Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 46f2f05f

History | View | Annotate | Download (236 KB)

1
/*
2
 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3
 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4
 *
5
 * This library is free software; you can redistribute it and/or
6
 * modify it under the terms of the GNU Lesser General Public
7
 * License as published by the Free Software Foundation; either
8
 * version 2 of the License, or (at your option) any later version.
9
 *
10
 * This library is distributed in the hope that it will be useful,
11
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13
 * Lesser General Public License for more details.
14
 *
15
 * You should have received a copy of the GNU Lesser General Public
16
 * License along with this library; if not, write to the Free Software
17
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
18
 *
19
 */
20
 
21
/**
22
 * @file h264.c
23
 * H.264 / AVC / MPEG4 part10 codec.
24
 * @author Michael Niedermayer <michaelni@gmx.at>
25
 */
26

    
27
#include "common.h"
28
#include "dsputil.h"
29
#include "avcodec.h"
30
#include "mpegvideo.h"
31
#include "h264data.h"
32
#include "golomb.h"
33

    
34
#include "cabac.h"
35

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

    
39
#define interlaced_dct interlaced_dct_is_a_bad_name
40
#define mb_intra mb_intra_isnt_initalized_see_mb_type
41

    
42
#define LUMA_DC_BLOCK_INDEX   25
43
#define CHROMA_DC_BLOCK_INDEX 26
44

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

    
52
#define MAX_SPS_COUNT 32
53
#define MAX_PPS_COUNT 256
54

    
55
#define MAX_MMCO_COUNT 66
56

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

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

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

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

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

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

    
161
    int chroma_qp; //QPc
162

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

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

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

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

    
201
    int block_offset[16+8];
202
    int chroma_subblock_offset[16]; //FIXME remove
203
    
204
    uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
205
    uint16_t *mb2b8_xy;
206
    int b_stride; //FIXME use s->b4_stride
207
    int b8_stride;
208

    
209
    int halfpel_flag;
210
    int thirdpel_flag;
211

    
212
    int unknown_svq3_flag;
213
    int next_slice_index;
214

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

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

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

    
257
    //Weighted pred stuff
258
    int use_weight;
259
    int use_weight_chroma;
260
    int luma_log2_weight_denom;
261
    int chroma_log2_weight_denom;
262
    int luma_weight[2][16];
263
    int luma_offset[2][16];
264
    int chroma_weight[2][16][2];
265
    int chroma_offset[2][16][2];
266
    int implicit_weight[16][16];
267
   
268
    //deblock
269
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0 
270
    int slice_alpha_c0_offset;
271
    int slice_beta_offset;
272
     
273
    int redundant_pic_count;
274
    
275
    int direct_spatial_mv_pred;
276
    int dist_scale_factor[16];
277

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

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

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

    
325
}H264Context;
326

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

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

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

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

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

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

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

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

    
438
    if(for_deblock){
439
        topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
440
        top_type     = h->slice_table[top_xy     ] < 255 ? s->current_picture.mb_type[top_xy]     : 0;
441
        topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
442
        left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
443
        left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
444
    }else{
445
        topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
446
        top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
447
        topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
448
        left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
449
        left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
450
    }
451

    
452
    if(IS_INTRA(mb_type)){
453
        h->topleft_samples_available= 
454
        h->top_samples_available= 
455
        h->left_samples_available= 0xFFFF;
456
        h->topright_samples_available= 0xEEEA;
457

    
458
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
459
            h->topleft_samples_available= 0xB3FF;
460
            h->top_samples_available= 0x33FF;
461
            h->topright_samples_available= 0x26EA;
462
        }
463
        for(i=0; i<2; i++){
464
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
465
                h->topleft_samples_available&= 0xDF5F;
466
                h->left_samples_available&= 0x5F5F;
467
            }
468
        }
469
        
470
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
471
            h->topleft_samples_available&= 0x7FFF;
472
        
473
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
474
            h->topright_samples_available&= 0xFBFF;
475
    
476
        if(IS_INTRA4x4(mb_type)){
477
            if(IS_INTRA4x4(top_type)){
478
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
479
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
480
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
481
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
482
            }else{
483
                int pred;
484
                if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
485
                    pred= 2;
486
                else{
487
                    pred= -1;
488
                }
489
                h->intra4x4_pred_mode_cache[4+8*0]=
490
                h->intra4x4_pred_mode_cache[5+8*0]=
491
                h->intra4x4_pred_mode_cache[6+8*0]=
492
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
493
            }
494
            for(i=0; i<2; i++){
495
                if(IS_INTRA4x4(left_type[i])){
496
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
497
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
498
                }else{
499
                    int pred;
500
                    if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
501
                        pred= 2;
502
                    else{
503
                        pred= -1;
504
                    }
505
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
506
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
507
                }
508
            }
509
        }
510
    }
511
    
