->redundant_pic_count= get_ue_golomb(&s->gb);
3090
    }
3091

    
3092
    //set defaults, might be overriden a few line later
3093
    h->ref_count[0]= h->pps.ref_count[0];
3094
    h->ref_count[1]= h->pps.ref_count[1];
3095

    
3096
    if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3097
        if(h->slice_type == B_TYPE){
3098
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
3099
        }
3100
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
3101
    
3102
        if(num_ref_idx_active_override_flag){
3103
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3104
            if(h->slice_type==B_TYPE)
3105
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3106

    
3107
            if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3108
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3109
                return -1;
3110
            }
3111
        }
3112
    }
3113

    
3114
    if(first_mb_in_slice == 0){
3115
        fill_default_ref_list(h);
3116
    }
3117

    
3118
    decode_ref_pic_list_reordering(h);
3119

    
3120
    if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) 
3121
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3122
        pred_weight_table(h);
3123
    
3124
    if(s->current_picture.reference)
3125
        decode_ref_pic_marking(h);
3126

    
3127
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3128
        h->cabac_init_idc = get_ue_golomb(&s->gb);
3129

    
3130
    h->last_qscale_diff = 0;
3131
    s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3132
    if(s->qscale<0 || s->qscale>51){
3133
        av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3134
        return -1;
3135
    }
3136
    //FIXME qscale / qp ... stuff
3137
    if(h->slice_type == SP_TYPE){
3138
        get_bits1(&s->gb); /* sp_for_switch_flag */
3139
    }
3140
    if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3141
        get_se_golomb(&s->gb); /* slice_qs_delta */
3142
    }
3143

    
3144
    h->deblocking_filter = 1;
3145
    h->slice_alpha_c0_offset = 0;
3146
    h->slice_beta_offset = 0;
3147
    if( h->pps.deblocking_filter_parameters_present ) {
3148
        h->deblocking_filter= get_ue_golomb(&s->gb);
3149
        if(h->deblocking_filter < 2) 
3150
            h->deblocking_filter^= 1; // 1<->0
3151

    
3152
        if( h->deblocking_filter ) {
3153
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3154
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3155
        }
3156
    }
3157

    
3158
#if 0 //FMO
3159
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3160
        slice_group_change_cycle= get_bits(&s->gb, ?);
3161
#endif
3162

    
3163
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3164
        av_log(h->s.avctx, AV_LOG_DEBUG, "mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n", 
3165
               first_mb_in_slice, 
3166
               av_get_pict_type_char(h->slice_type),
3167
               pps_id, h->frame_num,
3168
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3169
               h->ref_count[0], h->ref_count[1],
3170
               s->qscale,
3171
               h->deblocking_filter
3172
               );
3173
    }
3174

    
3175
    return 0;
3176
}
3177

    
3178
/**
3179
 *
3180
 */
3181
static inline int get_level_prefix(GetBitContext *gb){
3182
    unsigned int buf;
3183
    int log;
3184
    
3185
    OPEN_READER(re, gb);
3186
    UPDATE_CACHE(re, gb);
3187
    buf=GET_CACHE(re, gb);
3188
    
3189
    log= 32 - av_log2(buf);
3190
#ifdef TRACE
3191
    print_bin(buf>>(32-log), log);
3192
    printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3193
#endif
3194

    
3195
    LAST_SKIP_BITS(re, gb, log);
3196
    CLOSE_READER(re, gb);
3197

    
3198
    return log-1;
3199
}
3200

    
3201
/**
3202
 * decodes a residual block.
3203
 * @param n block index
3204
 * @param scantable scantable
3205
 * @param max_coeff number of coefficients in the block
3206
 * @return <0 if an error occured
3207
 */
3208
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3209
    MpegEncContext * const s = &h->s;
3210
    const uint16_t *qmul= dequant_coeff[qp];
3211
    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};
3212
    int level[16], run[16];
3213
    int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3214

    
3215
    //FIXME put trailing_onex into the context
3216

    
3217
    if(n == CHROMA_DC_BLOCK_INDEX){
3218
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3219
        total_coeff= coeff_token>>2;
3220
    }else{    
3221
        if(n == LUMA_DC_BLOCK_INDEX){
3222
            total_coeff= pred_non_zero_count(h, 0);
3223
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3224
            total_coeff= coeff_token>>2;
3225
        }else{
3226
            total_coeff= pred_non_zero_count(h, n);
3227
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3228
            total_coeff= coeff_token>>2;
3229
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3230
        }
3231
    }
3232

    
3233
    //FIXME set last_non_zero?
3234

    
3235
    if(total_coeff==0)
3236
        return 0;
3237
        
3238
    trailing_ones= coeff_token&3;
3239
    tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3240
    assert(total_coeff<=16);
3241
    
3242
    for(i=0; i<trailing_ones; i++){
3243
        level[i]= 1 - 2*get_bits1(gb);
3244
    }
3245

    
3246
    suffix_length= total_coeff > 10 && trailing_ones < 3;
3247

    
3248
    for(; i<total_coeff; i++){
3249
        const int prefix= get_level_prefix(gb);
3250
        int level_code, mask;
3251

    
3252
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3253
            if(suffix_length)
3254
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3255
            else
3256
                level_code= (prefix<<suffix_length); //part
3257
        }else if(prefix==14){
3258
            if(suffix_length)
3259
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3260
            else
3261
                level_code= prefix + get_bits(gb, 4); //part
3262
        }else if(prefix==15){
3263
            level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3264
            if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3265
        }else{
3266
            av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3267
            return -1;
3268
        }
3269

    
3270
        if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3271

    
3272
        mask= -(level_code&1);
3273
        level[i]= (((2+level_code)>>1) ^ mask) - mask;
3274

    
3275
        if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3276

    
3277
#if 1
3278
        if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3279
#else        
3280
        if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3281
        /* ? == prefix > 2 or sth */
3282
#endif
3283
        tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3284
    }
3285

    
3286
    if(total_coeff == max_coeff)
3287
        zeros_left=0;
3288
    else{
3289
        if(n == CHROMA_DC_BLOCK_INDEX)
3290
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3291
        else
3292
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3293
    }
3294
    
