Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ e4cb187d

History | View | Annotate | Download (207 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
                continue;
1069
            }else //next start code
1070
                break;
1071
        }
1072

    
1073
        dst[di++]= src[si++];
1074
    }
1075

    
1076
    *dst_length= di;
1077
    *consumed= si + 1;//+1 for the header
1078
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1079
    return dst;
1080
}
1081

    
1082
#if 0
1083
/**
1084
 * @param src the data which should be escaped
1085
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1086
 * @param length the length of the src data
1087
 * @param dst_length the length of the dst array
1088
 * @returns length of escaped data in bytes or -1 if an error occured
1089
 */
1090
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1091
    int i, escape_count, si, di;
1092
    uint8_t *temp;
1093
    
1094
    assert(length>=0);
1095
    assert(dst_length>0);
1096
    
1097
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1098

1099
    if(length==0) return 1;
1100

1101
    escape_count= 0;
1102
    for(i=0; i<length; i+=2){
1103
        if(src[i]) continue;
1104
        if(i>0 && src[i-1]==0) 
1105
            i--;
1106
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1107
            escape_count++;
1108
            i+=2;
1109
        }
1110
    }
1111
    
1112
    if(escape_count==0){ 
1113
        if(dst+1 != src)
1114
            memcpy(dst+1, src, length);
1115
        return length + 1;
1116
    }
1117
    
1118
    if(length + escape_count + 1> dst_length)
1119
        return -1;
1120

1121
    //this should be damn rare (hopefully)
1122

1123
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1124
    temp= h->rbsp_buffer;
1125
//printf("encoding esc\n");
1126
    
1127
    si= 0;
1128
    di= 0;
1129
    while(si < length){
1130
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1131
            temp[di++]= 0; si++;
1132
            temp[di++]= 0; si++;
1133
            temp[di++]= 3; 
1134
            temp[di++]= src[si++];
1135
        }
1136
        else
1137
            temp[di++]= src[si++];
1138
    }
1139
    memcpy(dst+1, temp, length+escape_count);
1140
    
1141
    assert(di == length+escape_count);
1142
    
1143
    return di + 1;
1144
}
1145

1146
/**
1147
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1148
 */
1149
static void encode_rbsp_trailing(PutBitContext *pb){
1150
    int length;
1151
    put_bits(pb, 1, 1);
1152
    length= (-put_bits_count(pb))&7;
1153
    if(length) put_bits(pb, length, 0);
1154
}
1155
#endif
1156

    
1157
/**
1158
 * identifies the exact end of the bitstream
1159
 * @return the length of the trailing, or 0 if damaged
1160
 */
1161
static int decode_rbsp_trailing(uint8_t *src){
1162
    int v= *src;
1163
    int r;
1164

    
1165
    tprintf("rbsp trailing %X\n", v);
1166

    
1167
    for(r=1; r<9; r++){
1168
        if(v&1) return r;
1169
        v>>=1;
1170
    }
1171
    return 0;
1172
}
1173

    
1174
/**
1175
 * idct tranforms the 16 dc values and dequantize them.
1176
 * @param qp quantization parameter
1177
 */
1178
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1179
    const int qmul= dequant_coeff[qp][0];
1180
#define stride 16
1181
    int i;
1182
    int temp[16]; //FIXME check if this is a good idea
1183
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1184
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1185

    
1186
//memset(block, 64, 2*256);
1187
//return;
1188
    for(i=0; i<4; i++){
1189
        const int offset= y_offset[i];
1190
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1191
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1192
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1193
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1194

    
1195
        temp[4*i+0]= z0+z3;
1196
        temp[4*i+1]= z1+z2;
1197
        temp[4*i+2]= z1-z2;
1198
        temp[4*i+3]= z0-z3;
1199
    }
1200

    
1201
    for(i=0; i<4; i++){
1202
        const int offset= x_offset[i];
1203
        const int z0= temp[4*0+i] + temp[4*2+i];
1204
        const int z1= temp[4*0+i] - temp[4*2+i];
1205
        const int z2= temp[4*1+i] - temp[4*3+i];
1206
        const int z3= temp[4*1+i] + temp[4*3+i];
1207

    
1208
        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1209
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1210
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1211
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1212
    }
1213
}
1214

    
1215
#if 0
1216
/**
1217
 * dct tranforms the 16 dc values.
1218
 * @param qp quantization parameter ??? FIXME
1219
 */
1220
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1221
//    const int qmul= dequant_coeff[qp][0];
1222
    int i;
1223
    int temp[16]; //FIXME check if this is a good idea
1224
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1225
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1226

1227
    for(i=0; i<4; i++){
1228
        const int offset= y_offset[i];
1229
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1230
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1231
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1232
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1233

1234
        temp[4*i+0]= z0+z3;
1235
        temp[4*i+1]= z1+z2;
1236
        temp[4*i+2]= z1-z2;
1237
        temp[4*i+3]= z0-z3;
1238
    }
1239

1240
    for(i=0; i<4; i++){
1241
        const int offset= x_offset[i];
1242
        const int z0= temp[4*0+i] + temp[4*2+i];
1243
        const int z1= temp[4*0+i] - temp[4*2+i];
1244
        const int z2= temp[4*1+i] - temp[4*3+i];
1245
        const int z3= temp[4*1+i] + temp[4*3+i];
1246

1247
        block[stride*0 +offset]= (z0 + z3)>>1;
1248
        block[stride*2 +offset]= (z1 + z2)>>1;
1249
        block[stride*8 +offset]= (z1 - z2)>>1;
1250
        block[stride*10+offset]= (z0 - z3)>>1;
1251
    }
1252
}
1253
#endif
1254

    
1255
#undef xStride
1256
#undef stride
1257

    
1258
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1259
    const int qmul= dequant_coeff[qp][0];
1260
    const int stride= 16*2;
1261
    const int xStride= 16;
1262
    int a,b,c,d,e;
1263

    
1264
    a= block[stride*0 + xStride*0];
1265
    b= block[stride*0 + xStride*1];
1266
    c= block[stride*1 + xStride*0];
1267
    d= block[stride*1 + xStride*1];
1268

    
1269
    e= a-b;
1270
    a= a+b;
1271
    b= c-d;
1272
    c= c+d;
1273

    
1274
    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1275
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1276
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1277
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1278
}
1279

    
1280
#if 0
1281
static void chroma_dc_dct_c(DCTELEM *block){
1282
    const int stride= 16*2;
1283
    const int xStride= 16;
1284
    int a,b,c,d,e;
1285

1286
    a= block[stride*0 + xStride*0];
1287
    b= block[stride*0 + xStride*1];
1288
    c= block[stride*1 + xStride*0];
1289
    d= block[stride*1 + xStride*1];
1290

1291
    e= a-b;
1292
    a= a+b;
1293
    b= c-d;
1294
    c= c+d;
1295

1296
    block[stride*0 + xStride*0]= (a+c);
1297
    block[stride*0 + xStride*1]= (e+b);
1298
    block[stride*1 + xStride*0]= (a-c);
1299
    block[stride*1 + xStride*1]= (e-b);
1300
}
1301
#endif
1302

    
1303
/**
1304
 * gets the chroma qp.
1305
 */
1306
static inline int get_chroma_qp(H264Context *h, int qscale){
1307
    
1308
    return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1309
}
1310

    
1311

    
1312
/**
1313
 *
1314
 */
1315
static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1316
    int i;
1317
    uint8_t *cm = cropTbl + MAX_NEG_CROP;
1318

    
1319
    block[0] += 32;
1320

    
1321
    for(i=0; i<4; i++){
1322
        const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
1323
        const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
1324
        const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
1325
        const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
1326

    
1327
        block[0 + 4*i]= z0 + z3;
1328
        block[1 + 4*i]= z1 + z2;
1329
        block[2 + 4*i]= z1 - z2;
1330
        block[3 + 4*i]= z0 - z3;
1331
    }
1332

    
1333
    for(i=0; i<4; i++){
1334
        const int z0=  block[i + 4*0]     +  block[i + 4*2];
1335
        const int z1=  block[i + 4*0]     -  block[i + 4*2];
1336
        const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
1337
        const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
1338

    
1339
        dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1340
        dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1341
        dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1342
        dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1343
    }
1344
}
1345

    
1346
#if 0
1347
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1348
    int i;
1349
    //FIXME try int temp instead of block
1350
    
1351
    for(i=0; i<4; i++){
1352
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1353
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1354
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1355
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1356
        const int z0= d0 + d3;
1357
        const int z3= d0 - d3;
1358
        const int z1= d1 + d2;
1359
        const int z2= d1 - d2;
1360
        
1361
        block[0 + 4*i]=   z0 +   z1;
1362
        block[1 + 4*i]= 2*z3 +   z2;
1363
        block[2 + 4*i]=   z0 -   z1;
1364
        block[3 + 4*i]=   z3 - 2*z2;
1365
    }    
1366

1367
    for(i=0; i<4; i++){
1368
        const int z0= block[0*4 + i] + block[3*4 + i];
1369
        const int z3= block[0*4 + i] - block[3*4 + i];
1370
        const int z1= block[1*4 + i] + block[2*4 + i];
1371
        const int z2= block[1*4 + i] - block[2*4 + i];
1372
        
1373
        block[0*4 + i]=   z0 +   z1;
1374
        block[1*4 + i]= 2*z3 +   z2;
1375
        block[2*4 + i]=   z0 -   z1;
1376
        block[3*4 + i]=   z3 - 2*z2;
1377
    }
1378
}
1379
#endif
1380

    
1381
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1382
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1383
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1384
    int i;
1385
    const int * const quant_table= quant_coeff[qscale];
1386
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1387
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1388
    const unsigned int threshold2= (threshold1<<1);
1389
    int last_non_zero;
1390

    
1391
    if(seperate_dc){
1392
        if(qscale<=18){
1393
            //avoid overflows
1394
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1395
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1396
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1397

    
1398
            int level= block[0]*quant_coeff[qscale+18][0];
1399
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1400
                if(level>0){
1401
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1402
                    block[0]= level;
1403
                }else{
1404
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1405
                    block[0]= -level;
1406
                }
1407
//                last_non_zero = i;
1408
            }else{
1409
                block[0]=0;
1410
            }
1411
        }else{
1412
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1413
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1414
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1415

    
1416
            int level= block[0]*quant_table[0];
1417
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1418
                if(level>0){
1419
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1420
                    block[0]= level;
1421
                }else{
1422
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1423
                    block[0]= -level;
1424
                }
1425
//                last_non_zero = i;
1426
            }else{
1427
                block[0]=0;
1428
            }
1429
        }
1430
        last_non_zero= 0;
1431
        i=1;
1432
    }else{
1433
        last_non_zero= -1;
1434
        i=0;
1435
    }
1436

    
1437
    for(; i<16; i++){
1438
        const int j= scantable[i];
1439
        int level= block[j]*quant_table[j];
1440

    
1441
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1442
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1443
        if(((unsigned)(level+threshold1))>threshold2){
1444
            if(level>0){
1445
                level= (bias + level)>>QUANT_SHIFT;
1446
                block[j]= level;
1447
            }else{
1448
                level= (bias - level)>>QUANT_SHIFT;
1449
                block[j]= -level;
1450
            }
1451
            last_non_zero = i;
1452
        }else{
1453
            block[j]=0;
1454
        }
1455
    }
1456

    
1457
    return last_non_zero;
1458
}
1459

    
1460
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1461
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1462
    ((uint32_t*)(src+0*stride))[0]= a;
1463
    ((uint32_t*)(src+1*stride))[0]= a;
1464
    ((uint32_t*)(src+2*stride))[0]= a;
1465
    ((uint32_t*)(src+3*stride))[0]= a;
1466
}
1467

    
1468
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1469
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1470
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1471
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1472
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1473
}
1474

    
1475
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1476
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1477
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1478
    
1479
    ((uint32_t*)(src+0*stride))[0]= 
1480
    ((uint32_t*)(src+1*stride))[0]= 
1481
    ((uint32_t*)(src+2*stride))[0]= 
1482
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1483
}
1484

    
1485
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1486
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1487
    
1488
    ((uint32_t*)(src+0*stride))[0]= 
1489
    ((uint32_t*)(src+1*stride))[0]= 
1490
    ((uint32_t*)(src+2*stride))[0]= 
1491
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1492
}
1493

    
1494
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1495
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1496
    
1497
    ((uint32_t*)(src+0*stride))[0]= 
1498
    ((uint32_t*)(src+1*stride))[0]= 
1499
    ((uint32_t*)(src+2*stride))[0]= 
1500
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1501
}
1502

    
1503
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1504
    ((uint32_t*)(src+0*stride))[0]= 
1505
    ((uint32_t*)(src+1*stride))[0]= 
1506
    ((uint32_t*)(src+2*stride))[0]= 
1507
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1508
}
1509

    
1510

    
1511
#define LOAD_TOP_RIGHT_EDGE\
1512
    const int t4= topright[0];\
1513
    const int t5= topright[1];\
1514
    const int t6= topright[2];\
1515
    const int t7= topright[3];\
1516

    
1517
#define LOAD_LEFT_EDGE\
1518
    const int l0= src[-1+0*stride];\
1519
    const int l1= src[-1+1*stride];\
1520
    const int l2= src[-1+2*stride];\
1521
    const int l3= src[-1+3*stride];\
1522

    
1523
#define LOAD_TOP_EDGE\
1524
    const int t0= src[ 0-1*stride];\
1525
    const int t1= src[ 1-1*stride];\
1526
    const int t2= src[ 2-1*stride];\
1527
    const int t3= src[ 3-1*stride];\
1528