512
    
513
/*
514
0 . T T. T T T T 
515
1 L . .L . . . . 
516
2 L . .L . . . . 
517
3 . T TL . . . . 
518
4 L . .L . . . . 
519
5 L . .. . . . . 
520
*/
521
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
522
    if(top_type){
523
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
524
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
525
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
526
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
527
    
528
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
529
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
530
    
531
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
532
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
533
        
534
        h->top_cbp= h->cbp_table[top_xy];
535
    }else{
536
        h->non_zero_count_cache[4+8*0]=      
537
        h->non_zero_count_cache[5+8*0]=
538
        h->non_zero_count_cache[6+8*0]=
539
        h->non_zero_count_cache[7+8*0]=
540
    
541
        h->non_zero_count_cache[1+8*0]=
542
        h->non_zero_count_cache[2+8*0]=
543
    
544
        h->non_zero_count_cache[1+8*3]=
545
        h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
546
        
547
        if(IS_INTRA(mb_type)) h->top_cbp= 0x1C0;
548
        else                  h->top_cbp= 0;
549
    }
550
    
551
    if(left_type[0]){
552
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
553
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
554
        h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
555
        h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
556
        h->left_cbp= h->cbp_table[left_xy[0]]; //FIXME interlacing
557
    }else{
558
        h->non_zero_count_cache[3+8*1]= 
559
        h->non_zero_count_cache[3+8*2]= 
560
        h->non_zero_count_cache[0+8*1]= 
561
        h->non_zero_count_cache[0+8*4]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
562
        
563
        if(IS_INTRA(mb_type)) h->left_cbp= 0x1C0;//FIXME interlacing
564
        else                  h->left_cbp= 0;
565
    }
566
    
567
    if(left_type[1]){
568
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
569
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
570
        h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
571
        h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
572
    }else{
573
        h->non_zero_count_cache[3+8*3]= 
574
        h->non_zero_count_cache[3+8*4]= 
575
        h->non_zero_count_cache[0+8*2]= 
576
        h->non_zero_count_cache[0+8*5]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
577
    }
578
    
579
#if 1
580
    //FIXME direct mb can skip much of this
581
    if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
582
        int list;
583
        for(list=0; list<2; list++){
584
            if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list) && !IS_DIRECT(mb_type)){
585
                /*if(!h->mv_cache_clean[list]){
586
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
587
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
588
                    h->mv_cache_clean[list]= 1;
589
                }*/
590
                continue;
591
            }
592
            h->mv_cache_clean[list]= 0;
593
            
594
            if(IS_INTER(topleft_type)){
595
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
596
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
597
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
598
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
599
            }else{
600
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
601
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
602
            }
603
            
604
            if(IS_INTER(top_type)){
605
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
606
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
607
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
608
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
609
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
610
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
611
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
612
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
613
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
614
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
615
            }else{
616
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
617
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
618
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
619
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
620
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
621
            }
622

    
623
            if(IS_INTER(topright_type)){
624
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
625
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
626
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
627
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
628
            }else{
629
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
630
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
631
            }
632
            
633
            //FIXME unify cleanup or sth
634
            if(IS_INTER(left_type[0])){
635
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
636
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
637
                *(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]];
638
                *(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]];
639
                h->ref_cache[list][scan8[0] - 1 + 0*8]= 
640
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
641
            }else{
642
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
643
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
644
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
645
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
646
            }
647
            
648
            if(IS_INTER(left_type[1])){
649
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
650
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
651
                *(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]];
652
                *(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]];
653
                h->ref_cache[list][scan8[0] - 1 + 2*8]= 
654
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
655
            }else{
656
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
657
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
658
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
659
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
660
            }
661

    
662
            if(for_deblock)
663
                continue;
664

    
665
            h->ref_cache[list][scan8[5 ]+1] = 
666
            h->ref_cache[list][scan8[7 ]+1] = 
667
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewher else)
668
            h->ref_cache[list][scan8[4 ]] = 
669
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
670
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
671
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
672
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
673
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
674
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
675