3295
    for(i=0; i<total_coeff-1; i++){
3296
        if(zeros_left <=0)
3297
            break;
3298
        else if(zeros_left < 7){
3299
            run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3300
        }else{
3301
            run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3302
        }
3303
        zeros_left -= run[i];
3304
    }
3305

    
3306
    if(zeros_left<0){
3307
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3308
        return -1;
3309
    }
3310
    
3311
    for(; i<total_coeff-1; i++){
3312
        run[i]= 0;
3313
    }
3314

    
3315
    run[i]= zeros_left;
3316

    
3317
    coeff_num=-1;
3318
    if(n > 24){
3319
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3320
            int j;
3321

    
3322
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3323
            j= scantable[ coeff_num ];
3324

    
3325
            block[j]= level[i];
3326
        }
3327
    }else{
3328
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into  rundecode?
3329
            int j;
3330

    
3331
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3332
            j= scantable[ coeff_num ];
3333

    
3334
            block[j]= level[i] * qmul[j];
3335
//            printf("%d %d  ", block[j], qmul[j]);
3336
        }
3337
    }
3338
    return 0;
3339
}
3340

    
3341
/**
3342
 * decodes a macroblock
3343
 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3344
 */
3345
static int decode_mb_cavlc(H264Context *h){
3346
    MpegEncContext * const s = &h->s;
3347
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3348
    int mb_type, partition_count, cbp;
3349

    
3350
    s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?    
3351

    
3352
    tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3353
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3354
                down the code */
3355
    if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3356
        if(s->mb_skip_run==-1)
3357
            s->mb_skip_run= get_ue_golomb(&s->gb);
3358
        
3359
        if (s->mb_skip_run--) {
3360
            int mx, my;
3361
            /* skip mb */
3362
//FIXME b frame
3363
            mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3364

    
3365
            memset(h->non_zero_count[mb_xy], 0, 16);
3366
            memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3367

    
3368
            if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3369
                h->mb_field_decoding_flag= get_bits1(&s->gb);
3370
            }
3371

    
3372
            if(h->mb_field_decoding_flag)
3373
                mb_type|= MB_TYPE_INTERLACED;
3374
            
3375
            fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3376
            pred_pskip_motion(h, &mx, &my);
3377
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3378
            fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3379
            write_back_motion(h, mb_type);
3380

    
3381
            s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3382
            s->current_picture.qscale_table[mb_xy]= s->qscale;
3383
            h->slice_table[ mb_xy ]= h->slice_num;
3384

    
3385
            h->prev_mb_skiped= 1;
3386
            return 0;
3387
        }
3388
    }
3389
    if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3390
        if((s->mb_y&1)==0)
3391
            h->mb_field_decoding_flag = get_bits1(&s->gb);
3392
    }else
3393
        h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3394
    
3395
    h->prev_mb_skiped= 0;
3396
    
3397
    mb_type= get_ue_golomb(&s->gb);
3398
    if(h->slice_type == B_TYPE){
3399
        if(mb_type < 23){
3400
            partition_count= b_mb_type_info[mb_type].partition_count;
3401
            mb_type=         b_mb_type_info[mb_type].type;
3402
        }else{
3403
            mb_type -= 23;
3404
            goto decode_intra_mb;
3405
        }
3406
    }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3407
        if(mb_type < 5){
3408
            partition_count= p_mb_type_info[mb_type].partition_count;
3409
            mb_type=         p_mb_type_info[mb_type].type;
3410
        }else{
3411
            mb_type -= 5;
3412
            goto decode_intra_mb;
3413
        }
3414
    }else{
3415
       assert(h->slice_type == I_TYPE);
3416
decode_intra_mb:
3417
        if(mb_type > 25){
3418
            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);
3419
            return -1;
3420
        }
3421
        partition_count=0;
3422
        cbp= i_mb_type_info[mb_type].cbp;
3423
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3424
        mb_type= i_mb_type_info[mb_type].type;
3425
    }
3426

    
3427
    if(h->mb_field_decoding_flag)
3428
        mb_type |= MB_TYPE_INTERLACED;
3429

    
3430
    s->current_picture.mb_type[mb_xy]= mb_type;
3431
    h->slice_table[ mb_xy ]= h->slice_num;
3432
    
3433
    if(IS_INTRA_PCM(mb_type)){
3434
        const uint8_t *ptr;
3435
        int x, y;
3436
        
3437
        // we assume these blocks are very rare so we dont optimize it
3438
        align_get_bits(&s->gb);
3439
        
3440
        ptr= s->gb.buffer + get_bits_count(&s->gb);
3441
    
3442
        for(y=0; y<16; y++){
3443
            const int index= 4*(y&3) + 64*(y>>2);
3444
            for(x=0; x<16; x++){
3445
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3446
            }
3447
        }
3448
        for(y=0; y<8; y++){
3449
            const int index= 256 + 4*(y&3) + 32*(y>>2);
3450
            for(x=0; x<8; x++){
3451
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3452
            }
3453
        }
3454
        for(y=0; y<8; y++){
3455
            const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3456
            for(x=0; x<8; x++){
3457
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3458
            }
3459
        }
3460
    
3461
        skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3462
        
3463
        //FIXME deblock filter, non_zero_count_cache init ...
3464
        memset(h->non_zero_count[mb_xy], 16, 16);
3465
        s->current_picture.qscale_table[mb_xy]= s->qscale;
3466
        
3467
        return 0;
3468
    }
3469
        
3470
    fill_caches(h, mb_type);
3471

    
3472
    //mb_pred
3473
    if(IS_INTRA(mb_type)){
3474
//            init_top_left_availability(h);
3475
            if(IS_INTRA4x4(mb_type)){
3476
                int i;
3477

    
3478
//                fill_intra4x4_pred_table(h);
3479
                for(i=0; i<16; i++){
3480
                    const int mode_coded= !get_bits1(&s->gb);
3481
                    const int predicted_mode=  pred_intra_mode(h, i);
3482
                    int mode;
3483

    
3484
                    if(mode_coded){
3485
                        const int rem_mode= get_bits(&s->gb, 3);
3486
                        if(rem_mode<predicted_mode)
3487
                            mode= rem_mode;
3488
                        else
3489
                            mode= rem_mode + 1;
3490
                    }else{
3491
                        mode= predicted_mode;
3492
                    }
3493
                    