    
1529
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1530
    const int lt= src[-1-1*stride];
1531
    LOAD_TOP_EDGE
1532
    LOAD_LEFT_EDGE
1533

    
1534
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1535
    src[0+2*stride]=
1536
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1537
    src[0+1*stride]=
1538
    src[1+2*stride]=
1539
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1540
    src[0+0*stride]=
1541
    src[1+1*stride]=
1542
    src[2+2*stride]=
1543
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1544
    src[1+0*stride]=
1545
    src[2+1*stride]=
1546
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1547
    src[2+0*stride]=
1548
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1549
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1550
}
1551

    
1552
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1553
    LOAD_TOP_EDGE    
1554
    LOAD_TOP_RIGHT_EDGE    
1555
//    LOAD_LEFT_EDGE    
1556

    
1557
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1558
    src[1+0*stride]=
1559
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1560
    src[2+0*stride]=
1561
    src[1+1*stride]=
1562
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1563
    src[3+0*stride]=
1564
    src[2+1*stride]=
1565
    src[1+2*stride]=
1566
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1567
    src[3+1*stride]=
1568
    src[2+2*stride]=
1569
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1570
    src[3+2*stride]=
1571
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1572
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1573
}
1574

    
1575
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1576
    const int lt= src[-1-1*stride];
1577
    LOAD_TOP_EDGE    
1578
    LOAD_LEFT_EDGE    
1579
    const __attribute__((unused)) int unu= l3;
1580

    
1581
    src[0+0*stride]=
1582
    src[1+2*stride]=(lt + t0 + 1)>>1;
1583
    src[1+0*stride]=
1584
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1585
    src[2+0*stride]=
1586
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1587
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1588
    src[0+1*stride]=
1589
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1590
    src[1+1*stride]=
1591
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1592
    src[2+1*stride]=
1593
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1594
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1595
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1596
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1597
}
1598

    
1599
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1600
    LOAD_TOP_EDGE    
1601
    LOAD_TOP_RIGHT_EDGE    
1602
    const __attribute__((unused)) int unu= t7;
1603

    
1604
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1605
    src[1+0*stride]=
1606
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1607
    src[2+0*stride]=
1608
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1609
    src[3+0*stride]=
1610
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1611
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1612
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1613
    src[1+1*stride]=
1614
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1615
    src[2+1*stride]=
1616
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1617
    src[3+1*stride]=
1618
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1619
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1620
}
1621

    
1622
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1623
    LOAD_LEFT_EDGE    
1624

    
1625
    src[0+0*stride]=(l0 + l1 + 1)>>1;
1626
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1627
    src[2+0*stride]=
1628
    src[0+1*stride]=(l1 + l2 + 1)>>1;
1629
    src[3+0*stride]=
1630
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1631
    src[2+1*stride]=
1632
    src[0+2*stride]=(l2 + l3 + 1)>>1;
1633
    src[3+1*stride]=
1634
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1635
    src[3+2*stride]=
1636
    src[1+3*stride]=
1637
    src[0+3*stride]=
1638
    src[2+2*stride]=
1639
    src[2+3*stride]=
1640
    src[3+3*stride]=l3;
1641
}
1642
    
1643
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1644
    const int lt= src[-1-1*stride];
1645
    LOAD_TOP_EDGE    
1646
    LOAD_LEFT_EDGE    
1647
    const __attribute__((unused)) int unu= t3;
1648

    
1649
    src[0+0*stride]=
1650
    src[2+1*stride]=(lt + l0 + 1)>>1;
1651
    src[1+0*stride]=
1652
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1653
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1654
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1655
    src[0+1*stride]=
1656
    src[2+2*stride]=(l0 + l1 + 1)>>1;
1657
    src[1+1*stride]=
1658
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1659
    src[0+2*stride]=
1660
    src[2+3*stride]=(l1 + l2+ 1)>>1;
1661
    src[1+2*stride]=
1662
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1663
    src[0+3*stride]=(l2 + l3 + 1)>>1;
1664
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1665
}
1666

    
1667
static void pred16x16_vertical_c(uint8_t *src, int stride){
1668
    int i;
1669
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1670
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1671
    const uint32_t c= ((uint32_t*)(src-stride))[2];
1672
    const uint32_t d= ((uint32_t*)(src-stride))[3];
1673
    
1674
    for(i=0; i<16; i++){
1675
        ((uint32_t*)(src+i*stride))[0]= a;
1676
        ((uint32_t*)(src+i*stride))[1]= b;
1677
        ((uint32_t*)(src+i*stride))[2]= c;
1678
        ((uint32_t*)(src+i*stride))[3]= d;
1679
    }
1680
}
1681

    
1682
static void pred16x16_horizontal_c(uint8_t *src, int stride){
1683
    int i;
1684

    
1685
    for(i=0; i<16; i++){
1686
        ((uint32_t*)(src+i*stride))[0]=
1687
        ((uint32_t*)(src+i*stride))[1]=
1688
        ((uint32_t*)(src+i*stride))[2]=
1689
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1690
    }
1691
}
1692

    
1693
static void pred16x16_dc_c(uint8_t *src, int stride){
1694
    int i, dc=0;
1695

    
1696
    for(i=0;i<16; i++){
1697
        dc+= src[-1+i*stride];
1698
    }
1699
    
1700
    for(i=0;i<16; i++){
1701
        dc+= src[i-stride];
1702
    }
1703

    
1704
    dc= 0x01010101*((dc + 16)>>5);
1705

    
1706
    for(i=0; i<16; i++){
1707
        ((uint32_t*)(src+i*stride))[0]=
1708
        ((uint32_t*)(src+i*stride))[1]=
1709
        ((uint32_t*)(src+i*stride))[2]=
1710
        ((uint32_t*)(src+i*stride))[3]= dc;
1711
    }
1712
}
1713

    
1714
static void pred16x16_left_dc_c(uint8_t *src, int stride){
1715
    int i, dc=0;
1716

    
1717
    for(i=0;i<16; i++){
1718
        dc+= src[-1+i*stride];
1719
    }
1720
    
1721
    dc= 0x01010101*((dc + 8)>>4);
1722

    
1723
    for(i=0; i<16; i++){
1724
        ((uint32_t*)(src+i*stride))[0]=
1725
        ((uint32_t*)(src+i*stride))[1]=
1726
        ((uint32_t*)(src+i*stride))[2]=
1727
        ((uint32_t*)(src+i*stride))[3]= dc;
1728
    }
1729
}
1730

    
1731
static void pred16x16_top_dc_c(uint8_t *src, int stride){
1732
    int i, dc=0;
1733

    
1734
    for(i=0;i<16; i++){
1735
        dc+= src[i-stride];
1736
    }
1737
    dc= 0x01010101*((dc + 8)>>4);
1738

    
1739
    for(i=0; i<16; i++){
1740
        ((uint32_t*)(src+i*stride))[0]=
1741
        ((uint32_t*)(src+i*stride))[1]=
1742
        ((uint32_t*)(src+i*stride))[2]=
1743
        ((uint32_t*)(src+i*stride))[3]= dc;
1744
    }
1745
}
1746

    
1747
static void pred16x16_128_dc_c(uint8_t *src, int stride){
1748
    int i;
1749

    
1750
    for(i=0; i<16; i++){
1751
        ((uint32_t*)(src+i*stride))[0]=
1752
        ((uint32_t*)(src+i*stride))[1]=
1753
        ((uint32_t*)(src+i*stride))[2]=
1754
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1755
    }
1756
}
1757

    
1758
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1759
  int i, j, k;
1760
  int a;
1761
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
1762
  const uint8_t * const src0 = src+7-stride;
1763
  const uint8_t *src1 = src+8*stride-1;
1764
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
1765
  int H = src0[1] - src0[-1];
1766
  int V = src1[0] - src2[ 0];
1767
  for(k=2; k<=8; ++k) {
1768
    src1 += stride; src2 -= stride;
1769
    H += k*(src0[k] - src0[-k]);
1770
    V += k*(src1[0] - src2[ 0]);
1771
  }
1772
  if(svq3){
1773
    H = ( 5*(H/4) ) / 16;
1774
    V = ( 5*(V/4) ) / 16;
1775

    
1776
    /* required for 100% accuracy */
1777
    i = H; H = V; V = i;
1778
  }else{
1779
    H = ( 5*H+32 ) >> 6;
1780
    V = ( 5*V+32 ) >> 6;
1781
  }
1782

    
1783
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1784
  for(j=16; j>0; --j) {
1785
    int b = a;
1786
    a += V;
1787
    for(i=-16; i<0; i+=4) {
1788
      src[16+i] = cm[ (b    ) >> 5 ];
1789
      src[17+i] = cm[ (b+  H) >> 5 ];
1790
      src[18+i] = cm[ (b+2*H) >> 5 ];
1791
      src[19+i] = cm[ (b+3*H) >> 5 ];
1792
      b += 4*H;
1793
    }
1794
    src += stride;
1795
  }
1796
}
1797

    
1798
static void pred16x16_plane_c(uint8_t *src, int stride){
1799
    pred16x16_plane_compat_c(src, stride, 0);
1800
}
1801

    
1802
static void pred8x8_vertical_c(uint8_t *src, int stride){
1803
    int i;
1804
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1805
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1806
    
1807
    for(i=0; i<8; i++){
1808
        ((uint32_t*)(src+i*stride))[0]= a;
1809
        ((uint32_t*)(src+i*stride))[1]= b;
1810
    }
1811
}
1812

    
1813
static void pred8x8_horizontal_c(uint8_t *src, int stride){
1814
    int i;
1815

    
1816
    for(i=0; i<8; i++){
1817
        ((uint32_t*)(src+i*stride))[0]=
1818
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1819
    }
1820
}
1821

    
1822
static void pred8x8_128_dc_c(uint8_t *src, int stride){
1823
    int i;
1824

    
1825
    for(i=0; i<4; i++){
1826
        ((uint32_t*)(src+i*stride))[0]= 
1827
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1828
    }
1829
    for(i=4; i<8; i++){
1830
        ((uint32_t*)(src+i*stride))[0]= 
1831
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1832
    }
1833
}
1834

    
1835
static void pred8x8_left_dc_c(uint8_t *src, int stride){
1836
    int i;
1837
    int dc0, dc2;
1838

    
1839
    dc0=dc2=0;
1840
    for(i=0;i<4; i++){
1841
        dc0+= src[-1+i*stride];
1842
        dc2+= src[-1+(i+4)*stride];
1843
    }
1844
    dc0= 0x01010101*((dc0 + 2)>>2);
1845
    dc2= 0x01010101*((dc2 + 2)>>2);
1846

    
1847
    for(i=0; i<4; i++){
1848
        ((uint32_t*)(src+i*stride))[0]=
1849
        ((uint32_t*)(src+i*stride))[1]= dc0;
1850
    }
1851
    for(i=4; i<8; i++){
1852
        ((uint32_t*)(src+i*stride))[0]=
1853
        ((uint32_t*)(src+i*stride))[1]= dc2;
1854
    }
1855
}
1856

    
1857
static void pred8x8_top_dc_c(uint8_t *src, int stride){
1858
    int i;
1859
    int dc0, dc1;
1860

    
1861
    dc0=dc1=0;
1862
    for(i=0;i<4; i++){
1863
        dc0+= src[i-stride];
1864
        dc1+= src[4+i-stride];
1865
    }
1866
    dc0= 0x01010101*((dc0 + 2)>>2);
1867
    dc1= 0x01010101*((dc1 + 2)>>2);
1868

    
1869
    for(i=0; i<4; i++){
1870
        ((uint32_t*)(src+i*stride))[0]= dc0;
1871
        ((uint32_t*)(src+i*stride))[1]= dc1;
1872
    }
1873
    for(i=4; i<8; i++){
1874
        ((uint32_t*)(src+i*stride))[0]= dc0;
1875
        ((uint32_t*)(src+i*stride))[1]= dc1;
1876
    }
1877
}
1878

    
1879

    
1880
static void pred8x8_dc_c(uint8_t *src, int stride){
1881
    int i;
1882
    int dc0, dc1, dc2, dc3;
1883

    
1884
    dc0=dc1=dc2=0;
1885
    for(i=0;i<4; i++){
1886
        dc0+= src[-1+i*stride] + src[i-stride];
1887
        dc1+= src[4+i-stride];
1888
        dc2+= src[-1+(i+4)*stride];
1889
    }
1890
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1891
    dc0= 0x01010101*((dc0 + 4)>>3);
1892
    dc1= 0x01010101*((dc1 + 2)>>2);
1893
    dc2= 0x01010101*((dc2 + 2)>>2);
1894

    
1895
    for(i=0; i<4; i++){
1896
        ((uint32_t*)(src+i*stride))[0]= dc0;
1897
        ((uint32_t*)(src+i*stride))[1]= dc1;
1898
    }
1899
    for(i=4; i<8; i++){
1900
        ((uint32_t*)(src+i*stride))[0]= dc2;
1901
        ((uint32_t*)(src+i*stride))[1]= dc3;
1902
    }
1903
}
1904

    
1905
static void pred8x8_plane_c(uint8_t *src, int stride){
1906
  int j, k;
1907
  int a;
1908
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
1909
  const uint8_t * const src0 = src+3-stride;
1910
  const uint8_t *src1 = src+4*stride-1;
1911
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
1912
  int H = src0[1] - src0[-1];
1913
  int V = src1[0] - src2[ 0];
1914
  for(k=2; k<=4; ++k) {
1915
    src1 += stride; src2 -= stride;
1916
    H += k*(src0[k] - src0[-k]);
1917
    V += k*(src1[0] - src2[ 0]);
1918
  }
1919
  H = ( 17*H+16 ) >> 5;
1920
  V = ( 17*V+16 ) >> 5;
1921