    
676
            if( h->pps.cabac ) {
677
                /* XXX beurk, Load mvd */
678
                if(IS_INTER(topleft_type)){
679
                    const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
680
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
681
                }else{
682
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
683
                }
684

    
685
                if(IS_INTER(top_type)){
686
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
687
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
688
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
689
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
690
                    *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
691
                }else{
692
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]= 
693
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]= 
694
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]= 
695
                    *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
696
                }
697
                if(IS_INTER(left_type[0])){
698
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
699
                    *(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]];
700
                    *(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]];
701
                }else{
702
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
703
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
704
                }
705
                if(IS_INTER(left_type[1])){
706
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
707
                    *(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]];
708
                    *(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]];
709
                }else{
710
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
711
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
712
                }
713
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
714
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
715
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
716
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
717
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
718

    
719
                if(h->slice_type == B_TYPE){
720
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
721

    
722
                    if(IS_DIRECT(top_type)){
723
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
724
                    }else if(IS_8X8(top_type)){
725
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
726
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
727
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
728
                    }else{
729
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
730
                    }
731
                    
732
                    //FIXME interlacing
733
                    if(IS_DIRECT(left_type[0])){
734
                        h->direct_cache[scan8[0] - 1 + 0*8]=
735
                        h->direct_cache[scan8[0] - 1 + 2*8]= 1;
736
                    }else if(IS_8X8(left_type[0])){
737
                        int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
738
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
739
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
740
                    }else{
741
                        h->direct_cache[scan8[0] - 1 + 0*8]=
742
                        h->direct_cache[scan8[0] - 1 + 2*8]= 0;
743
                    }
744
                }
745
            }
746
        }
747
    }
748
#endif
749
}
750

    
751
static inline void write_back_intra_pred_mode(H264Context *h){
752
    MpegEncContext * const s = &h->s;
753
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
754

    
755
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
756
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
757
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
758
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
759
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
760
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
761
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
762
}
763

    
764
/**
765
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
766
 */
767
static inline int check_intra4x4_pred_mode(H264Context *h){
768
    MpegEncContext * const s = &h->s;
769
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
770
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
771
    int i;
772
    
773
    if(!(h->top_samples_available&0x8000)){
774
        for(i=0; i<4; i++){
775
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
776
            if(status<0){
777
                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);
778
                return -1;
779
            } else if(status){
780
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
781
            }
782
        }
783
    }
784
    
785
    if(!(h->left_samples_available&0x8000)){
786
        for(i=0; i<4; i++){
787
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
788
            if(status<0){
789
                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);
790
                return -1;
791
            } else if(status){
792
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
793
            }
794
        }
795
    }
796

    
797
    return 0;
798
} //FIXME cleanup like next
799

    
800
/**
801
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
802
 */
803
static inline int check_intra_pred_mode(H264Context *h, int mode){
804
    MpegEncContext * const s = &h->s;
805
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
806
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
807
    
808
    if(mode < 0 || mode > 6) {
809
        av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
810
        return -1;
811
    }
812
    
813
    if(!(h->top_samples_available&0x8000)){
814
        mode= top[ mode ];
815
        if(mode<0){
816
            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);
817
            return -1;
818
        }
819
    }
820
    
821
    if(!(h->left_samples_available&0x8000)){
822
        mode= left[ mode ];
823
        if(mode<0){
824
            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);
825
            return -1;
826
        } 
827
    }
828

    
829
    return mode;
830
}
831

    
832
/**
833
 * gets the predicted intra4x4 prediction mode.
834
 */
835
static inline int pred_intra_mode(H264Context *h, int n){
836
    const int index8= scan8[n];
837
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
838
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
839
    const int min= FFMIN(left, top);
840

    
841
    tprintf("mode:%d %d min:%d\n", left ,top, min);
842

    
843
    if(min<0) return DC_PRED;
844
    else      return min;
845
}
846

    
847
static inline void write_back_non_zero_count(H264Context *h){
848
    MpegEncContext * const s = &h->s;
849
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
850

    
851
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
852
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
853
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
854
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
855
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
856
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
857
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
858
    
859
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
860
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
861
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
862

    
863
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
864
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
865
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
866
}
867