3494
                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3495
                }
3496
                write_back_intra_pred_mode(h);
3497
                if( check_intra4x4_pred_mode(h) < 0)
3498
                    return -1;
3499
            }else{
3500
                h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3501
                if(h->intra16x16_pred_mode < 0)
3502
                    return -1;
3503
            }
3504
            h->chroma_pred_mode= get_ue_golomb(&s->gb);
3505

    
3506
            h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3507
            if(h->chroma_pred_mode < 0)
3508
                return -1;
3509
    }else if(partition_count==4){
3510
        int i, j, sub_partition_count[4], list, ref[2][4];
3511
        
3512
        if(h->slice_type == B_TYPE){
3513
            for(i=0; i<4; i++){
3514
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3515
                if(h->sub_mb_type[i] >=13){
3516
                    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);
3517
                    return -1;
3518
                }
3519
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3520
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3521
            }
3522
        }else{
3523
            assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3524
            for(i=0; i<4; i++){
3525
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3526
                if(h->sub_mb_type[i] >=4){
3527
                    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);
3528
                    return -1;
3529
                }
3530
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3531
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3532
            }
3533
        }
3534
        
3535
        for(list=0; list<2; list++){
3536
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3537
            if(ref_count == 0) continue;
3538
            for(i=0; i<4; i++){
3539
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3540
                    ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3541
                }else{
3542
                 //FIXME
3543
                    ref[list][i] = -1;
3544
                }
3545
            }
3546
        }
3547
        
3548
        for(list=0; list<2; list++){
3549
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3550
            if(ref_count == 0) continue;
3551

    
3552
            for(i=0; i<4; i++){
3553
                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3554
                h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3555

    
3556
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3557
                    const int sub_mb_type= h->sub_mb_type[i];
3558
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3559
                    for(j=0; j<sub_partition_count[i]; j++){
3560
                        int mx, my;
3561
                        const int index= 4*i + block_width*j;
3562
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3563
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3564
                        mx += get_se_golomb(&s->gb);
3565
                        my += get_se_golomb(&s->gb);
3566
                        tprintf("final mv:%d %d\n", mx, my);
3567

    
3568
                        if(IS_SUB_8X8(sub_mb_type)){
3569
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= 
3570
                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3571
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= 
3572
                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3573
                        }else if(IS_SUB_8X4(sub_mb_type)){
3574
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3575
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3576
                        }else if(IS_SUB_4X8(sub_mb_type)){
3577
                            mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3578
                            mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3579
                        }else{
3580
                            assert(IS_SUB_4X4(sub_mb_type));
3581
                            mv_cache[ 0 ][0]= mx;
3582
                            mv_cache[ 0 ][1]= my;
3583
                        }
3584
                    }
3585
                }else{
3586
                    uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3587
                    p[0] = p[1]=
3588
                    p[8] = p[9]= 0;
3589
                }
3590
            }
3591
        }
3592
    }else if(!IS_DIRECT(mb_type)){
3593
        int list, mx, my, i;
3594
         //FIXME we should set ref_idx_l? to 0 if we use that later ...
3595
        if(IS_16X16(mb_type)){
3596
            for(list=0; list<2; list++){
3597
                if(h->ref_count[0]>0){
3598
                    if(IS_DIR(mb_type, 0, list)){
3599
                        const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3600
                        fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3601
                    }
3602
                }
3603
            }
3604
            for(list=0; list<2; list++){
3605
                if(IS_DIR(mb_type, 0, list)){
3606
                    pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3607
                    mx += get_se_golomb(&s->gb);
3608
                    my += get_se_golomb(&s->gb);
3609
                    tprintf("final mv:%d %d\n", mx, my);
3610

    
3611
                    fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
3612
                }
3613
            }
3614
        }
3615
        else if(IS_16X8(mb_type)){
3616
            for(list=0; list<2; list++){
3617
                if(h->ref_count[list]>0){
3618
                    for(i=0; i<2; i++){
3619
                        if(IS_DIR(mb_type, i, list)){
3620
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3621
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3622
                        }
3623
                    }
3624
                }
3625
            }
3626
            for(list=0; list<2; list++){
3627
                for(i=0; i<2; i++){
3628
                    if(IS_DIR(mb_type, i, list)){
3629
                        pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3630
                        mx += get_se_golomb(&s->gb);
3631
                        my += get_se_golomb(&s->gb);
3632
                        tprintf("final mv:%d %d\n", mx, my);
3633

    
3634
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
3635
                    }
3636
                }
3637
            }
3638
        }else{
3639
            assert(IS_8X16(mb_type));
3640
            for(list=0; list<2; list++){
3641
                if(h->ref_count[list]>0){
3642
                    for(i=0; i<2; i++){
3643
                        if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3644
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3645
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3646
                        }
3647
                    }
3648
                }
3649
            }
3650
            for(list=0; list<2; list++){
3651
                for(i=0; i<2; i++){
3652
                    if(IS_DIR(mb_type, i, list)){
3653
                        pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3654
                        mx += get_se_golomb(&s->gb);
3655
                        my += get_se_golomb(&s->gb);
3656
                        tprintf("final mv:%d %d\n", mx, my);
3657

    
3658
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
3659
                    }
3660
                }
3661
            }
3662
        }
3663
    }
3664
    
3665
    if(IS_INTER(mb_type))
3666
        write_back_motion(h, mb_type);
3667
    
3668
    if(!IS_INTRA16x16(mb_type)){
3669
        cbp= get_ue_golomb(&s->gb);
3670
        if(cbp > 47){
3671
            av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3672
            return -1;
3673
        }
3674
        
3675
        if(IS_INTRA4x4(mb_type))
3676
            cbp= golomb_to_intra4x4_cbp[cbp];
3677
        else
3678
            cbp= golomb_to_inter_cbp[cbp];
3679
    }
3680

    
3681
    if(cbp || IS_INTRA16x16(mb_type)){
3682
        int i8x8, i4x4, chroma_idx;
3683
        int chroma_qp, dquant;
3684
        GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3685
        const uint8_t *scan, *dc_scan;
3686
        