    
1922
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1923
  for(j=8; j>0; --j) {
1924
    int b = a;
1925
    a += V;
1926
    src[0] = cm[ (b    ) >> 5 ];
1927
    src[1] = cm[ (b+  H) >> 5 ];
1928
    src[2] = cm[ (b+2*H) >> 5 ];
1929
    src[3] = cm[ (b+3*H) >> 5 ];
1930
    src[4] = cm[ (b+4*H) >> 5 ];
1931
    src[5] = cm[ (b+5*H) >> 5 ];
1932
    src[6] = cm[ (b+6*H) >> 5 ];
1933
    src[7] = cm[ (b+7*H) >> 5 ];
1934
    src += stride;
1935
  }
1936
}
1937

    
1938
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1939
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1940
                           int src_x_offset, int src_y_offset,
1941
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1942
    MpegEncContext * const s = &h->s;
1943
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1944
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1945
    const int luma_xy= (mx&3) + ((my&3)<<2);
1946
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1947
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1948
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1949
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1950
    int extra_height= extra_width;
1951
    int emu=0;
1952
    const int full_mx= mx>>2;
1953
    const int full_my= my>>2;
1954
    
1955
    assert(pic->data[0]);
1956
    
1957
    if(mx&7) extra_width -= 3;
1958
    if(my&7) extra_height -= 3;
1959
    
1960
    if(   full_mx < 0-extra_width 
1961
       || full_my < 0-extra_height 
1962
       || full_mx + 16/*FIXME*/ > s->width + extra_width 
1963
       || full_my + 16/*FIXME*/ > s->height + extra_height){
1964
        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);
1965
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1966
        emu=1;
1967
    }
1968
    
1969
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1970
    if(!square){
1971
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1972
    }
1973
    
1974
    if(s->flags&CODEC_FLAG_GRAY) return;
1975
    
1976
    if(emu){
1977
        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);
1978
            src_cb= s->edge_emu_buffer;
1979
    }
1980
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1981

    
1982
    if(emu){
1983
        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);
1984
            src_cr= s->edge_emu_buffer;
1985
    }
1986
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1987
}
1988

    
1989
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1990
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1991
                           int x_offset, int y_offset,
1992
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1993
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1994
                           int list0, int list1){
1995
    MpegEncContext * const s = &h->s;
1996
    qpel_mc_func *qpix_op=  qpix_put;
1997
    h264_chroma_mc_func chroma_op= chroma_put;
1998
    
1999
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2000
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2001
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2002
    x_offset += 8*s->mb_x;
2003
    y_offset += 8*s->mb_y;
2004
    
2005
    if(list0){
2006
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2007
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2008
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2009
                           qpix_op, chroma_op);
2010

    
2011
        qpix_op=  qpix_avg;
2012
        chroma_op= chroma_avg;
2013
    }
2014

    
2015
    if(list1){
2016
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2017
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2018
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2019
                           qpix_op, chroma_op);
2020
    }
2021
}
2022

    
2023
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2024
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2025
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
2026
    MpegEncContext * const s = &h->s;
2027
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2028
    const int mb_type= s->current_picture.mb_type[mb_xy];
2029
    
2030
    assert(IS_INTER(mb_type));
2031
    
2032
    if(IS_16X16(mb_type)){
2033
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2034
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2035
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2036
    }else if(IS_16X8(mb_type)){
2037
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2038
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2039
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2040
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2041
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2042
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2043
    }else if(IS_8X16(mb_type)){
2044
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2045
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2046
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2047
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2048
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2049
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2050
    }else{
2051
        int i;
2052
        
2053
        assert(IS_8X8(mb_type));
2054

    
2055
        for(i=0; i<4; i++){
2056
            const int sub_mb_type= h->sub_mb_type[i];
2057
            const int n= 4*i;
2058
            int x_offset= (i&1)<<2;
2059
            int y_offset= (i&2)<<1;
2060

    
2061
            if(IS_SUB_8X8(sub_mb_type)){
2062
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2063
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2064
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2065
            }else if(IS_SUB_8X4(sub_mb_type)){
2066
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2067
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2068
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2069
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2070
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2071
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2072
            }else if(IS_SUB_4X8(sub_mb_type)){
2073
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2074
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2075
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2076
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2077
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2078
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2079
            }else{
2080
                int j;
2081
                assert(IS_SUB_4X4(sub_mb_type));
2082
                for(j=0; j<4; j++){
2083
                    int sub_x_offset= x_offset + 2*(j&1);
2084
                    int sub_y_offset= y_offset +   (j&2);
2085
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2086
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2087
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2088
                }
2089
            }
2090
        }
2091
    }
2092
}
2093

    
2094
static void decode_init_vlc(H264Context *h){
2095
    static int done = 0;
2096

    
2097
    if (!done) {
2098
        int i;
2099
        done = 1;
2100

    
2101
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2102
                 &chroma_dc_coeff_token_len [0], 1, 1,
2103
                 &chroma_dc_coeff_token_bits[0], 1, 1);
2104

    
2105
        for(i=0; i<4; i++){
2106
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2107
                     &coeff_token_len [i][0], 1, 1,
2108
                     &coeff_token_bits[i][0], 1, 1);
2109
        }
2110

    
2111
        for(i=0; i<3; i++){
2112
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2113
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2114
                     &chroma_dc_total_zeros_bits[i][0], 1, 1);
2115
        }
2116
        for(i=0; i<15; i++){
2117
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2118
                     &total_zeros_len [i][0], 1, 1,
2119
                     &total_zeros_bits[i][0], 1, 1);
2120
        }
2121

    
2122
        for(i=0; i<6; i++){
2123
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2124
                     &run_len [i][0], 1, 1,
2125
                     &run_bits[i][0], 1, 1);
2126
        }
2127
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2128
                 &run_len [6][0], 1, 1,
2129
                 &run_bits[6][0], 1, 1);
2130
    }
2131
}
2132

    
2133
/**
2134
 * Sets the intra prediction function pointers.
2135
 */
2136
static void init_pred_ptrs(H264Context *h){
2137
//    MpegEncContext * const s = &h->s;
2138

    
2139
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2140
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2141
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2142
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2143
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2144
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2145
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2146
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2147
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2148
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2149
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2150
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2151

    
2152
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2153
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2154
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2155
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2156
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2157
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2158
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2159

    
2160
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2161
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2162
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2163
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2164
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2165
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2166
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2167
}
2168

    
2169
static void free_tables(H264Context *h){
2170
    av_freep(&h->intra4x4_pred_mode);
2171
    av_freep(&h->chroma_pred_mode_table);
2172
    av_freep(&h->cbp_table);
2173
    av_freep(&h->mvd_table[0]);
2174
    av_freep(&h->mvd_table[1]);
2175
    av_freep(&h->non_zero_count);
2176
    av_freep(&h->slice_table_base);
2177
    av_freep(&h->top_border);
2178
    h->slice_table= NULL;
2179

    
2180
    av_freep(&h->mb2b_xy);
2181
    av_freep(&h->mb2b8_xy);
2182
}
2183

    
2184
/**
2185
 * allocates tables.
2186
 * needs widzh/height
2187
 */
2188
static int alloc_tables(H264Context *h){
2189
    MpegEncContext * const s = &h->s;
2190
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2191
    int x,y;
2192

    
2193
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2194

    
2195
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2196
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2197
    CHECKED_ALLOCZ(h->top_border       , s->mb_width * (16+8+8) * sizeof(uint8_t))
2198

    
2199
    if( h->pps.cabac ) {
2200
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2201
        CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2202
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2203
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2204
    }
2205

    
2206
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2207
    h->slice_table= h->slice_table_base + s->mb_stride + 1;
2208

    
2209
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint16_t));
2210
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2211
    for(y=0; y<s->mb_height; y++){
2212
        for(x=0; x<s->mb_width; x++){
2213
            const int mb_xy= x + y*s->mb_stride;
2214
            const int b_xy = 4*x + 4*y*h->b_stride;
2215
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2216
        
2217
            h->mb2b_xy [mb_xy]= b_xy;
2218
            h->mb2b8_xy[mb_xy]= b8_xy;
2219
        }
2220
    }
2221
    
2222
    return 0;
2223
fail:
2224
    free_tables(h);
2225
    return -1;
2226
}
2227

    
2228
static void common_init(H264Context *h){
2229
    MpegEncContext * const s = &h->s;
2230

    
2231
    s->width = s->avctx->width;
2232
    s->height = s->avctx->height;
2233
    s->codec_id= s->avctx->codec->id;
2234
    
2235
    init_pred_ptrs(h);
2236

    
2237
    s->unrestricted_mv=1;
2238
    s->decode=1; //FIXME
2239
}
2240

    
2241
static int decode_init(AVCodecContext *avctx){
2242
    H264Context *h= avctx->priv_data;
2243
    MpegEncContext * const s = &h->s;
2244

    
2245
    MPV_decode_defaults(s);
2246
    
2247
    s->avctx = avctx;
2248
    common_init(h);
2249

    
2250
    s->out_format = FMT_H264;
2251
    s->workaround_bugs= avctx->workaround_bugs;
2252

    
2253
    // set defaults
2254
//    s->decode_mb= ff_h263_decode_mb;
2255
    s->low_delay= 1;
2256
    avctx->pix_fmt= PIX_FMT_YUV420P;
2257

    
2258
    decode_init_vlc(h);
2259
    
2260
    return 0;
2261
}
2262

    
2263
static void frame_start(H264Context *h){
2264
    MpegEncContext * const s = &h->s;
2265
    int i;
2266

    
2267
    MPV_frame_start(s, s->avctx);
2268
    ff_er_frame_start(s);
2269
    h->mmco_index=0;
2270

    
2271
    assert(s->linesize && s->uvlinesize);
2272

    
2273
    for(i=0; i<16; i++){
2274
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2275
        h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2276
    }
2277
    for(i=0; i<4; i++){
2278
        h->block_offset[16+i]=
2279
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2280
    }
2281

    
2282
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2283
}
2284

    
2285
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2286
    MpegEncContext * const s = &h->s;
2287
    int i;
2288
    
2289
    src_y  -=   linesize;
2290
    src_cb -= uvlinesize;
2291
    src_cr -= uvlinesize;
2292

    
2293
    h->left_border[0]= h->top_border[s->mb_x][15];
2294
    for(i=1; i<17; i++){
2295
        h->left_border[i]= src_y[15+i*  linesize];
2296
    }
2297
    
2298
    *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2299
    *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2300

    
2301
    if(!(s->flags&CODEC_FLAG_GRAY)){
2302
        h->left_border[17  ]= h->top_border[s->mb_x][16+7];
2303
        h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2304
        for(i=1; i<9; i++){
2305
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
2306
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2307
        }
2308
        *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2309
        *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2310
    }
2311
}
2312

    
2313
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){
2314
    MpegEncContext * const s = &h->s;
2315
    int temp8, i;
2316
    uint64_t temp64;
2317

    
2318
    src_y  -=   linesize + 1;
2319
    src_cb -= uvlinesize + 1;
2320
    src_cr -= uvlinesize + 1;
2321

    
2322
#define XCHG(a,b,t,xchg)\
2323
t= a;\
2324
if(xchg)\
2325
    a= b;\
2326
b= t;
2327
    
2328
    for(i=0; i<17; i++){
2329
        XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2330
    }
2331
    
2332
    XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2333
    XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2334

    
2335
    if(!(s->flags&CODEC_FLAG_GRAY)){
2336
        for(i=0; i<9; i++){
2337
            XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
2338
            XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2339
        }
2340
        XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2341
        XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2342
    }
2343
}
2344

    
2345
static void hl_decode_mb(H264Context *h){
2346
    MpegEncContext * const s = &h->s;
2347
    const int mb_x= s->mb_x;
2348
    const int mb_y= s->mb_y;
2349
    const int mb_xy= mb_x + mb_y*s->mb_stride;
2350
    const int mb_type= s->current_picture.mb_type[mb_xy];
2351
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2352
    int linesize, uvlinesize /*dct_offset*/;
2353
    int i;
2354

    
2355
    if(!s->decode)
2356
        return;
2357

    
2358
    if(s->mb_skiped){
2359
    }
2360

    
2361
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2362
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2363
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2364

    
2365
    if (h->mb_field_decoding_flag) {
2366
        linesize = s->linesize * 2;
2367
        uvlinesize = s->uvlinesize * 2;
2368
        if(mb_y&1){ //FIXME move out of this func?
2369
            dest_y -= s->linesize*15;
2370
            dest_cb-= s->linesize*7;
2371
            dest_cr-= s->linesize*7;
2372
        }
2373
    } else {
2374
        linesize = s->linesize;
2375
        uvlinesize = s->uvlinesize;
2376
//        dct_offset = s->linesize * 16;
2377
    }
2378

    
2379
    if(IS_INTRA(mb_type)){
2380
        if(h->deblocking_filter)
2381
            xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2382

    
2383
        if(!(s->flags&CODEC_FLAG_GRAY)){
2384
            h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2385
            h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2386
        }
2387

    
2388
        if(IS_INTRA4x4(mb_type)){
2389
            if(!s->encoding){
2390
                for(i=0; i<16; i++){
2391
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2392
                    uint8_t *topright= ptr + 4 - linesize;
2393
                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2394
                    const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2395
                    int tr;
2396

    
2397
                    if(!topright_avail){
2398
                        tr= ptr[3 - linesize]*0x01010101;
2399
                        topright= (uint8_t*) &tr;
2400
                    }else if(i==5 && h->deblocking_filter){
2401
                        tr= *(uint32_t*)h->top_border[mb_x+1];
2402
                        topright= (uint8_t*) &tr;
2403
                    }
2404