    
868
/**
869
 * gets the predicted number of non zero coefficients.
870
 * @param n block index
871
 */
872
static inline int pred_non_zero_count(H264Context *h, int n){
873
    const int index8= scan8[n];
874
    const int left= h->non_zero_count_cache[index8 - 1];
875
    const int top = h->non_zero_count_cache[index8 - 8];
876
    int i= left + top;
877
    
878
    if(i<64) i= (i+1)>>1;
879

    
880
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
881

    
882
    return i&31;
883
}
884

    
885
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
886
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
887

    
888
    if(topright_ref != PART_NOT_AVAILABLE){
889
        *C= h->mv_cache[list][ i - 8 + part_width ];
890
        return topright_ref;
891
    }else{
892
        tprintf("topright MV not available\n");
893

    
894
        *C= h->mv_cache[list][ i - 8 - 1 ];
895
        return h->ref_cache[list][ i - 8 - 1 ];
896
    }
897
}
898

    
899
/**
900
 * gets the predicted MV.
901
 * @param n the block index
902
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
903
 * @param mx the x component of the predicted motion vector
904
 * @param my the y component of the predicted motion vector
905
 */
906
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
907
    const int index8= scan8[n];
908
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
909
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
910
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
911
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
912
    const int16_t * C;
913
    int diagonal_ref, match_count;
914

    
915
    assert(part_width==1 || part_width==2 || part_width==4);
916

    
917
/* mv_cache
918
  B . . A T T T T 
919
  U . . L . . , .
920
  U . . L . . . .
921
  U . . L . . , .
922
  . . . L . . . .
923
*/
924

    
925
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
926
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
927
    tprintf("pred_motion match_count=%d\n", match_count);
928
    if(match_count > 1){ //most common
929
        *mx= mid_pred(A[0], B[0], C[0]);
930
        *my= mid_pred(A[1], B[1], C[1]);
931
    }else if(match_count==1){
932
        if(left_ref==ref){
933
            *mx= A[0];
934
            *my= A[1];        
935
        }else if(top_ref==ref){
936
            *mx= B[0];
937
            *my= B[1];        
938
        }else{
939
            *mx= C[0];
940
            *my= C[1];        
941
        }
942
    }else{
943
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
944
            *mx= A[0];
945
            *my= A[1];        
946
        }else{
947
            *mx= mid_pred(A[0], B[0], C[0]);
948
            *my= mid_pred(A[1], B[1], C[1]);
949
        }
950
    }
951
        
952
    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);
953
}
954

    
955
/**
956
 * gets the directionally predicted 16x8 MV.
957
 * @param n the block index
958
 * @param mx the x component of the predicted motion vector
959
 * @param my the y component of the predicted motion vector
960
 */
961
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
962
    if(n==0){
963
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
964
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
965

    
966
        tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
967
        
968
        if(top_ref == ref){
969
            *mx= B[0];
970
            *my= B[1];
971
            return;
972
        }
973
    }else{
974
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
975
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
976
        
977
        tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
978

    
979
        if(left_ref == ref){
980
            *mx= A[0];
981
            *my= A[1];
982
            return;
983
        }
984
    }
985

    
986
    //RARE
987
    pred_motion(h, n, 4, list, ref, mx, my);
988
}
989

    
990
/**
991
 * gets the directionally predicted 8x16 MV.
992
 * @param n the block index
993
 * @param mx the x component of the predicted motion vector
994
 * @param my the y component of the predicted motion vector
995
 */
996
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
997
    if(n==0){
998
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
999
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
1000
        
1001
        tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
1002

    
1003
        if(left_ref == ref){
1004
            *mx= A[0];
1005
            *my= A[1];
1006
            return;
1007
        }
1008
    }else{
1009
        const int16_t * C;
1010
        int diagonal_ref;
1011

    
1012
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1013
        
1014
        tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
1015

    
1016
        if(diagonal_ref == ref){ 
1017
            *mx= C[0];
1018
            *my= C[1];
1019
            return;
1020
        }
1021
    }
1022

    
1023
    //RARE
1024
    pred_motion(h, n, 2, list, ref, mx, my);
1025
}
1026

    
1027
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1028
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1029
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1030

    
1031
    tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1032

    
1033
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1034
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1035
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1036
       