3687
//        fill_non_zero_count_cache(h);
3688

    
3689
        if(IS_INTERLACED(mb_type)){
3690
            scan= field_scan;
3691
            dc_scan= luma_dc_field_scan;
3692
        }else{
3693
            scan= zigzag_scan;
3694
            dc_scan= luma_dc_zigzag_scan;
3695
        }
3696

    
3697
        dquant= get_se_golomb(&s->gb);
3698

    
3699
        if( dquant > 25 || dquant < -26 ){
3700
            av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3701
            return -1;
3702
        }
3703
        
3704
        s->qscale += dquant;
3705
        if(((unsigned)s->qscale) > 51){
3706
            if(s->qscale<0) s->qscale+= 52;
3707
            else            s->qscale-= 52;
3708
        }
3709
        
3710
        h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3711
        if(IS_INTRA16x16(mb_type)){
3712
            if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3713
                return -1; //FIXME continue if partotioned and other retirn -1 too
3714
            }
3715

    
3716
            assert((cbp&15) == 0 || (cbp&15) == 15);
3717

    
3718
            if(cbp&15){
3719
                for(i8x8=0; i8x8<4; i8x8++){
3720
                    for(i4x4=0; i4x4<4; i4x4++){
3721
                        const int index= i4x4 + 4*i8x8;
3722
                        if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3723
                            return -1;
3724
                        }
3725
                    }
3726
                }
3727
            }else{
3728
                fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
3729
            }
3730
        }else{
3731
            for(i8x8=0; i8x8<4; i8x8++){
3732
                if(cbp & (1<<i8x8)){
3733
                    for(i4x4=0; i4x4<4; i4x4++){
3734
                        const int index= i4x4 + 4*i8x8;
3735
                        
3736
                        if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3737
                            return -1;
3738
                        }
3739
                    }
3740
                }else{
3741
                    uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3742
                    nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3743
                }
3744
            }
3745
        }
3746
        
3747
        if(cbp&0x30){
3748
            for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3749
                if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3750
                    return -1;
3751
                }
3752
        }
3753

    
3754
        if(cbp&0x20){
3755
            for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3756
                for(i4x4=0; i4x4<4; i4x4++){
3757
                    const int index= 16 + 4*chroma_idx + i4x4;
3758
                    if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3759
                        return -1;
3760
                    }
3761
                }
3762
            }
3763
        }else{
3764
            uint8_t * const nnz= &h->non_zero_count_cache[0];
3765
            nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3766
            nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3767
        }
3768
    }else{
3769
        uint8_t * const nnz= &h->non_zero_count_cache[0];
3770
        fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
3771
        nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3772
        nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3773
    }
3774
    s->current_picture.qscale_table[mb_xy]= s->qscale;
3775
    write_back_non_zero_count(h);
3776

    
3777
    return 0;
3778
}
3779

    
3780
static int decode_cabac_mb_type( H264Context *h ) {
3781
    MpegEncContext * const s = &h->s;
3782

    
3783
    if( h->slice_type == I_TYPE ) {
3784
        const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3785
        int ctx = 0;
3786
        int mb_type;
3787

    
3788
        if( s->mb_x > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-1] ) )
3789
            ctx++;
3790
        if( s->mb_y > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
3791
            ctx++;
3792

    
3793
        if( get_cabac( &h->cabac, &h->cabac_state[3+ctx] ) == 0 )
3794
            return 0;   /* I4x4 */
3795

    
3796
        if( get_cabac_terminate( &h->cabac ) )
3797
            return 25;  /* PCM */
3798

    
3799
        mb_type = 1;    /* I16x16 */
3800
        if( get_cabac( &h->cabac, &h->cabac_state[3+3] ) )
3801
            mb_type += 12;  /* cbp_luma != 0 */
3802

    
3803
        if( get_cabac( &h->cabac, &h->cabac_state[3+4] ) ) {
3804
            if( get_cabac( &h->cabac, &h->cabac_state[3+5] ) )
3805
                mb_type += 4 * 2;   /* cbp_chroma == 2 */
3806
            else
3807
                mb_type += 4 * 1;   /* cbp_chroma == 1 */
3808
        }
3809
        if( get_cabac( &h->cabac, &h->cabac_state[3+6] ) )
3810
            mb_type += 2;
3811
        if( get_cabac( &h->cabac, &h->cabac_state[3+7] ) )
3812
            mb_type += 1;
3813
        return mb_type;
3814

    
3815
    } else if( h->slice_type == P_TYPE ) {
3816
        if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
3817
            /* P-type */
3818
            if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
3819
                if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
3820
                    return 0; /* P_L0_D16x16; */
3821
                else
3822
                    return 3; /* P_8x8; */
3823
            } else {
3824
                if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
3825
                    return 2; /* P_L0_D8x16; */
3826
                else
3827
                    return 1; /* P_L0_D16x8; */
3828
            }
3829
        } else {
3830
            int mb_type;
3831
            /* I-type */
3832
            if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
3833
                return 5+0; /* I_4x4 */
3834
            if( get_cabac_terminate( &h->cabac ) )
3835
                return 5+25; /*I_PCM */
3836
            mb_type = 5+1;    /* I16x16 */
3837
            if( get_cabac( &h->cabac, &h->cabac_state[17+1] ) )
3838
                mb_type += 12;  /* cbp_luma != 0 */
3839

    
3840
            if( get_cabac( &h->cabac, &h->cabac_state[17+2] ) ) {
3841
                if( get_cabac( &h->cabac, &h->cabac_state[17+2] ) )
3842
                    mb_type += 4 * 2;   /* cbp_chroma == 2 */
3843
                else
3844
                    mb_type += 4 * 1;   /* cbp_chroma == 1 */
3845
            }
3846
            if( get_cabac( &h->cabac, &h->cabac_state[17+3] ) )
3847
                mb_type += 2;
3848
            if( get_cabac( &h->cabac, &h->cabac_state[17+3] ) )
3849
                mb_type += 1;
3850

    
3851
            return mb_type;
3852
        }
3853
    } else {
3854
        /* TODO do others frames types */
3855
        return -1;
3856
    }
3857
}
3858