    
2405
                    h->pred4x4[ dir ](ptr, topright, linesize);
2406
                    if(h->non_zero_count_cache[ scan8[i] ]){
2407
                        if(s->codec_id == CODEC_ID_H264)
2408
                            h264_add_idct_c(ptr, h->mb + i*16, linesize);
2409
                        else
2410
                            svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2411
                    }
2412
                }
2413
            }
2414
        }else{
2415
            h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2416
            if(s->codec_id == CODEC_ID_H264)
2417
                h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2418
            else
2419
                svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2420
        }
2421
        if(h->deblocking_filter)
2422
            xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2423
    }else if(s->codec_id == CODEC_ID_H264){
2424
        hl_motion(h, dest_y, dest_cb, dest_cr,
2425
                  s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
2426
                  s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2427
    }
2428

    
2429

    
2430
    if(!IS_INTRA4x4(mb_type)){
2431
        if(s->codec_id == CODEC_ID_H264){
2432
            for(i=0; i<16; i++){
2433
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2434
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2435
                    h264_add_idct_c(ptr, h->mb + i*16, linesize);
2436
                }
2437
            }
2438
        }else{
2439
            for(i=0; i<16; i++){
2440
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2441
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2442
                    svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2443
                }
2444
            }
2445
        }
2446
    }
2447

    
2448
    if(!(s->flags&CODEC_FLAG_GRAY)){
2449
        chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2450
        chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2451
        if(s->codec_id == CODEC_ID_H264){
2452
            for(i=16; i<16+4; i++){
2453
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2454
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2455
                    h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2456
                }
2457
            }
2458
            for(i=20; i<20+4; i++){
2459
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2460
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2461
                    h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2462
                }
2463
            }
2464
        }else{
2465
            for(i=16; i<16+4; i++){
2466
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2467
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2468
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2469
                }
2470
            }
2471
            for(i=20; i<20+4; i++){
2472
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2473
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2474
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2475
                }
2476
            }
2477
        }
2478
    }
2479
    if(h->deblocking_filter) {
2480
        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2481
        filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2482
    }
2483
}
2484

    
2485
/**
2486
 * fills the default_ref_list.
2487
 */
2488
static int fill_default_ref_list(H264Context *h){
2489
    MpegEncContext * const s = &h->s;
2490
    int i;
2491
    Picture sorted_short_ref[16];
2492
    
2493
    if(h->slice_type==B_TYPE){
2494
        int out_i;
2495
        int limit= -1;
2496

    
2497
        for(out_i=0; out_i<h->short_ref_count; out_i++){
2498
            int best_i=-1;
2499
            int best_poc=-1;
2500

    
2501
            for(i=0; i<h->short_ref_count; i++){
2502
                const int poc= h->short_ref[i]->poc;
2503
                if(poc > limit && poc < best_poc){
2504
                    best_poc= poc;
2505
                    best_i= i;
2506
                }
2507
            }
2508
            
2509
            assert(best_i != -1);
2510
            
2511
            limit= best_poc;
2512
            sorted_short_ref[out_i]= *h->short_ref[best_i];
2513
        }
2514
    }
2515

    
2516
    if(s->picture_structure == PICT_FRAME){
2517
        if(h->slice_type==B_TYPE){
2518
            const int current_poc= s->current_picture_ptr->poc;
2519
            int list;
2520

    
2521
            for(list=0; list<2; list++){
2522
                int index=0;
2523

    
2524
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2525
                    const int i2= list ? h->short_ref_count - i - 1 : i;
2526
                    const int poc= sorted_short_ref[i2].poc;
2527
                    
2528
                    if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2529

    
2530
                    if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2531
                        h->default_ref_list[list][index  ]= sorted_short_ref[i2];
2532
                        h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2533
                    }
2534
                }
2535

    
2536
                for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2537
                    if(h->long_ref[i]->reference != 3) continue;
2538

    
2539
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
2540
                    h->default_ref_list[ list ][index++].pic_id= i;;
2541
                }
2542
                
2543
                if(h->long_ref_count > 1 && h->short_ref_count==0){
2544
                    Picture temp= h->default_ref_list[1][0];
2545
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
2546
                    h->default_ref_list[1][0] = temp;
2547
                }
2548

    
2549
                if(index < h->ref_count[ list ])
2550
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2551
            }
2552
        }else{
2553
            int index=0;
2554
            for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2555
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2556
                h->default_ref_list[0][index  ]= *h->short_ref[i];
2557
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2558
            }
2559
            for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2560
                if(h->long_ref[i]->reference != 3) continue;
2561
                h->default_ref_list[0][index  ]= *h->long_ref[i];
2562
                h->default_ref_list[0][index++].pic_id= i;;
2563
            }
2564
            if(index < h->ref_count[0])
2565
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2566
        }
2567
    }else{ //FIELD
2568
        if(h->slice_type==B_TYPE){
2569
        }else{
2570
            //FIXME second field balh
2571
        }
2572
    }
2573
    return 0;
2574
}
2575

    
2576
static int decode_ref_pic_list_reordering(H264Context *h){
2577
    MpegEncContext * const s = &h->s;
2578
    int list;
2579
    
2580
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2581
    
2582
    for(list=0; list<2; list++){
2583
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2584

    
2585
        if(get_bits1(&s->gb)){
2586
            int pred= h->curr_pic_num;
2587
            int index;
2588

    
2589
            for(index=0; ; index++){
2590
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2591
                int pic_id;
2592
                int i;
2593
                
2594
                
2595
                if(index >= h->ref_count[list]){
2596
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2597
                    return -1;
2598
                }
2599
                
2600
                if(reordering_of_pic_nums_idc<3){
2601
                    if(reordering_of_pic_nums_idc<2){
2602
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2603

    
2604
                        if(abs_diff_pic_num >= h->max_pic_num){
2605
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2606
                            return -1;
2607
                        }
2608

    
2609
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2610
                        else                                pred+= abs_diff_pic_num;
2611
                        pred &= h->max_pic_num - 1;
2612
                    
2613
                        for(i= h->ref_count[list]-1; i>=index; i--){
2614
                            if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2615
                                break;
2616
                        }
2617
                    }else{
2618
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2619

    
2620
                        for(i= h->ref_count[list]-1; i>=index; i--){
2621
                            if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2622
                                break;
2623
                        }
2624
                    }
2625

    
2626
                    if(i < index){
2627
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2628
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2629
                    }else if(i > index){
2630
                        Picture tmp= h->ref_list[list][i];
2631
                        for(; i>index; i--){
2632
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2633
                        }
2634
                        h->ref_list[list][index]= tmp;
2635
                    }
2636
                }else if(reordering_of_pic_nums_idc==3) 
2637
                    break;
2638
                else{
2639
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2640
                    return -1;
2641
                }
2642
            }
2643
        }
2644

    
2645
        if(h->slice_type!=B_TYPE) break;
2646
    }
2647
    return 0;    
2648
}
2649

    
2650
static int pred_weight_table(H264Context *h){
2651
    MpegEncContext * const s = &h->s;
2652
    int list, i;
2653
    
2654
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2655
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2656

    
2657
    for(list=0; list<2; list++){
2658
        for(i=0; i<h->ref_count[list]; i++){
2659
            int luma_weight_flag, chroma_weight_flag;
2660
            
2661
            luma_weight_flag= get_bits1(&s->gb);
2662
            if(luma_weight_flag){
2663
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
2664
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
2665
            }
2666

    
2667
            chroma_weight_flag= get_bits1(&s->gb);
2668
            if(chroma_weight_flag){
2669
                int j;
2670
                for(j=0; j<2; j++){
2671
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2672
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2673
                }
2674
            }
2675
        }
2676
        if(h->slice_type != B_TYPE) break;
2677
    }
2678
    return 0;
2679
}
2680

    
2681
/**
2682
 * instantaneos decoder refresh.
2683
 */
2684
static void idr(H264Context *h){
2685
    int i;
2686

    
2687
    for(i=0; i<h->long_ref_count; i++){
2688
        h->long_ref[i]->reference=0;
2689
        h->long_ref[i]= NULL;
2690
    }
2691
    h->long_ref_count=0;
2692

    
2693
    for(i=0; i<h->short_ref_count; i++){
2694
        h->short_ref[i]->reference=0;
2695
        h->short_ref[i]= NULL;
2696
    }
2697
    h->short_ref_count=0;
2698
}
2699

    
2700
/**
2701
 *
2702
 * @return the removed picture or NULL if an error occures
2703
 */
2704
static Picture * remove_short(H264Context *h, int frame_num){
2705
    MpegEncContext * const s = &h->s;
2706
    int i;
2707
    
2708
    if(s->avctx->debug&FF_DEBUG_MMCO)
2709
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
2710
    
2711
    for(i=0; i<h->short_ref_count; i++){
2712
        Picture *pic= h->short_ref[i];
2713
        if(s->avctx->debug&FF_DEBUG_MMCO)
2714
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
2715
        if(pic->frame_num == frame_num){
2716
            h->short_ref[i]= NULL;
2717
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2718
            h->short_ref_count--;
2719
            return pic;
2720
        }
2721
    }
2722
    return NULL;
2723
}
2724

    
2725
/**
2726
 *
2727
 * @return the removed picture or NULL if an error occures
2728
 */
2729
static Picture * remove_long(H264Context *h, int i){
2730
    Picture *pic;
2731

    
2732
    if(i >= h->long_ref_count) return NULL;
2733
    pic= h->long_ref[i];
2734
    if(pic==NULL) return NULL;
2735
    
2736
    h->long_ref[i]= NULL;
2737
    memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2738
    h->long_ref_count--;
2739

    
2740
    return pic;
2741
}
2742

    
2743
/**
2744
 * Executes the reference picture marking (memory management control operations).
2745
 */
2746
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2747
    MpegEncContext * const s = &h->s;
2748
    int i;
2749
    int current_is_long=0;
2750
    Picture *pic;
2751
    
2752
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2753
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
2754
        
2755
    for(i=0; i<mmco_count; i++){
2756
        if(s->avctx->debug&FF_DEBUG_MMCO)
2757
            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);
2758

    
2759
        switch(mmco[i].opcode){
2760
        case MMCO_SHORT2UNUSED:
2761
            pic= remove_short(h, mmco[i].short_frame_num);
2762
            if(pic==NULL) return -1;
2763
            pic->reference= 0;
2764
            break;
2765
        case MMCO_SHORT2LONG:
2766
            pic= remove_long(h, mmco[i].long_index);
2767
            if(pic) pic->reference=0;
2768
            
2769
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2770
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
2771
            break;
2772
        case MMCO_LONG2UNUSED:
2773
            pic= remove_long(h, mmco[i].long_index);
2774
            if(pic==NULL) return -1;
2775
            pic->reference= 0;
2776
            break;
2777
        case MMCO_LONG:
2778
            pic= remove_long(h, mmco[i].long_index);
2779
            if(pic) pic->reference=0;
2780
            
2781
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2782
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
2783
            h->long_ref_count++;
2784
            
2785
            current_is_long=1;
2786
            break;
2787
        case MMCO_SET_MAX_LONG:
2788
            assert(mmco[i].long_index <= 16);
2789
            while(mmco[i].long_index < h->long_ref_count){
2790
                pic= remove_long(h, mmco[i].long_index);
2791
                pic->reference=0;
2792
            }
2793
            while(mmco[i].long_index > h->long_ref_count){
2794
                h->long_ref[ h->long_ref_count++ ]= NULL;
2795
            }
2796
            break;
2797
        case MMCO_RESET:
2798
            while(h->short_ref_count){
2799
                pic= remove_short(h, h->short_ref[0]->frame_num);
2800
                pic->reference=0;
2801
            }
2802
            while(h->long_ref_count){
2803
                pic= remove_long(h, h->long_ref_count-1);
2804
                pic->reference=0;
2805
            }
2806
            break;
2807
        default: assert(0);
2808
        }
2809
    }
2810
    
2811
    if(!current_is_long){
2812
        pic= remove_short(h, s->current_picture_ptr->frame_num);
2813
        if(pic){
2814
            pic->reference=0;
2815
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
2816
        }
2817
        
2818
        if(h->short_ref_count)
2819
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2820

    
2821
        h->short_ref[0]= s->current_picture_ptr;
2822
        h->short_ref[0]->long_ref=0;
2823
        h->short_ref_count++;
2824
    }
2825
    
2826
    return 0; 
2827
}
2828

    
2829
static int decode_ref_pic_marking(H264Context *h){
2830
    MpegEncContext * const s = &h->s;
2831
    int i;
2832
    
2833
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2834
        s->broken_link= get_bits1(&s->gb) -1;
2835
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2836
        if(h->mmco[0].long_index == -1)
2837
            h->mmco_index= 0;
2838
        else{
2839
            h->mmco[0].opcode= MMCO_LONG;
2840
            h->mmco_index= 1;
2841
        } 
2842
    }else{
2843
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2844
            for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) { 
2845
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2846

    
2847
                h->mmco[i].opcode= opcode;
2848
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2849
                    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
2850
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2851
                        fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2852
                        return -1;
2853
                    }*/
2854
                }
2855
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2856
                    h->mmco[i].long_index= get_ue_golomb(&s->gb);
2857
                    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){
2858
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
2859
                        return -1;
2860
                    }
2861
                }
2862
                    
2863
                if(opcode > MMCO_LONG){
2864
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
2865
                    return -1;
2866
                }
2867
            }
2868
            h->mmco_index= i;
2869
        }else{
2870
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2871