1037
        *mx = *my = 0;
1038
        return;
1039
    }
1040
        
1041
    pred_motion(h, 0, 4, 0, 0, mx, my);
1042

    
1043
    return;
1044
}
1045

    
1046
static inline void direct_dist_scale_factor(H264Context * const h){
1047
    const int poc = h->s.current_picture_ptr->poc;
1048
    const int poc1 = h->ref_list[1][0].poc;
1049
    int i;
1050
    for(i=0; i<h->ref_count[0]; i++){
1051
        int poc0 = h->ref_list[0][i].poc;
1052
        int td = clip(poc1 - poc0, -128, 127);
1053
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
1054
            h->dist_scale_factor[i] = 256;
1055
        }else{
1056
            int tb = clip(poc - poc0, -128, 127);
1057
            int tx = (16384 + (ABS(td) >> 1)) / td;
1058
            h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1059
        }
1060
    }
1061
}
1062

    
1063
static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1064
    MpegEncContext * const s = &h->s;
1065
    const int mb_xy =   s->mb_x +   s->mb_y*s->mb_stride;
1066
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1067
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1068
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1069
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1070
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1071
    const int is_b8x8 = IS_8X8(*mb_type);
1072
    int sub_mb_type;
1073
    int i8, i4;
1074

    
1075
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1076
        /* FIXME save sub mb types from previous frames (or derive from MVs)
1077
         * so we know exactly what block size to use */
1078
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1079
        *mb_type =    MB_TYPE_8x8;
1080
    }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1081
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1082
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1083
    }else{
1084
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1085
        *mb_type =    MB_TYPE_8x8;
1086
    }
1087
    if(!is_b8x8)
1088
        *mb_type |= MB_TYPE_DIRECT2;
1089

    
1090
    tprintf("mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);
1091
    
1092
    if(h->direct_spatial_mv_pred){
1093
        int ref[2];
1094
        int mv[2][2];
1095
        int list;
1096

    
1097
        /* ref = min(neighbors) */
1098
        for(list=0; list<2; list++){
1099
            int refa = h->ref_cache[list][scan8[0] - 1];
1100
            int refb = h->ref_cache[list][scan8[0] - 8];
1101
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1102
            if(refc == -2)
1103
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
1104
            ref[list] = refa;
1105
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1106
                ref[list] = refb;
1107
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1108
                ref[list] = refc;
1109
            if(ref[list] < 0)
1110
                ref[list] = -1;
1111
        }
1112

    
1113
        if(ref[0] < 0 && ref[1] < 0){
1114
            ref[0] = ref[1] = 0;
1115
            mv[0][0] = mv[0][1] =
1116
            mv[1][0] = mv[1][1] = 0;
1117
        }else{
1118
            for(list=0; list<2; list++){
1119
                if(ref[list] >= 0)
1120
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1121
                else
1122
                    mv[list][0] = mv[list][1] = 0;
1123
            }
1124
        }
1125

    
1126
        if(ref[1] < 0){
1127
            *mb_type &= ~MB_TYPE_P0L1;
1128
            sub_mb_type &= ~MB_TYPE_P0L1;
1129
        }else if(ref[0] < 0){
1130
            *mb_type &= ~MB_TYPE_P0L0;
1131
            sub_mb_type &= ~MB_TYPE_P0L0;
1132
        }
1133

    
1134
        if(IS_16X16(*mb_type)){
1135
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1136
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1137
            if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1138
                ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1139
                if(ref[0] > 0)
1140
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1141
                else
1142
                    fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1143
                if(ref[1] > 0)
1144
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1145
                else
1146
                    fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1147
            }else{
1148
                fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1149
                fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1150
            }
1151
        }else{
1152
            for(i8=0; i8<4; i8++){
1153
                const int x8 = i8&1;
1154
                const int y8 = i8>>1;
1155
    
1156
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1157
                    continue;
1158
                h->sub_mb_type[i8] = sub_mb_type;
1159
    
1160
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1161
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1162
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1163
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1164
    
1165
                /* col_zero_flag */
1166
                if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1167
                    for(i4=0; i4<4; i4++){
1168
                        const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1169
                        if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1170
                            if(ref[0] == 0)
1171
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1172
                            if(ref[1] == 0)
1173
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1174
                        }
1175
                    }
1176
                }
1177
            }
1178
        }
1179
    }else{ /* direct temporal mv pred */
1180
        /* FIXME assumes that L1ref0 used the same ref lists as current frame */
1181
        if(IS_16X16(*mb_type)){
1182
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1183
            if(IS_INTRA(mb_type_col)){
1184
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1185
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1186
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1187
            }else{
1188
                const int ref0 = l1ref0[0];
1189
                const int dist_scale_factor = h->dist_scale_factor[ref0];
1190
                const int16_t *mv_col = l1mv0[0];
1191
                int mv_l0[2];
1192
                mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1193
                mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1194
                fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1195
                fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1196
                fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]), 4);
1197
            }
1198
        }else{
1199
            for(i8=0; i8<4; i8++){
1200
                const int x8 = i8&1;
1201
                const int y8 = i8>>1;
1202
                int ref0, dist_scale_factor;
1203
    