    
3859
static int decode_cabac_mb_skip( H264Context *h) {
3860
    MpegEncContext * const s = &h->s;
3861
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3862
    const int mba_xy = mb_xy - 1;
3863
    const int mbb_xy = mb_xy - s->mb_stride;
3864
    int ctx = 0;
3865

    
3866
    if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mba_xy] ) )
3867
        ctx++;
3868
    if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ) )
3869
        ctx++;
3870

    
3871
    if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
3872
        return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
3873
    else /* B-frame */
3874
        return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
3875
}
3876

    
3877
static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
3878
    int mode = 0;
3879

    
3880
    if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
3881
        return pred_mode;
3882

    
3883
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3884
        mode += 1;
3885
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3886
        mode += 2;
3887
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3888
        mode += 4;
3889
    if( mode >= pred_mode )
3890
        return mode + 1;
3891
    else
3892
        return mode;
3893
}
3894

    
3895
static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
3896
    MpegEncContext * const s = &h->s;
3897
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3898
    const int mba_xy = mb_xy - 1;
3899
    const int mbb_xy = mb_xy - s->mb_stride;
3900

    
3901
    int ctx = 0;
3902

    
3903
    /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
3904
    if( s->mb_x > 0 && h->chroma_pred_mode_table[mba_xy] != 0 )
3905
        ctx++;
3906

    
3907
    if( s->mb_y > 0 && h->chroma_pred_mode_table[mbb_xy] != 0 )
3908
        ctx++;
3909

    
3910
    if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
3911
        return 0;
3912

    
3913
    if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
3914
        return 1;
3915
    if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
3916
        return 2;
3917
    else
3918
        return 3;
3919
}
3920

    
3921
static const uint8_t block_idx_x[16] = {
3922
    0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
3923
};
3924
static const uint8_t block_idx_y[16] = {
3925
    0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
3926
};
3927
static const uint8_t block_idx_xy[4][4] = {
3928
    { 0, 2, 8,  10},
3929
    { 1, 3, 9,  11},
3930
    { 4, 6, 12, 14},
3931
    { 5, 7, 13, 15}
3932
};
3933

    
3934
static int decode_cabac_mb_cbp_luma( H264Context *h) {
3935
    MpegEncContext * const s = &h->s;
3936
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3937

    
3938
    int cbp = 0;
3939
    int i8x8;
3940

    
3941
    h->cbp_table[mb_xy] = 0;  /* FIXME aaahahahah beurk */
3942

    
3943
    for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
3944
        int mba_xy = -1;
3945
        int mbb_xy = -1;
3946
        int x, y;
3947
        int ctx = 0;
3948

    
3949
        x = block_idx_x[4*i8x8];
3950
        y = block_idx_y[4*i8x8];
3951

    
3952
        if( x > 0 )
3953
            mba_xy = mb_xy;
3954
        else if( s->mb_x > 0 )
3955
            mba_xy = mb_xy - 1;
3956

    
3957
        if( y > 0 )
3958
            mbb_xy = mb_xy;
3959
        else if( s->mb_y > 0 )
3960
            mbb_xy = mb_xy - s->mb_stride;
3961

    
3962
        /* No need to test for skip as we put 0 for skip block */
3963
        if( mba_xy >= 0 ) {
3964
            int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
3965
            if( ((h->cbp_table[mba_xy] >> i8x8a)&0x01) == 0 )
3966
                ctx++;
3967
        }
3968

    
3969
        if( mbb_xy >= 0 ) {
3970
            int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
3971
            if( ((h->cbp_table[mbb_xy] >> i8x8b)&0x01) == 0 )
3972
                ctx += 2;
3973
        }
3974

    
3975
        if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
3976
            cbp |= 1 << i8x8;
3977
            h->cbp_table[mb_xy] = cbp;  /* FIXME aaahahahah beurk */
3978
        }
3979
    }
3980
    return cbp;
3981
}
3982
static int decode_cabac_mb_cbp_chroma( H264Context *h) {
3983
    MpegEncContext * const s = &h->s;
3984
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3985
    int ctx;
3986
    int cbp_a, cbp_b;
3987

    
3988
    /* No need to test for skip */
3989
    if( s->mb_x > 0 )
3990
        cbp_a = (h->cbp_table[mb_xy-1]>>4)&0x03;
3991
    else
3992
        cbp_a = -1;
3993

    
3994
    if( s->mb_y > 0 )
3995
        cbp_b = (h->cbp_table[mb_xy-s->mb_stride]>>4)&0x03;
3996
    else
3997
        cbp_b = -1;
3998

    
3999
    ctx = 0;
4000
    if( cbp_a > 0 ) ctx++;
4001
    if( cbp_b > 0 ) ctx += 2;
4002
    if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
4003
        return 0;
4004

    
4005
    ctx = 4;
4006
    if( cbp_a == 2 ) ctx++;
4007
    if( cbp_b == 2 ) ctx += 2;
4008
    if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) )
4009
        return 2;
4010
    else
4011
        return 1;
4012
}
4013
static int decode_cabac_mb_dqp( H264Context *h) {
4014
    MpegEncContext * const s = &h->s;
4015
    int mbn_xy;
4016
    int   ctx = 0;
4017
    int   val = 0;
4018

    
4019
    if( s->mb_x > 0 )
4020
        mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;
4021
    else
4022
        mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;
4023

    
4024
    if( mbn_xy >= 0 && h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
4025
        ctx++;
4026

    
4027
    while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
4028
        if( ctx < 2 )
4029
            ctx = 2;
4030
        else
4031
            ctx = 3;
4032
        val++;
4033
    }
4034

    
4035
    if( val&0x01 )
4036
        return (val + 1)/2;
4037
    else
4038
        return -(val + 1)/2;
4039
}
4040
static int decode_cabac_mb_sub_type( H264Context *h ) {
4041
    if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
4042
        return 0;   /* 8x8 */
4043
    if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
4044
        return 1;   /* 8x4 */
4045
    if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
4046
        return 2;   /* 4x8 */
4047
    return 3;       /* 4x4 */
4048
}
4049

    
4050
static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
4051
    int refa = h->ref_cache[list][scan8[n] - 1];
4052
    int refb = h->ref_cache[list][scan8[n] - 8];
4053
    int ref  = 0;
4054
    int ctx  = 0;
4055