    
2872
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2873
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2874
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2875
                h->mmco_index= 1;
2876
            }else
2877
                h->mmco_index= 0;
2878
        }
2879
    }
2880
    
2881
    return 0; 
2882
}
2883

    
2884
static int init_poc(H264Context *h){
2885
    MpegEncContext * const s = &h->s;
2886
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2887
    int field_poc[2];
2888

    
2889
    if(h->nal_unit_type == NAL_IDR_SLICE){
2890
        h->frame_num_offset= 0;
2891
    }else{
2892
        if(h->frame_num < h->prev_frame_num)
2893
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2894
        else
2895
            h->frame_num_offset= h->prev_frame_num_offset;
2896
    }
2897

    
2898
    if(h->sps.poc_type==0){
2899
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2900

    
2901
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2902
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2903
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2904
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2905
        else
2906
            h->poc_msb = h->prev_poc_msb;
2907
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2908
        field_poc[0] = 
2909
        field_poc[1] = h->poc_msb + h->poc_lsb;
2910
        if(s->picture_structure == PICT_FRAME) 
2911
            field_poc[1] += h->delta_poc_bottom;
2912
    }else if(h->sps.poc_type==1){
2913
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2914
        int i;
2915

    
2916
        if(h->sps.poc_cycle_length != 0)
2917
            abs_frame_num = h->frame_num_offset + h->frame_num;
2918
        else
2919
            abs_frame_num = 0;
2920

    
2921
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
2922
            abs_frame_num--;
2923
            
2924
        expected_delta_per_poc_cycle = 0;
2925
        for(i=0; i < h->sps.poc_cycle_length; i++)
2926
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2927

    
2928
        if(abs_frame_num > 0){
2929
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2930
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2931

    
2932
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2933
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
2934
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2935
        } else
2936
            expectedpoc = 0;
2937

    
2938
        if(h->nal_ref_idc == 0) 
2939
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2940
        
2941
        field_poc[0] = expectedpoc + h->delta_poc[0];
2942
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2943

    
2944
        if(s->picture_structure == PICT_FRAME)
2945
            field_poc[1] += h->delta_poc[1];
2946
    }else{
2947
        int poc;
2948
        if(h->nal_unit_type == NAL_IDR_SLICE){
2949
            poc= 0;
2950
        }else{
2951
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2952
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2953
        }
2954
        field_poc[0]= poc;
2955
        field_poc[1]= poc;
2956
    }
2957
    
2958
    if(s->picture_structure != PICT_BOTTOM_FIELD)
2959
        s->current_picture_ptr->field_poc[0]= field_poc[0];
2960
    if(s->picture_structure != PICT_TOP_FIELD)
2961
        s->current_picture_ptr->field_poc[1]= field_poc[1];
2962
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2963
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2964

    
2965
    return 0;
2966
}
2967

    
2968
/**
2969
 * decodes a slice header.
2970
 * this will allso call MPV_common_init() and frame_start() as needed
2971
 */
2972
static int decode_slice_header(H264Context *h){
2973
    MpegEncContext * const s = &h->s;
2974
    int first_mb_in_slice, pps_id;
2975
    int num_ref_idx_active_override_flag;
2976
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2977

    
2978
    s->current_picture.reference= h->nal_ref_idc != 0;
2979

    
2980
    first_mb_in_slice= get_ue_golomb(&s->gb);
2981

    
2982
    h->slice_type= get_ue_golomb(&s->gb);
2983
    if(h->slice_type > 9){
2984
        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);
2985
    }
2986
    if(h->slice_type > 4){
2987
        h->slice_type -= 5;
2988
        h->slice_type_fixed=1;
2989
    }else
2990
        h->slice_type_fixed=0;
2991
    
2992
    h->slice_type= slice_type_map[ h->slice_type ];
2993
    
2994
    s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
2995
        
2996
    pps_id= get_ue_golomb(&s->gb);
2997
    if(pps_id>255){
2998
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2999
        return -1;
3000
    }
3001
    h->pps= h->pps_buffer[pps_id];
3002
    if(h->pps.slice_group_count == 0){
3003
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3004
        return -1;
3005
    }
3006

    
3007
    h->sps= h->sps_buffer[ h->pps.sps_id ];
3008
    if(h->sps.log2_max_frame_num == 0){
3009
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3010
        return -1;
3011
    }
3012
    
3013
    s->mb_width= h->sps.mb_width;
3014
    s->mb_height= h->sps.mb_height;
3015
    
3016
    h->b_stride=  s->mb_width*4;
3017
    h->b8_stride= s->mb_width*2;
3018

    
3019
    s->mb_x = first_mb_in_slice % s->mb_width;
3020
    s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
3021
    
3022
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3023
    if(h->sps.frame_mbs_only_flag)
3024
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
3025
    else
3026
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
3027
    
3028
    if (s->context_initialized 
3029
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3030
        free_tables(h);
3031
        MPV_common_end(s);
3032
    }
3033
    if (!s->context_initialized) {
3034
        if (MPV_common_init(s) < 0)
3035
            return -1;
3036

    
3037
        alloc_tables(h);
3038

    
3039
        s->avctx->width = s->width;
3040
        s->avctx->height = s->height;
3041
        s->avctx->sample_aspect_ratio= h->sps.sar;
3042
    }
3043

    
3044
    if(first_mb_in_slice == 0){
3045
        frame_start(h);
3046
    }
3047

    
3048
    s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3049
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3050

    
3051
    if(h->sps.frame_mbs_only_flag){
3052
        s->picture_structure= PICT_FRAME;
3053
    }else{
3054
        if(get_bits1(&s->gb)) //field_pic_flag
3055
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3056
        else
3057
            s->picture_structure= PICT_FRAME;
3058
    }
3059

    
3060
    if(s->picture_structure==PICT_FRAME){
3061
        h->curr_pic_num=   h->frame_num;
3062
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3063
    }else{
3064
        h->curr_pic_num= 2*h->frame_num;
3065
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3066
    }
3067
        
3068
    if(h->nal_unit_type == NAL_IDR_SLICE){
3069
        get_ue_golomb(&s->gb); /* idr_pic_id */
3070
    }
3071
   
3072
    if(h->sps.poc_type==0){
3073
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3074
        
3075
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3076
            h->delta_poc_bottom= get_se_golomb(&s->gb);
3077
        }
3078
    }
3079
    
3080
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3081
        h->delta_poc[0]= get_se_golomb(&s->gb);
3082
        
3083
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3084
            h->delta_poc[1]= get_se_golomb(&s->gb);
3085
    }
3086
    
3087
    init_poc(h);
3088
    
3089
    if(h->pps.redundant_pic_cnt_present){
3090
        h->redundant_pic_count= get_ue_golomb(&s->gb);
3091
    }
3092

    
3093
    //set defaults, might be overriden a few line later
3094
    h->ref_count[0]= h->pps.ref_count[0];
3095
    h->ref_count[1]= h->pps.ref_count[1];
3096

    
3097
    if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3098
        if(h->slice_type == B_TYPE){
3099
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
3100
        }
3101
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
3102
    
3103
        if(num_ref_idx_active_override_flag){
3104
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3105
            if(h->slice_type==B_TYPE)
3106
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3107

    
3108
            if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3109
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3110
                return -1;
3111
            }
3112
        }
3113
    }
3114

    
3115
    if(first_mb_in_slice == 0){
3116
        fill_default_ref_list(h);
3117
    }
3118

    
3119
    decode_ref_pic_list_reordering(h);
3120

    
3121
    if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) 
3122
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3123
        pred_weight_table(h);
3124
    
3125
    if(s->current_picture.reference)
3126
        decode_ref_pic_marking(h);
3127

    
3128
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3129
        h->cabac_init_idc = get_ue_golomb(&s->gb);
3130

    
3131
    h->last_qscale_diff = 0;
3132
    s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3133
    if(s->qscale<0 || s->qscale>51){
3134
        av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3135
        return -1;
3136
    }
3137
    //FIXME qscale / qp ... stuff
3138
    if(h->slice_type == SP_TYPE){
3139
        get_bits1(&s->gb); /* sp_for_switch_flag */
3140
    }
3141
    if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3142
        get_se_golomb(&s->gb); /* slice_qs_delta */
3143
    }
3144

    
3145
    h->deblocking_filter = 1;
3146
    h->slice_alpha_c0_offset = 0;
3147
    h->slice_beta_offset = 0;
3148
    if( h->pps.deblocking_filter_parameters_present ) {
3149
        h->deblocking_filter= get_ue_golomb(&s->gb);
3150
        if(h->deblocking_filter < 2) 
3151
            h->deblocking_filter^= 1; // 1<->0
3152

    
3153
        if( h->deblocking_filter ) {
3154
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3155
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3156
        }
3157
    }
3158

    
3159
#if 0 //FMO
3160
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3161
        slice_group_change_cycle= get_bits(&s->gb, ?);
3162
#endif
3163

    
3164
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3165
        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", 
3166
               first_mb_in_slice, 
3167
               av_get_pict_type_char(h->slice_type),
3168
               pps_id, h->frame_num,
3169
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3170
               h->ref_count[0], h->ref_count[1],
3171
               s->qscale,
3172
               h->deblocking_filter
3173
               );
3174
    }
3175

    
3176
    return 0;
3177
}
3178

    
3179
/**
3180
 *
3181
 */
3182
static inline int get_level_prefix(GetBitContext *gb){
3183
    unsigned int buf;
3184
    int log;
3185
    
3186
    OPEN_READER(re, gb);
3187
    UPDATE_CACHE(re, gb);
3188
    buf=GET_CACHE(re, gb);
3189
    
3190
    log= 32 - av_log2(buf);
3191
#ifdef TRACE
3192
    print_bin(buf>>(32-log), log);
3193
    printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3194
#endif
3195

    
3196
    LAST_SKIP_BITS(re, gb, log);
3197
    CLOSE_READER(re, gb);
3198

    
3199
    return log-1;
3200
}
3201

    
3202
/**
3203
 * decodes a residual block.
3204
 * @param n block index
3205
 * @param scantable scantable
3206
 * @param max_coeff number of coefficients in the block
3207
 * @return <0 if an error occured
3208
 */
3209
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3210
    MpegEncContext * const s = &h->s;
3211
    const uint16_t *qmul= dequant_coeff[qp];
3212
    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};
3213
    int level[16], run[16];
3214
    int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3215

    
3216
    //FIXME put trailing_onex into the context
3217

    
3218
    if(n == CHROMA_DC_BLOCK_INDEX){
3219
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3220
        total_coeff= coeff_token>>2;
3221
    }else{    
3222
        if(n == LUMA_DC_BLOCK_INDEX){
3223
            total_coeff= pred_non_zero_count(h, 0);
3224
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3225
            total_coeff= coeff_token>>2;
3226
        }else{
3227
            total_coeff= pred_non_zero_count(h, n);
3228
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3229
            total_coeff= coeff_token>>2;
3230
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3231
        }
3232
    }
3233

    
3234
    //FIXME set last_non_zero?
3235

    
3236
    if(total_coeff==0)
3237
        return 0;
3238
        
3239
    trailing_ones= coeff_token&3;
3240
    tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3241
    assert(total_coeff<=16);
3242
    
3243
    for(i=0; i<trailing_ones; i++){
3244
        level[i]= 1 - 2*get_bits1(gb);
3245
    }
3246

    
3247
    suffix_length= total_coeff > 10 && trailing_ones < 3;
3248

    
3249
    for(; i<total_coeff; i++){
3250
        const int prefix= get_level_prefix(gb);
3251
        int level_code, mask;
3252

    
3253
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3254
            if(suffix_length)
3255
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3256
            else
3257
                level_code= (prefix<<suffix_length); //part
3258
        }else if(prefix==14){
3259
            if(suffix_length)
3260
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3261
            else
3262
                level_code= prefix + get_bits(gb, 4); //part
3263
        }else if(prefix==15){
3264
            level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3265
            if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3266
        }else{
3267
            av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3268
            return -1;
3269
        }
3270

    
3271
        if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3272

    
3273
        mask= -(level_code&1);
3274
        level[i]= (((2+level_code)>>1) ^ mask) - mask;
3275

    
3276
        if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3277

    
3278
#if 1
3279
        if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3280
#else        
3281
        if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3282
        /* ? == prefix > 2 or sth */
3283
#endif
3284
        tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3285
    }
3286

    
3287
    if(total_coeff == max_coeff)
3288
        zeros_left=0;
3289
    else{
3290
        if(n == CHROMA_DC_BLOCK_INDEX)
3291
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3292
        else
3293
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3294
    }
3295
    
3296
    for(i=0; i<total_coeff-1; i++){
3297
        if(zeros_left <=0)
3298
            break;
3299
        else if(zeros_left < 7){
3300
            run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3301
        }else{
3302
            run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3303
        }
3304
        zeros_left -= run[i];
3305
    }
3306

    
3307
    if(zeros_left<0){
3308
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3309
        return -1;
3310
    }
3311
    
3312
    for(; i<total_coeff-1; i++){
3313
        run[i]= 0;
3314
    }
3315

    
3316
    run[i]= zeros_left;
3317

    
3318
    coeff_num=-1;
3319
    if(n > 24){
3320
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3321
            int j;
3322

    
3323
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3324
            j= scantable[ coeff_num ];
3325

    
3326
            block[j]= level[i];
3327
        }
3328
    }else{
3329
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into  rundecode?
3330
            int j;
3331

    
3332
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3333
            j= scantable[ coeff_num ];
3334

    
3335
            block[j]= level[i] * qmul[j];
3336
//            printf("%d %d  ", block[j], qmul[j]);
3337
        }
3338
    }
3339
    return 0;
3340
}
3341