    
4056
    if( refa > 0 )
4057
        ctx++;
4058
    if( refb > 0 )
4059
        ctx += 2;
4060

    
4061
    while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
4062
        ref++;
4063
        if( ctx < 4 )
4064
            ctx = 4;
4065
        else
4066
            ctx = 5;
4067
    }
4068
    return ref;
4069
}
4070

    
4071
static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
4072
    int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
4073
               abs( h->mvd_cache[list][scan8[n] - 8][l] );
4074
    int ctxbase = (l == 0) ? 40 : 47;
4075
    int ctx;
4076
    int mvd = 0;
4077

    
4078
    if( amvd < 3 )
4079
        ctx = 0;
4080
    else if( amvd > 32 )
4081
        ctx = 2;
4082
    else
4083
        ctx = 1;
4084

    
4085
    while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
4086
        mvd++;
4087
        if( ctx < 3 )
4088
            ctx = 3;
4089
        else if( ctx < 6 )
4090
            ctx++;
4091
    }
4092

    
4093
    if( mvd >= 9 ) {
4094
        int k = 3;
4095
        while( get_cabac_bypass( &h->cabac ) ) {
4096
            mvd += 1 << k;
4097
            k++;
4098
        }
4099
        while( k-- ) {
4100
            if( get_cabac_bypass( &h->cabac ) )
4101
                mvd += 1 << k;
4102
        }
4103
    }
4104
    if( mvd != 0 && get_cabac_bypass( &h->cabac ) )
4105
        return -mvd;
4106
    return mvd;
4107
}
4108

    
4109

    
4110
static int get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {
4111
    MpegEncContext * const s = &h->s;
4112
    const int mb_xy  = s->mb_x + s->mb_y*s->mb_stride;
4113
    int mba_xy = -1;
4114
    int mbb_xy = -1;
4115

    
4116
    int nza = -1;
4117
    int nzb = -1;
4118
    int ctx = 0;
4119

    
4120
    if( cat == 0 ) {
4121
        if( s->mb_x > 0 ) {
4122
            mba_xy = mb_xy - 1;
4123
            if( IS_INTRA16x16(s->current_picture.mb_type[mba_xy] ) )
4124
                    nza = h->cbp_table[mba_xy]&0x100;
4125
        }
4126
        if( s->mb_y > 0 ) {
4127
            mbb_xy = mb_xy - s->mb_stride;
4128
            if( IS_INTRA16x16(s->current_picture.mb_type[mbb_xy] ) )
4129
                    nzb = h->cbp_table[mbb_xy]&0x100;
4130
        }
4131
    } else if( cat == 1 || cat == 2 ) {
4132
        int i8x8a, i8x8b;
4133
        int x, y;
4134

    
4135
        x = block_idx_x[idx];
4136
        y = block_idx_y[idx];
4137

    
4138
        if( x > 0 )
4139
            mba_xy = mb_xy;
4140
        else if( s->mb_x > 0 )
4141
            mba_xy = mb_xy - 1;
4142

    
4143
        if( y > 0 )
4144
            mbb_xy = mb_xy;
4145
        else if( s->mb_y > 0 )
4146
            mbb_xy = mb_xy - s->mb_stride;
4147

    
4148
        /* No need to test for skip */
4149
        if( mba_xy >= 0 ) {
4150
            i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4151

    
4152
            if( !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
4153
                ((h->cbp_table[mba_xy]&0x0f)>>i8x8a))
4154
                nza = h->non_zero_count_cache[scan8[idx] - 1];
4155
        }
4156

    
4157
        if( mbb_xy >= 0 ) {
4158
            i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4159

    
4160
            if( !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
4161
                ((h->cbp_table[mbb_xy]&0x0f)>>i8x8b))
4162
                nzb = h->non_zero_count_cache[scan8[idx] - 8];
4163
        }
4164
    } else if( cat == 3 ) {
4165
        if( s->mb_x > 0 ) {
4166
            mba_xy = mb_xy - 1;
4167

    
4168
            if( !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
4169
                (h->cbp_table[mba_xy]&0x30) )
4170
                nza = (h->cbp_table[mba_xy]>>(6+idx))&0x01;
4171
        }
4172
        if( s->mb_y > 0 ) {
4173
            mbb_xy = mb_xy - s->mb_stride;
4174

    
4175
            if( !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
4176
                (h->cbp_table[mbb_xy]&0x30) )
4177
                nzb = (h->cbp_table[mbb_xy]>>(6+idx))&0x01;
4178
        }
4179
    } else if( cat == 4 ) {
4180
        int idxc = idx % 4 ;
4181
        if( idxc == 1 || idxc == 3 )
4182
            mba_xy = mb_xy;
4183
        else if( s->mb_x > 0 )
4184
            mba_xy = mb_xy -1;
4185

    
4186
        if( idxc == 2 || idxc == 3 )
4187
            mbb_xy = mb_xy;
4188
        else if( s->mb_y > 0 )
4189
            mbb_xy = mb_xy - s->mb_stride;
4190

    
4191
        if( mba_xy >= 0 &&
4192
            !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
4193
            (h->cbp_table[mba_xy]&0x30) == 0x20 )
4194
            nza = h->non_zero_count_cache[scan8[16+idx] - 1];
4195

    
4196
        if( mbb_xy >= 0 &&
4197
            !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
4198
            (h->cbp_table[mbb_xy]&0x30) == 0x20 )
4199
            nzb = h->non_zero_count_cache[scan8[16+idx] - 8];
4200
    }
4201

    
4202
    if( ( mba_xy < 0 && IS_INTRA( s->current_picture.mb_type[mb_xy] ) ) ||
4203
        ( mba_xy >= 0 && IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) ) ||
4204
          nza > 0 )
4205
        ctx++;
4206

    
4207
    if( ( mbb_xy < 0 && IS_INTRA( s->current_picture.mb_type[mb_xy] ) ) ||
4208
        ( mbb_xy >= 0 && IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) ) ||
4209
          nzb > 0 )
4210
        ctx += 2;
4211

    
4212
    return ctx + 4 * cat;
4213
}
4214

    
4215
static int decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, int qp, int max_coeff) {
4216
    const int mb_xy  = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
4217
    const uint16_t *qmul= dequant_coeff[qp];
4218
    static const int significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
4219
    static const int last_significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
4220
    static const int coeff_abs_level_m1_offset[5] = { 0, 10, 20, 30, 39 };
4221

    
4222
    int coeff[16];
4223

    
4224
    int last = 0;
4225
    int coeff_count = 0;
4226
    int nz[16] = {0};
4227
    int i;
4228

    
4229
    int abslevel1 = 0;
4230
    int abslevelgt1 = 0;
4231

    
4232
    /* cat: 0-> DC 16x16  n = 0
4233
     *      1-> AC 16x16  n = luma4x4idx
4234
     *      2-> Luma4x4   n = luma4x4idx
4235
     *      3-> DC Chroma n = iCbCr
4236
     *      4-> AC Chroma n = 4 * iCbCr + chroma4x4idx
4237
     */
4238

    
4239
    /* read coded block flag */
4240
    if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {
4241
        if( cat == 1 || cat == 2 )
4242
            h->non_zero_count_cache[scan8[n]] = 0;
4243
        else if( cat == 4 )
4244
            h->non_zero_count_cache[scan8[16+n]] = 0;
4245

    
4246
        return 0;
4247
    }
4248

    
4249
    while( last < max_coeff - 1 ) {
4250
        int ctx = FFMIN( last, max_coeff - 2 );
4251

    
4252
        if( get_cabac( &h->cabac, &h->cabac_state[105+significant_coeff_flag_offset[cat]+ctx] ) == 0 ) {
4253
            nz[last++] = 0;
4254
        }
4255
        else {
4256
            nz[last++] = 1;
4257
            coeff_count++;
4258
            if( get_cabac( &h->cabac, &h->cabac_state[166+last_significant_coeff_flag_offset[cat]+ctx] ) ) {
4259
                while( last < max_coeff ) {
4260
                    nz[last++] = 0;
4261
                }
4262
                break;
4263
            }
4264
        }
4265
    }
4266
    if( last == max_coeff -1 ) {
4267
        nz[last++] = 1;
4268
        coeff_count++;
4269
    }
4270

    
4271
    if( cat == 0 && coeff_count > 0 )
4272
        h->cbp_table[mb_xy] |= 0x100;
4273
    else if( cat == 1 || cat == 2 )
4274
        h->non_zero_count_cache[scan8[n]] = coeff_count;
4275
    else if( cat == 3 && coeff_count > 0 )
4276
        h->cbp_table[mb_xy] |= 0x40 << n;
4277
    else if( cat == 4 )
4278
        h->non_zero_count_cache[scan8[16+n]] = coeff_count;
4279

    
4280
    for( i = coeff_count - 1; i >= 0; i-- ) {
4281
        int coeff_abs_m1;
4282

    
4283
        int ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 + 1 )) + coeff_abs_level_m1_offset[cat];
4284

    
4285
        if( get_cabac( &h->cabac, &h->cabac_state[227+ctx] ) == 0 ) {
4286
            coeff_abs_m1 = 0;
4287
        } else {
4288
            coeff_abs_m1 = 1;
4289
            ctx = 5 + FFMIN( 4, abslevelgt1 ) + coeff_abs_level_m1_offset[cat];
4290
            while( coeff_abs_m1 < 14 && get_cabac( &h->cabac, &h->cabac_state[227+ctx] ) ) {
4291
                coeff_abs_m1++;
4292
            }
4293
        }
4294

    
4295
        if( coeff_abs_m1 >= 14 ) {
4296
            int j = 0;
4297
            while( get_cabac_bypass( &h->cabac ) ) {
4298
                coeff_abs_m1 += 1 << j;
4299
                j++;
4300
            }
4301

    
4302
            while( j-- ) {
4303
                if( get_cabac_bypass( &h->cabac ) )
4304
                    coeff_abs_m1 += 1 << j ;
4305
            }
4306
        }
4307
        if( get_cabac_bypass( &h->cabac ) )
4308
            coeff[i] = -1 *( coeff_abs_m1 + 1 );
4309
        else
4310
            coeff[i] = coeff_abs_m1 + 1;
4311

    
4312
        if( coeff_abs_m1 == 0 )
4313
            abslevel1++;
4314
        else
4315
            abslevelgt1++;
4316
    }
4317

    
4318
    if( cat == 0 || cat == 3 ) { /* DC */
4319
        int j;
4320
        for( i = 0, j = 0; j < coeff_count; i++ ) {
4321
            if( nz[i] ) {
4322
                block[scantable[i]] = coeff[j];
4323

    
4324
                j++;
4325
            }
4326
        }
4327

    
4328
    } else { /* AC */
4329
        int j;
4330
        for( i = 0, j = 0; j < coeff_count; i++ ) {
4331
            if( nz[i] ) {
4332
                block[scantable[i]] = coeff[j] * qmul[scantable[i]];
4333

    
4334
                j++;
4335
            }
4336
        }
4337
    }
4338
    return 0;
4339
}
4340

    
4341
/**
4342
 * decodes a macroblock
4343
 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4344
 */
4345
static int decode_mb_cabac(H264Context *h) {
4346
    MpegEncContext * const s = &h->s;
4347
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4348
    int mb_type, partition_count, cbp = 0;
4349

    
4350
    s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?)
4351

    
4352
    if( h->slice_type == B_TYPE ) {
4353
        av_log( h->s.avctx, AV_LOG_ERROR, "B-frame not supported with CABAC\n" );
4354
        return -1;
4355
    }
4356
    if( h->sps.mb_aff ) {
4357
        av_log( h->s.avctx, AV_LOG_ERROR, "Fields not supported with CABAC\n" );
4358
        return -1;
4359
    }
4360

    
4361
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
4362
        /* read skip flags */
4363
        if( decode_cabac_mb_skip( h ) ) {
4364
            int mx, my;
4365

    
4366
            /* skip mb */
4367
            mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4368

    
4369
            memset(h->non_zero_count[mb_xy], 0, 16);
4370
            memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4371
#if 0
4372
            if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
4373
                h->mb_field_decoding_flag= get_bits1(&s->gb);
4374
            }
4375
            if(h->mb_field_decoding_flag)
4376
                mb_type|= MB_TYPE_INTERLACED;
4377
#endif
4378