    
3342
/**
3343
 * decodes a macroblock
3344
 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3345
 */
3346
static int decode_mb_cavlc(H264Context *h){
3347
    MpegEncContext * const s = &h->s;
3348
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3349
    int mb_type, partition_count, cbp;
3350

    
3351
    s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?    
3352

    
3353
    tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3354
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3355
                down the code */
3356
    if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3357
        if(s->mb_skip_run==-1)
3358
            s->mb_skip_run= get_ue_golomb(&s->gb);
3359
        
3360
        if (s->mb_skip_run--) {
3361
            int mx, my;
3362
            /* skip mb */
3363
//FIXME b frame
3364
            mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3365

    
3366
            memset(h->non_zero_count[mb_xy], 0, 16);
3367
            memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3368

    
3369
            if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3370
                h->mb_field_decoding_flag= get_bits1(&s->gb);
3371
            }
3372

    
3373
            if(h->mb_field_decoding_flag)
3374
                mb_type|= MB_TYPE_INTERLACED;
3375
            
3376
            fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3377
            pred_pskip_motion(h, &mx, &my);
3378
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3379
            fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3380
            write_back_motion(h, mb_type);
3381

    
3382
            s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3383
            s->current_picture.qscale_table[mb_xy]= s->qscale;
3384
            h->slice_table[ mb_xy ]= h->slice_num;
3385

    
3386
            h->prev_mb_skiped= 1;
3387
            return 0;
3388
        }
3389
    }
3390
    if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3391
        if((s->mb_y&1)==0)
3392
            h->mb_field_decoding_flag = get_bits1(&s->gb);
3393
    }else
3394
        h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3395
    
3396
    h->prev_mb_skiped= 0;
3397
    
3398
    mb_type= get_ue_golomb(&s->gb);
3399
    if(h->slice_type == B_TYPE){
3400
        if(mb_type < 23){
3401
            partition_count= b_mb_type_info[mb_type].partition_count;
3402
            mb_type=         b_mb_type_info[mb_type].type;
3403
        }else{
3404
            mb_type -= 23;
3405
            goto decode_intra_mb;
3406
        }
3407
    }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3408
        if(mb_type < 5){
3409
            partition_count= p_mb_type_info[mb_type].partition_count;
3410
            mb_type=         p_mb_type_info[mb_type].type;
3411
        }else{
3412
            mb_type -= 5;
3413
            goto decode_intra_mb;
3414
        }
3415
    }else{
3416
       assert(h->slice_type == I_TYPE);
3417
decode_intra_mb:
3418
        if(mb_type > 25){
3419
            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);
3420
            return -1;
3421
        }
3422
        partition_count=0;
3423
        cbp= i_mb_type_info[mb_type].cbp;
3424
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3425
        mb_type= i_mb_type_info[mb_type].type;
3426
    }
3427

    
3428
    if(h->mb_field_decoding_flag)
3429
        mb_type |= MB_TYPE_INTERLACED;
3430

    
3431
    s->current_picture.mb_type[mb_xy]= mb_type;
3432
    h->slice_table[ mb_xy ]= h->slice_num;
3433
    
3434
    if(IS_INTRA_PCM(mb_type)){
3435
        const uint8_t *ptr;
3436
        int x, y;
3437
        
3438
        // we assume these blocks are very rare so we dont optimize it
3439
        align_get_bits(&s->gb);
3440
        
3441
        ptr= s->gb.buffer + get_bits_count(&s->gb);
3442
    
3443
        for(y=0; y<16; y++){
3444
            const int index= 4*(y&3) + 64*(y>>2);
3445
            for(x=0; x<16; x++){
3446
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3447
            }
3448
        }
3449
        for(y=0; y<8; y++){
3450
            const int index= 256 + 4*(y&3) + 32*(y>>2);
3451
            for(x=0; x<8; x++){
3452
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3453
            }
3454
        }
3455
        for(y=0; y<8; y++){
3456
            const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3457
            for(x=0; x<8; x++){
3458
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3459
            }
3460
        }
3461
    
3462
        skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3463
        
3464
        //FIXME deblock filter, non_zero_count_cache init ...
3465
        memset(h->non_zero_count[mb_xy], 16, 16);
3466
        s->current_picture.qscale_table[mb_xy]= s->qscale;
3467
        
3468
        return 0;
3469
    }
3470
        
3471
    fill_caches(h, mb_type);
3472

    
3473
    //mb_pred
3474
    if(IS_INTRA(mb_type)){
3475
//            init_top_left_availability(h);
3476
            if(IS_INTRA4x4(mb_type)){
3477
                int i;
3478

    
3479
//                fill_intra4x4_pred_table(h);
3480
                for(i=0; i<16; i++){
3481
                    const int mode_coded= !get_bits1(&s->gb);
3482
                    const int predicted_mode=  pred_intra_mode(h, i);
3483
                    int mode;
3484

    
3485
                    if(mode_coded){
3486
                        const int rem_mode= get_bits(&s->gb, 3);
3487
                        if(rem_mode<predicted_mode)
3488
                            mode= rem_mode;
3489
                        else
3490
                            mode= rem_mode + 1;
3491
                    }else{
3492
                        mode= predicted_mode;
3493
                    }
3494
                    
3495
                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3496
                }
3497
                write_back_intra_pred_mode(h);
3498
                if( check_intra4x4_pred_mode(h) < 0)
3499
                    return -1;
3500
            }else{
3501
                h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3502
                if(h->intra16x16_pred_mode < 0)
3503
                    return -1;
3504
            }
3505
            h->chroma_pred_mode= get_ue_golomb(&s->gb);
3506

    
3507
            h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3508
            if(h->chroma_pred_mode < 0)
3509
                return -1;
3510
    }else if(partition_count==4){
3511
        int i, j, sub_partition_count[4], list, ref[2][4];
3512
        
3513
        if(h->slice_type == B_TYPE){
3514
            for(i=0; i<4; i++){
3515
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3516
                if(h->sub_mb_type[i] >=13){
3517
                    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);
3518
                    return -1;
3519
                }
3520
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3521
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3522
            }
3523
        }else{
3524
            assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3525
            for(i=0; i<4; i++){
3526
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3527
                if(h->sub_mb_type[i] >=4){
3528
                    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);
3529
                    return -1;
3530
                }
3531
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3532
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3533
            }
3534
        }
3535
        
3536
        for(list=0; list<2; list++){
3537
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3538
            if(ref_count == 0) continue;
3539
            for(i=0; i<4; i++){
3540
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3541
                    ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3542
                }else{
3543
                 //FIXME
3544
                    ref[list][i] = -1;
3545
                }
3546
            }
3547
        }
3548
        
3549
        for(list=0; list<2; list++){
3550
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3551
            if(ref_count == 0) continue;
3552

    
3553
            for(i=0; i<4; i++){
3554
                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3555
                h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3556

    
3557
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3558
                    const int sub_mb_type= h->sub_mb_type[i];
3559
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3560
                    for(j=0; j<sub_partition_count[i]; j++){
3561
                        int mx, my;
3562
                        const int index= 4*i + block_width*j;
3563
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3564
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3565
                        mx += get_se_golomb(&s->gb);
3566
                        my += get_se_golomb(&s->gb);
3567
                        tprintf("final mv:%d %d\n", mx, my);
3568

    
3569
                        if(IS_SUB_8X8(sub_mb_type)){
3570
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= 
3571
                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3572
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= 
3573
                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3574
                        }else if(IS_SUB_8X4(sub_mb_type)){
3575
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3576
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3577
                        }else if(IS_SUB_4X8(sub_mb_type)){
3578
                            mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3579
                            mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3580
                        }else{
3581
                            assert(IS_SUB_4X4(sub_mb_type));
3582
                            mv_cache[ 0 ][0]= mx;
3583
                            mv_cache[ 0 ][1]= my;
3584
                        }
3585
                    }
3586
                }else{
3587
                    uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3588
                    p[0] = p[1]=
3589
                    p[8] = p[9]= 0;
3590
                }
3591
            }
3592
        }
3593
    }else if(!IS_DIRECT(mb_type)){
3594
        int list, mx, my, i;
3595
         //FIXME we should set ref_idx_l? to 0 if we use that later ...
3596
        if(IS_16X16(mb_type)){
3597
            for(list=0; list<2; list++){
3598
                if(h->ref_count[0]>0){
3599
                    if(IS_DIR(mb_type, 0, list)){
3600
                        const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3601
                        fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3602
                    }
3603
                }
3604
            }
3605
            for(list=0; list<2; list++){
3606
                if(IS_DIR(mb_type, 0, list)){
3607
                    pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3608
                    mx += get_se_golomb(&s->gb);
3609
                    my += get_se_golomb(&s->gb);
3610
                    tprintf("final mv:%d %d\n", mx, my);
3611

    
3612
                    fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
3613
                }
3614
            }
3615
        }
3616
        else if(IS_16X8(mb_type)){
3617
            for(list=0; list<2; list++){
3618
                if(h->ref_count[list]>0){
3619
                    for(i=0; i<2; i++){
3620
                        if(IS_DIR(mb_type, i, list)){
3621
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3622
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3623
                        }
3624
                    }
3625
                }
3626
            }
3627
            for(list=0; list<2; list++){
3628
                for(i=0; i<2; i++){
3629
                    if(IS_DIR(mb_type, i, list)){
3630
                        pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3631
                        mx += get_se_golomb(&s->gb);
3632
                        my += get_se_golomb(&s->gb);
3633
                        tprintf("final mv:%d %d\n", mx, my);
3634

    
3635
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
3636
                    }
3637
                }
3638
            }
3639
        }else{
3640
            assert(IS_8X16(mb_type));
3641
            for(list=0; list<2; list++){
3642
                if(h->ref_count[list]>0){
3643
                    for(i=0; i<2; i++){
3644
                        if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3645
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3646
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3647
                        }
3648
                    }
3649
                }
3650
            }
3651
            for(list=0; list<2; list++){
3652
                for(i=0; i<2; i++){
3653
                    if(IS_DIR(mb_type, i, list)){
3654
                        pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3655
                        mx += get_se_golomb(&s->gb);
3656
                        my += get_se_golomb(&s->gb);
3657
                        tprintf("final mv:%d %d\n", mx, my);
3658

    
3659
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
3660
                    }
3661
                }
3662
            }
3663
        }
3664
    }
3665
    
3666
    if(IS_INTER(mb_type))
3667
        write_back_motion(h, mb_type);
3668
    
3669
    if(!IS_INTRA16x16(mb_type)){
3670
        cbp= get_ue_golomb(&s->gb);
3671
        if(cbp > 47){
3672
            av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3673
            return -1;
3674
        }
3675
        
3676
        if(IS_INTRA4x4(mb_type))
3677
            cbp= golomb_to_intra4x4_cbp[cbp];
3678
        else
3679
            cbp= golomb_to_inter_cbp[cbp];
3680
    }
3681

    
3682
    if(cbp || IS_INTRA16x16(mb_type)){
3683
        int i8x8, i4x4, chroma_idx;
3684
        int chroma_qp, dquant;
3685
        GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3686
        const uint8_t *scan, *dc_scan;
3687
        
3688
//        fill_non_zero_count_cache(h);
3689

    
3690
        if(IS_INTERLACED(mb_type)){
3691
            scan= field_scan;
3692
            dc_scan= luma_dc_field_scan;
3693
        }else{
3694
            scan= zigzag_scan;
3695
            dc_scan= luma_dc_zigzag_scan;
3696
        }
3697

    
3698
        dquant= get_se_golomb(&s->gb);
3699

    
3700
        if( dquant > 25 || dquant < -26 ){
3701
            av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3702
            return -1;
3703
        }
3704
        
3705
        s->qscale += dquant;
3706
        if(((unsigned)s->qscale) > 51){
3707
            if(s->qscale<0) s->qscale+= 52;
3708
            else            s->qscale-= 52;
3709
        }
3710
        
3711
        h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3712
        if(IS_INTRA16x16(mb_type)){
3713
            if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3714
                return -1; //FIXME continue if partotioned and other retirn -1 too
3715
            }
3716

    
3717
            assert((cbp&15) == 0 || (cbp&15) == 15);
3718

    
3719
            if(cbp&15){
3720
                for(i8x8=0; i8x8<4; i8x8++){
3721
                    for(i4x4=0; i4x4<4; i4x4++){
3722
                        const int index= i4x4 + 4*i8x8;
3723
                        if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3724
                            return -1;
3725
                        }
3726
                    }
3727
                }
3728
            }else{
3729
                fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
3730
            }
3731
        }else{
3732
            for(i8x8=0; i8x8<4; i8x8++){
3733
                if(cbp & (1<<i8x8)){
3734
                    for(i4x4=0; i4x4<4; i4x4++){
3735
                        const int index= i4x4 + 4*i8x8;
3736
                        
3737
                        if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3738
                            return -1;
3739
                        }
3740
                    }
3741
                }else{
3742
                    uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3743
                    nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3744
                }
3745
            }
3746
        }
3747
        
3748
        if(cbp&0x30){
3749
            for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3750
                if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3751
                    return -1;
3752
                }
3753
        }
3754

    
3755
        if(cbp&0x20){
3756
            for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3757
                for(i4x4=0; i4x4<4; i4x4++){
3758
                    const int index= 16 + 4*chroma_idx + i4x4;
3759
                    if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3760
                        return -1;
3761
                    }
3762
                }
3763
            }
3764
        }else{
3765
            uint8_t * const nnz= &h->non_zero_count_cache[0];
3766
            nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3767
            nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3768
        }
3769
    }else{
3770
        uint8_t * const nnz= &h->non_zero_count_cache[0];
3771
        fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
3772
        nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3773
        nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3774
    }
3775
    s->current_picture.qscale_table[mb_xy]= s->qscale;
3776
    write_back_non_zero_count(h);
3777