    
4379
            fill_caches(h, mb_type); //FIXME check what is needed and what not ...
4380
            pred_pskip_motion(h, &mx, &my);
4381
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4382
            fill_rectangle(  h->mvd_cache[0][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
4383
            fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4384
            write_back_motion(h, mb_type);
4385

    
4386
            s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
4387
            s->current_picture.qscale_table[mb_xy]= s->qscale;
4388
            h->slice_table[ mb_xy ]= h->slice_num;
4389
            h->cbp_table[mb_xy] = 0;
4390
            h->chroma_pred_mode_table[mb_xy] = 0;
4391
            h->last_qscale_diff = 0;
4392

    
4393
            h->prev_mb_skiped= 1;
4394

    
4395
            return 0;
4396

    
4397
        }
4398
    }
4399
    h->prev_mb_skiped = 0;
4400

    
4401
    if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {
4402
        av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );
4403
        return -1;
4404
    }
4405

    
4406
    if( h->slice_type == P_TYPE ) {
4407
        if( mb_type < 5) {
4408
            partition_count= p_mb_type_info[mb_type].partition_count;
4409
            mb_type=         p_mb_type_info[mb_type].type;
4410
        } else {
4411
            mb_type -= 5;
4412
            goto decode_intra_mb;
4413
        }
4414
    } else {
4415
       assert(h->slice_type == I_TYPE);
4416
decode_intra_mb:
4417
        partition_count = 0;
4418
        cbp= i_mb_type_info[mb_type].cbp;
4419
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4420
        mb_type= i_mb_type_info[mb_type].type;
4421
    }
4422
#if 0
4423
    if(h->mb_field_decoding_flag)
4424
        mb_type |= MB_TYPE_INTERLACED;
4425
#endif
4426

    
4427
    s->current_picture.mb_type[mb_xy]= mb_type;
4428
    h->slice_table[ mb_xy ]= h->slice_num;
4429

    
4430
    if(IS_INTRA_PCM(mb_type)) {
4431
        /* TODO */
4432
        h->cbp_table[mb_xy] = 0xf +4*2;
4433
        h->chroma_pred_mode_table[mb_xy] = 0;
4434
        s->current_picture.qscale_table[mb_xy]= s->qscale;
4435
        return -1;
4436
    }
4437

    
4438
    fill_caches(h, mb_type);
4439

    
4440
    if( IS_INTRA( mb_type ) ) {
4441
        if( IS_INTRA4x4( mb_type ) ) {
4442
            int i;
4443
            for( i = 0; i < 16; i++ ) {
4444
                int pred = pred_intra_mode( h, i );
4445
                h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
4446

    
4447
                //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
4448
            }
4449
            write_back_intra_pred_mode(h);
4450
            if( check_intra4x4_pred_mode(h) < 0 ) return -1;
4451
        } else {
4452
            h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
4453
            if( h->intra16x16_pred_mode < 0 ) return -1;
4454
        }
4455
        h->chroma_pred_mode_table[mb_xy] =
4456
            h->chroma_pred_mode          = decode_cabac_mb_chroma_pre_mode( h );
4457

    
4458
        h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );
4459
        if( h->chroma_pred_mode < 0 ) return -1;
4460
    } else if( partition_count == 4 ) {
4461
        int i, j, sub_partition_count[4], list, ref[2][4];
4462

    
4463
        /* Only P-frame */
4464
        for( i = 0; i < 4; i++ ) {
4465
            h->sub_mb_type[i] = decode_cabac_mb_sub_type( h );
4466
            sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4467
            h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4468
        }
4469

    
4470
        for( list = 0; list < 2; list++ ) {
4471
            if( h->ref_count[list] > 0 ) {
4472
                for( i = 0; i < 4; i++ ) {
4473
                    if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
4474
                        if( h->ref_count[list] > 1 )
4475
                            ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
4476
                        else
4477
                            ref[list][i] = 0;
4478
                    } else {
4479
                        ref[list][i] = -1;
4480
                    }
4481
                    h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4482
                    h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4483
                }
4484
            }
4485
        }
4486

    
4487
        for(list=0; list<2; list++){
4488

    
4489
            for(i=0; i<4; i++){
4490
                //h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4491
                //h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4492

    
4493
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
4494
                    const int sub_mb_type= h->sub_mb_type[i];
4495
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4496
                    for(j=0; j<sub_partition_count[i]; j++){
4497
                        int mpx, mpy;
4498
                        int mx, my;
4499
                        const int index= 4*i + block_width*j;
4500
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4501
                        int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
4502
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
4503

    
4504
                        mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
4505
                        my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
4506
                        tprintf("final mv:%d %d\n", mx, my);
4507

    
4508
                        if(IS_SUB_8X8(sub_mb_type)){
4509
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
4510
                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4511
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
4512
                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4513

    
4514
                            mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=
4515
                            mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
4516
                            mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=
4517
                            mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
4518
                        }else if(IS_SUB_8X4(sub_mb_type)){
4519
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4520
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4521

    
4522
                            mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;
4523
                            mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;
4524
                        }else if(IS_SUB_4X8(sub_mb_type)){
4525
                            mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4526
                            mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4527

    
4528
                            mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;
4529
                            mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;
4530
                        }else{
4531
                            assert(IS_SUB_4X4(sub_mb_type));
4532
                            mv_cache[ 0 ][0]= mx;
4533
                            mv_cache[ 0 ][1]= my;
4534

    
4535
                            mvd_cache[ 0 ][0]= mx - mpx;
4536
                            mvd_cache[ 0 ][1]= my - mpy;
4537
                        }
4538
                    }
4539
                }else{
4540
                    uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4541
                    uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];
4542
                    p[0] = p[1] = p[8] = p[9] = 0;
4543
                    pd[0]= pd[1]= pd[8]= pd[9]= 0;
4544
                }
4545
            }
4546
        }
4547
    } else if( !IS_DIRECT(mb_type) ) {
4548
        int list, mx, my, i, mpx, mpy;
4549
        if(IS_16X16(mb_type)){
4550
            for(list=0; list<2; list++){
4551