    
3778
    return 0;
3779
}
3780

    
3781
static int decode_cabac_mb_type( H264Context *h ) {
3782
    MpegEncContext * const s = &h->s;
3783

    
3784
    if( h->slice_type == I_TYPE ) {
3785
        const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3786
        int ctx = 0;
3787
        int mb_type;
3788

    
3789
        if( s->mb_x > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-1] ) )
3790
            ctx++;
3791
        if( s->mb_y > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
3792
            ctx++;
3793

    
3794
        if( get_cabac( &h->cabac, &h->cabac_state[3+ctx] ) == 0 )
3795
            return 0;   /* I4x4 */
3796

    
3797
        if( get_cabac_terminate( &h->cabac ) )
3798
            return 25;  /* PCM */
3799

    
3800
        mb_type = 1;    /* I16x16 */
3801
        if( get_cabac( &h->cabac, &h->cabac_state[3+3] ) )
3802
            mb_type += 12;  /* cbp_luma != 0 */
3803

    
3804
        if( get_cabac( &h->cabac, &h->cabac_state[3+4] ) ) {
3805
            if( get_cabac( &h->cabac, &h->cabac_state[3+5] ) )
3806
                mb_type += 4 * 2;   /* cbp_chroma == 2 */
3807
            else
3808
                mb_type += 4 * 1;   /* cbp_chroma == 1 */
3809
        }
3810
        if( get_cabac( &h->cabac, &h->cabac_state[3+6] ) )
3811
            mb_type += 2;
3812
        if( get_cabac( &h->cabac, &h->cabac_state[3+7] ) )
3813
            mb_type += 1;
3814
        return mb_type;
3815

    
3816
    } else if( h->slice_type == P_TYPE ) {
3817
        if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
3818
            /* P-type */
3819
            if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
3820
                if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
3821
                    return 0; /* P_L0_D16x16; */
3822
                else
3823
                    return 3; /* P_8x8; */
3824
            } else {
3825
                if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
3826
                    return 2; /* P_L0_D8x16; */
3827
                else
3828
                    return 1; /* P_L0_D16x8; */
3829
            }
3830
        } else {
3831
            int mb_type;
3832
            /* I-type */
3833
            if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
3834
                return 5+0; /* I_4x4 */
3835
            if( get_cabac_terminate( &h->cabac ) )
3836
                return 5+25; /*I_PCM */
3837
            mb_type = 5+1;    /* I16x16 */
3838
            if( get_cabac( &h->cabac, &h->cabac_state[17+1] ) )
3839
                mb_type += 12;  /* cbp_luma != 0 */
3840

    
3841
            if( get_cabac( &h->cabac, &h->cabac_state[17+2] ) ) {
3842
                if( get_cabac( &h->cabac, &h->cabac_state[17+2] ) )
3843
                    mb_type += 4 * 2;   /* cbp_chroma == 2 */
3844
                else
3845
                    mb_type += 4 * 1;   /* cbp_chroma == 1 */
3846
            }
3847
            if( get_cabac( &h->cabac, &h->cabac_state[17+3] ) )
3848
                mb_type += 2;
3849
            if( get_cabac( &h->cabac, &h->cabac_state[17+3] ) )
3850
                mb_type += 1;
3851

    
3852
            return mb_type;
3853
        }
3854
    } else {
3855
        /* TODO do others frames types */
3856
        return -1;
3857
    }
3858
}
3859

    
3860
static int decode_cabac_mb_skip( H264Context *h) {
3861
    MpegEncContext * const s = &h->s;
3862
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3863
    const int mba_xy = mb_xy - 1;
3864
    const int mbb_xy = mb_xy - s->mb_stride;
3865
    int ctx = 0;
3866

    
3867
    if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mba_xy] ) )
3868
        ctx++;
3869
    if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ) )
3870
        ctx++;
3871

    
3872
    if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
3873
        return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
3874
    else /* B-frame */
3875
        return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
3876
}
3877

    
3878
static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
3879
    int mode = 0;
3880

    
3881
    if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
3882
        return pred_mode;
3883

    
3884
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3885
        mode += 1;
3886
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3887
        mode += 2;
3888
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3889
        mode += 4;
3890
    if( mode >= pred_mode )
3891
        return mode + 1;
3892
    else
3893
        return mode;
3894
}
3895

    
3896
static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
3897
    MpegEncContext * const s = &h->s;
3898
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3899
    const int mba_xy = mb_xy - 1;
3900
    const int mbb_xy = mb_xy - s->mb_stride;
3901

    
3902
    int ctx = 0;
3903

    
3904
    /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
3905
    if( s->mb_x > 0 && h->chroma_pred_mode_table[mba_xy] != 0 )
3906
        ctx++;
3907

    
3908
    if( s->mb_y > 0 && h->chroma_pred_mode_table[mbb_xy] != 0 )
3909
        ctx++;
3910

    
3911
    if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
3912
        return 0;
3913

    
3914
    if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
3915
        return 1;
3916
    if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
3917
        return 2;
3918
    else
3919
        return 3;
3920
}
3921

    
3922
static const uint8_t block_idx_x[16] = {
3923
    0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
3924
};
3925
static const uint8_t block_idx_y[16] = {
3926
    0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
3927
};
3928
static const uint8_t block_idx_xy[4][4] = {
3929
    { 0, 2, 8,  10},
3930
    { 1, 3, 9,  11},
3931
    { 4, 6, 12, 14},
3932
    { 5, 7, 13, 15}
3933
};
3934

    
3935
static int decode_cabac_mb_cbp_luma( H264Context *h) {
3936
    MpegEncContext * const s = &h->s;
3937
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3938

    
3939
    int cbp = 0;
3940
    int i8x8;
3941

    
3942
    h->cbp_table[mb_xy] = 0;  /* FIXME aaahahahah beurk */
3943

    
3944
    for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
3945
        int mba_xy = -1;
3946
        int mbb_xy = -1;
3947
        int x, y;
3948
        int ctx = 0;
3949

    
3950
        x = block_idx_x[4*i8x8];
3951
        y = block_idx_y[4*i8x8];
3952

    
3953
        if( x > 0 )
3954
            mba_xy = mb_xy;
3955
        else if( s->mb_x > 0 )
3956
            mba_xy = mb_xy - 1;
3957

    
3958
        if( y > 0 )
3959
            mbb_xy = mb_xy;
3960
        else if( s->mb_y > 0 )
3961
            mbb_xy = mb_xy - s->mb_stride;
3962

    
3963
        /* No need to test for skip as we put 0 for skip block */
3964
        if( mba_xy >= 0 ) {
3965
            int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
3966
            if( ((h->cbp_table[mba_xy] >> i8x8a)&0x01) == 0 )
3967
                ctx++;
3968
        }
3969

    
3970
        if( mbb_xy >= 0 ) {
3971
            int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
3972
            if( ((h->cbp_table[mbb_xy] >> i8x8b)&0x01) == 0 )
3973
                ctx += 2;
3974
        }
3975

    
3976
        if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
3977
            cbp |= 1 << i8x8;
3978
            h->cbp_table[mb_xy] = cbp;  /* FIXME aaahahahah beurk */
3979
        }
3980
    }
3981
    return cbp;
3982
}
3983
static int decode_cabac_mb_cbp_chroma( H264Context *h) {
3984
    MpegEncContext * const s = &h->s;
3985
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3986
    int ctx;
3987
    int cbp_a, cbp_b;
3988

    
3989
    /* No need to test for skip */
3990
    if( s->mb_x > 0 )
3991
        cbp_a = (h->cbp_table[mb_xy-1]>>4)&0x03;
3992
    else
3993
        cbp_a = -1;
3994

    
3995
    if( s->mb_y > 0 )
3996
        cbp_b = (h->cbp_table[mb_xy-s->mb_stride]>>4)&0x03;
3997
    else
3998
        cbp_b = -1;
3999

    
4000
    ctx = 0;
4001
    if( cbp_a > 0 ) ctx++;
4002
    if( cbp_b > 0 ) ctx += 2;
4003
    if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
4004
        return 0;
4005

    
4006
    ctx = 4;
4007
    if( cbp_a == 2 ) ctx++;
4008
    if( cbp_b == 2 ) ctx += 2;
4009
    if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) )
4010
        return 2;
4011
    else
4012
        return 1;
4013
}
4014
static int decode_cabac_mb_dqp( H264Context *h) {
4015
    MpegEncContext * const s = &h->s;
4016
    int mbn_xy;
4017
    int   ctx = 0;
4018
    int   val = 0;
4019

    
4020
    if( s->mb_x > 0 )
4021
        mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;
4022
    else
4023
        mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;
4024

    
4025
    if( mbn_xy >= 0 && h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
4026
        ctx++;
4027

    
4028
    while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
4029
        if( ctx < 2 )
4030
            ctx = 2;
4031
        else
4032
            ctx = 3;
4033
        val++;
4034
    }
4035

    
4036
    if( val&0x01 )
4037
        return (val + 1)/2;
4038
    else
4039
        return -(val + 1)/2;
4040
}
4041
static int decode_cabac_mb_sub_type( H264Context *h ) {
4042
    if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
4043
        return 0;   /* 8x8 */
4044
    if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
4045
        return 1;   /* 8x4 */
4046
    if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
4047
        return 2;   /* 4x8 */
4048
    return 3;       /* 4x4 */
4049
}
4050

    
4051
static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
4052
    int refa = h->ref_cache[list][scan8[n] - 1];
4053
    int refb = h->ref_cache[list][scan8[n] - 8];
4054
    int ref  = 0;
4055
    int ctx  = 0;
4056

    
4057
    if( refa > 0 )
4058
        ctx++;
4059
    if( refb > 0 )
4060
        ctx += 2;
4061

    
4062
    while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
4063
        ref++;
4064
        if( ctx < 4 )
4065
            ctx = 4;
4066
        else
4067
            ctx = 5;
4068
    }
4069
    return ref;
4070
}
4071

    
4072
static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
4073
    int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
4074
               abs( h->mvd_cache[list][scan8[n] - 8][l] );
4075
    int ctxbase = (l == 0) ? 40 : 47;
4076
    int ctx;
4077
    int mvd = 0;
4078

    
4079
    if( amvd < 3 )
4080
        ctx = 0;
4081
    else if( amvd > 32 )
4082
        ctx = 2;
4083
    else
4084
        ctx = 1;
4085

    
4086
    while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
4087
        mvd++;
4088
        if( ctx < 3 )
4089
            ctx = 3;
4090
        else if( ctx < 6 )
4091
            ctx++;
4092
    }
4093

    
4094
    if( mvd >= 9 ) {
4095
        int k = 3;
4096
        while( get_cabac_bypass( &h->cabac ) ) {
4097
            mvd += 1 << k;
4098
            k++;
4099
        }
4100
        while( k-- ) {
4101
            if( get_cabac_bypass( &h->cabac ) )
4102
                mvd += 1 << k;
4103
        }
4104
    }
4105
    if( mvd != 0 && get_cabac_bypass( &h->cabac ) )
4106
        return -mvd;
4107
    return mvd;
4108
}
4109

    
4110

    
4111
static int get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {
4112
    MpegEncContext * const s = &h->s;
4113
    const int mb_xy  = s->mb_x + s->mb_y*s->mb_stride;
4114
    int mba_xy = -1;
4115
    int mbb_xy = -1;
4116

    
4117
    int nza = -1;
4118
    int nzb = -1;
4119
    int ctx = 0;
4120

    
4121
    if( cat == 0 ) {
4122
        if( s->mb_x > 0 ) {
4123
            mba_xy = mb_xy - 1;
4124
            if( IS_INTRA16x16(s->current_picture.mb_type[mba_xy] ) )
4125
                    nza = h->cbp_table[mba_xy]&0x100;
4126
        }
4127
        if( s->mb_y > 0 ) {
4128
            mbb_xy = mb_xy - s->mb_stride;
4129
            if( IS_INTRA16x16(s->current_picture.mb_type[mbb_xy] ) )
4130
                    nzb = h->cbp_table[mbb_xy]&0x100;
4131
        }
4132
    } else if( cat == 1 || cat == 2 ) {
4133
        int i8x8a, i8x8b;
4134
        int x, y;
4135

    
4136
        x = block_idx_x[idx];
4137
        y = block_idx_y[idx];
4138

    
4139
        if( x > 0 )
4140
            mba_xy = mb_xy;
4141
        else if( s->mb_x > 0 )
4142
            mba_xy = mb_xy - 1;
4143

    
4144
        if( y > 0 )
4145
            mbb_xy = mb_xy;
4146
        else if( s->mb_y > 0 )
4147
            mbb_xy = mb_xy - s->mb_stride;
4148

    
4149
        /* No need to test for skip */
4150
        if( mba_xy >= 0 ) {
4151
            i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4152

    
4153
            if( !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
4154
                ((h->cbp_table[mba_xy]&0x0f)>>i8x8a))
4155
                nza = h->non_zero_count_cache[scan8[idx] - 1];
4156
        }
4157

    
4158
        if( mbb_xy >= 0 ) {
4159
            i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4160

    
4161
            if( !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
4162
                ((h->cbp_table[mbb_xy]&0x0f)>>i8x8b))
4163
                nzb = h->non_zero_count_cache[scan8[idx] - 8];
4164
        }
4165
    } else if( cat == 3 ) {
4166
        if( s->mb_x > 0 ) {
4167
            mba_xy = mb_xy - 1;
4168

    
4169
            if( !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
4170
                (h->cbp_table[mba_xy]&0x30) )
4171
                nza = (h->cbp_table[mba_xy]>>(6+idx))&0x01;
4172
        }
4173
        if( s->mb_y > 0 ) {
4174
            mbb_xy = mb_xy - s->mb_stride;
4175

    
4176
            if( !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
4177
                (h->cbp_table[mbb_xy]&0x30) )
4178
                nzb = (h->cbp_table[mbb_xy]>>(6+idx))&0x01;
4179
        }
4180
    } else if( cat == 4 ) {
4181
        int idxc = idx % 4 ;
4182
        if( idxc == 1 || idxc == 3 )
4183
            mba_xy = mb_xy;
4184
        else if( s->mb_x > 0 )
4185
            mba_xy = mb_xy -1;
4186

    
4187
        if( idxc == 2 || idxc == 3 )
4188
            mbb_xy = mb_xy;
4189
        else if( s->mb_y > 0 )
4190
            mbb_xy = mb_xy - s->mb_stride;
4191

    
4192
        if( mba_xy >= 0 &&
4193
            !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
4194
            (h->cbp_table[mba_xy]&0x30) == 0x20 )
4195
            nza = h->non_zero_count_cache[scan8[16+idx] - 1];
4196

    
4197
        if( mbb_xy >= 0 &&
4198
            !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
4199
            (h->cbp_table[mbb_xy]&0x30) == 0x20 )
4200
            nzb = h->non_zero_count_cache[scan8[16+idx] - 8];
4201
    }
4202

    
4203
    if( ( mba_xy < 0 && IS_INTRA( s->current_picture.mb_type[mb_xy] ) ) ||
4204
        ( mba_xy >= 0 && IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) ) ||
4205
          nza > 0 )
4206
        ctx++;
4207

    
4208
    if( ( mbb_xy < 0 && IS_INTRA( s->current_picture.mb_type[mb_xy] ) ) ||
4209
        ( mbb_xy >= 0 && IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) ) ||
4210
          nzb > 0 )
4211
        ctx += 2;
4212

    
4213
    return ctx + 4 * cat;
4214
}
4215

    
4216
static int decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, int qp, int max_coeff) {
4217
    const int mb_xy  = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
4218
    const uint16_t *qmul= dequant_coeff[qp];
4219
    static const int significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
4220
    static const int last_significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
4221
    static const int coeff_abs_level_m1_offset[5] = { 0, 10, 20, 30, 39 };
4222

    
4223
    int coeff[16];
4224

    
4225
    int last = 0;
4226
    int coeff_count = 0;
4227
    int nz[16] = {0};
4228
    int i;
4229

    
4230
    int abslevel1 = 0;
4231
    int abslevelgt1 = 0;
4232

    
4233
    /* cat: 0-> DC 16x16  n = 0
4234
     *      1-> AC 16x16  n = luma4x4idx
4235
     *      2-> Luma4x4   n = luma4x4idx
4236
     *      3-> DC Chroma n = iCbCr
4237
     *      4-> AC Chroma n = 4 * iCbCr + chroma4x4idx
4238
     */
4239

    
4240
    /* read coded block flag */
4241
    if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {
4242
        if( cat == 1 || cat == 2 )
4243
            h->non_zero_count_cache[scan8[n]] = 0;
4244
        else if( cat == 4 )
4245
            h->non_zero_count_cache[scan8[16+n]] = 0;
4246

    
4247
        return 0;
4248
    }
4249

    
4250
    while( last < max_coeff - 1 ) {
4251
        int ctx = FFMIN( last, max_coeff - 2 );
4252

    
4253
        if( get_cabac( &h->cabac, &h->cabac_state[105+significant_coeff_flag_offset[cat]+ctx] ) == 0 ) {
4254
            nz[last++] = 0;
4255
        }
4256
        else {
4257
            nz[last++] = 1;
4258
            coeff_count++;
4259
            if( get_cabac( &h->cabac, &h->cabac_state[166+last_significant_coeff_flag_offset[cat]+ctx] ) ) {
4260
                while( last < max_coeff ) {
4261
                    nz[last++] = 0;
4262
                }
4263
                break;
4264
            }
4265
        }
4266
    }
4267
    if( last == max_coeff -1 ) {
4268
        nz[last++] = 1;
4269
        coeff_count++;
4270
    }
4271

    
4272
    if( cat == 0 && coeff_count > 0 )
4273
        h->cbp_table[mb_xy] |= 0x100;
4274
    else if( cat == 1 || cat == 2 )
4275
        h->non_zero_count_cache[scan8[n]] = coeff_count;
4276
    else if( cat == 3 && coeff_count > 0 )
4277
        h->cbp_table[mb_xy] |= 0x40 << n;
4278
    else if( cat == 4 )
4279
        h->non_zero_count_cache[scan8[16+n]] = coeff_count;
4280

    
4281
    for( i = coeff_count - 1; i >= 0; i-- ) {
4282
        int coeff_abs_m1;
4283

    
4284
        int ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 + 1 )) + coeff_abs_level_m1_offset[cat];
4285

    
4286
        if( get_cabac( &h->cabac, &h->cabac_state[227+ctx] ) == 0 ) {
4287
            coeff_abs_m1 = 0;
4288
        } else {
4289
            coeff_abs_m1 = 1;
4290
            ctx = 5 + FFMIN( 4, abslevelgt1 ) + coeff_abs_level_m1_offset[cat];
4291
            while( coeff_abs_m1 < 14 && get_cabac( &h->cabac, &h->cabac_state[227+ctx] ) ) {
4292
                coeff_abs_m1++;
4293
            }
4294
        }
4295

    
4296
        if( coeff_abs_m1 >= 14 ) {
4297
            int j = 0;
4298
            while( get_cabac_bypass( &h->cabac ) ) {
4299
                coeff_abs_m1 += 1 << j;
4300
                j++;
4301
            }
4302

    
4303
            while( j-- ) {
4304
                if( get_cabac_bypass( &h->cabac ) )
4305
                    coeff_abs_m1 += 1 << j ;
4306
            }
4307
        }
4308
        if( get_cabac_bypass( &h->cabac ) )
4309
            coeff[i] = -1 *( coeff_abs_m1 + 1 );
4310
        else
4311
            coeff[i] = coeff_abs_m1 + 1;
4312

    
4313
        if( coeff_abs_m1 == 0 )
4314
            abslevel1++;
4315
        else
4316
            abslevelgt1++;
4317
    }
4318

    
4319
    if( cat == 0 || cat == 3 ) { /* DC */
4320
        int j;
4321
        for( i = 0, j = 0; j < coeff_count; i++ ) {
4322
            if( nz[i] ) {
4323
                block[scantable[i]] = coeff[j];
4324

    
4325
                j++;
4326
            }
4327
        }
4328

    
4329
    } else { /* AC */
4330
        int j;
4331
        for( i = 0, j = 0; j < coeff_count; i++ ) {
4332
            if( nz[i] ) {
4333
                block[scantable[i]] = coeff[j] * qmul[scantable[i]];
4334

    
4335
                j++;
4336
            }
4337
        }
4338
    }
4339
    return 0;
4340
}
4341

    
4342
/**
4343
 * decodes a macroblock
4344
 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4345
 */
4346
static int decode_mb_cabac(H264Context *h) {
4347
    MpegEncContext * const s = &h->s;
4348
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4349
    int mb_type, partition_count, cbp = 0;
4350

    
4351
    s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?)
4352

    
4353
    if( h->slice_type == B_TYPE ) {
4354
        av_log( h->s.avctx, AV_LOG_ERROR, "B-frame not supported with CABAC\n" );
4355
        return -1;
4356
    }
4357
    if( h->sps.mb_aff ) {
4358
        av_log( h->s.avctx, AV_LOG_ERROR, "Fields not supported with CABAC\n" );
4359
        return -1;
4360
    }
4361

    
4362
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
4363
        /* read skip flags */
4364
        if( decode_cabac_mb_skip( h ) ) {
4365
            int mx, my;
4366

    
4367
            /* skip mb */
4368
            mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4369

    
4370
            memset(h->non_zero_count[mb_xy], 0, 16);
4371
            memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4372
#if 0
4373
            if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
4374
                h->mb_field_decoding_flag= get_bits1(&s->gb);
4375
            }
4376
            if(h->mb_field_decoding_flag)
4377
                mb_type|= MB_TYPE_INTERLACED;
4378
#endif
4379

    
4380
            fill_caches(h, mb_type); //FIXME check what is needed and what not ...
4381
            pred_pskip_motion(h, &mx, &my);
4382
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4383
            fill_rectangle(  h->mvd_cache[0][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
4384
            fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4385
            write_back_motion(h, mb_type);
4386

    
4387
            s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
4388
            s->current_picture.qscale_table[mb_xy]= s->qscale;
4389
            h->slice_table[ mb_xy ]= h->slice_num;
4390
            h->cbp_table[mb_xy] = 0;
4391
            h->chroma_pred_mode_table[mb_xy] = 0;
4392
            h->last_qscale_diff = 0;
4393

    
4394
            h->prev_mb_skiped= 1;
4395

    
4396
            return 0;
4397

    
4398
        }
4399
    }
4400
    h->prev_mb_skiped = 0;
4401

    
4402
    if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {
4403
        av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );
4404
        return -1;
4405
    }
4406

    
4407
    if( h->slice_type == P_TYPE ) {
4408
        if( mb_type < 5) {
4409
            partition_count= p_mb_type_info[mb_type].partition_count;
4410
            mb_type=         p_mb_type_info[mb_type].type;
4411
        } else {
4412
            mb_type -= 5;
4413
            goto decode_intra_mb;
4414
        }
4415
    } else {
4416
       assert(h->slice_type == I_TYPE);
4417
decode_intra_mb:
4418
        partition_count = 0;
4419
        cbp= i_mb_type_info[mb_type].cbp;
4420
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4421
        mb_type= i_mb_type_info[mb_type].type;
4422
    }
4423
#if 0
4424
    if(h->mb_field_decoding_flag)
4425
        mb_type |= MB_TYPE_INTERLACED;
4426
#endif
4427

    
4428
    s->current_picture.mb_type[mb_xy]= mb_type;
4429
    h->slice_table[ mb_xy ]= h->slice_num;
4430

    
4431
    if(IS_INTRA_PCM(mb_type)) {
4432
        /* TODO */
4433
        h->cbp_table[mb_xy] = 0xf +4*2;
4434
        h->chroma_pred_mode_table[mb_xy] = 0;
4435
        s->current_picture.qscale_table[mb_xy]= s->qscale;
4436
        return -1;
4437
    }
4438

    
4439
    fill_caches(h, mb_type);
4440

    
4441
    if( IS_INTRA( mb_type ) ) {
4442
        if( IS_INTRA4x4( mb_type ) ) {
4443
            int i;
4444
            for( i = 0; i < 16; i++ ) {
4445
                int pred = pred_intra_mode( h, i );
4446
                h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
4447

    
4448
                //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
4449
            }
4450
            write_back_intra_pred_mode(h);
4451
            if( check_intra4x4_pred_mode(h) < 0 ) return -1;
4452
        } else {
4453
            h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
4454
            if( h->intra16x16_pred_mode < 0 ) return -1;
4455
        }
4456
        h->chroma_pred_mode_table[mb_xy] =
4457
            h->chroma_pred_mode          = decode_cabac_mb_chroma_pre_mode( h );
4458

    
4459
        h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );
4460
        if( h->chroma_pred_mode < 0 ) return -1;
4461
    } else if( partition_count == 4 ) {
4462
        int i, j, sub_partition_count[4], list, ref[2][4];
4463

    
4464
        /* Only P-frame */
4465
        for( i = 0; i < 4; i++ ) {
4466
            h->sub_mb_type[i] = decode_cabac_mb_sub_type( h );
4467
            sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4468
            h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4469
        }
4470

    
4471
        for( list = 0; list < 2; list++ ) {
4472
            if( h->ref_count[list] > 0 ) {
4473
                for( i = 0; i < 4; i++ ) {
4474
                    if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
4475
                        if( h->ref_count[list] > 1 )
4476
                            ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
4477
                        else
4478
                            ref[list][i] = 0;
4479
                    } else {
4480
                        ref[list][i] = -1;
4481
                    }
4482
                    h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4483
                    h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4484
                }
4485
            }
4486
        }
4487

    
4488
        for(list=0; list<2; list++){
4489

    
4490
            for(i=0; i<4; i++){
4491
                //h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4492
                //h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4493

    
4494
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
4495
                    const int sub_mb_type= h->sub_mb_type[i];
4496
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4497
                    for(j=0; j<sub_partition_count[i]; j++){
4498
                        int mpx, mpy;
4499
                        int mx, my;
4500
                        const int index= 4*i + block_width*j;
4501
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4502
                        int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
4503
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
4504

    
4505
                        mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
4506
                        my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
4507
                        tprintf("final mv:%d %d\n", mx, my);
4508

    
4509
                        if(IS_SUB_8X8(sub_mb_type)){
4510
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
4511
                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4512
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
4513
                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4514

    
4515
                            mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=
4516
                            mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
4517
                            mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=
4518
                            mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
4519
                        }else if(IS_SUB_8X4(sub_mb_type)){
4520
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4521
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4522

    
4523
                            mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;
4524
                            mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;
4525
                        }else if(IS_SUB_4X8(sub_mb_type)){
4526
                            mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4527
                            mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4528

    
4529
                            mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;
4530
                            mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;
4531
                        }else{
4532
                            assert(IS_SUB_4X4(sub_mb_type));
4533
                            mv_cache[ 0 ][0]= mx;
4534
                            mv_cache[ 0 ][1]= my;
4535

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