1204
                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1205
                    continue;
1206
                h->sub_mb_type[i8] = sub_mb_type;
1207
                if(IS_INTRA(mb_type_col)){
1208
                    fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1209
                    fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1210
                    fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1211
                    fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1212
                    continue;
1213
                }
1214
    
1215
                ref0 = l1ref0[x8 + y8*h->b8_stride];
1216
                dist_scale_factor = h->dist_scale_factor[ref0];
1217
    
1218
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1219
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1220
                for(i4=0; i4<4; i4++){
1221
                    const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1222
                    int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1223
                    mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1224
                    mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1225
                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1226
                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1227
                }
1228
            }
1229
        }
1230
    }
1231
}
1232

    
1233
static inline void write_back_motion(H264Context *h, int mb_type){
1234
    MpegEncContext * const s = &h->s;
1235
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1236
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1237
    int list;
1238

    
1239
    for(list=0; list<2; list++){
1240
        int y;
1241
        if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
1242
            if(1){ //FIXME skip or never read if mb_type doesnt use it
1243
                for(y=0; y<4; y++){
1244
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1245
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1246
                }
1247
                if( h->pps.cabac ) {
1248
                    /* FIXME needed ? */
1249
                    for(y=0; y<4; y++){
1250
                        *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1251
                        *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1252
                    }
1253
                }
1254
                for(y=0; y<2; y++){
1255
                    *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1256
                }
1257
            }
1258
            continue;
1259
        }
1260
        
1261
        for(y=0; y<4; y++){
1262
            *(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];
1263
            *(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];
1264
        }
1265
        if( h->pps.cabac ) {
1266
            for(y=0; y<4; y++){
1267
                *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1268
                *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1269
            }
1270
        }
1271
        for(y=0; y<2; y++){
1272
            s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1273
            s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1274
        }
1275
    }
1276
    
1277
    if(h->slice_type == B_TYPE && h->pps.cabac){
1278
        if(IS_8X8(mb_type)){
1279
            h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1280
            h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1281
            h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1282
        }
1283
    }
1284
}
1285

    
1286
/**
1287
 * Decodes a network abstraction layer unit.
1288
 * @param consumed is the number of bytes used as input
1289
 * @param length is the length of the array
1290
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1291
 * @returns decoded bytes, might be src+1 if no escapes 
1292
 */
1293
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1294
    int i, si, di;
1295
    uint8_t *dst;
1296

    
1297
//    src[0]&0x80;                //forbidden bit
1298
    h->nal_ref_idc= src[0]>>5;
1299
    h->nal_unit_type= src[0]&0x1F;
1300

    
1301
    src++; length--;
1302
#if 0    
1303
    for(i=0; i<length; i++)
1304
        printf("%2X ", src[i]);
1305
#endif
1306
    for(i=0; i+1<length; i+=2){
1307
        if(src[i]) continue;
1308
        if(i>0 && src[i-1]==0) i--;
1309
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1310
            if(src[i+2]!=3){
1311
                /* startcode, so we must be past the end */
1312
                length=i;
1313
            }
1314
            break;
1315
        }
1316
    }
1317

    
1318
    if(i>=length-1){ //no escaped 0
1319
        *dst_length= length;
1320
        *consumed= length+1; //+1 for the header
1321
        return src; 
1322
    }
1323

    
1324
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1325
    dst= h->rbsp_buffer;
1326

    
1327
//printf("deoding esc\n");
1328
    si=di=0;
1329
    while(si<length){ 
1330
        //remove escapes (very rare 1:2^22)
1331
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1332
            if(src[si+2]==3){ //escape
1333
                dst[di++]= 0;
1334
                dst[di++]= 0;
1335
                si+=3;
1336
                continue;
1337
            }else //next start code
1338
                break;
1339
        }
1340

    
1341
        dst[di++]= src[si++];
1342
    }
1343

    
1344
    *dst_length= di;
1345
    *consumed= si + 1;//+1 for the header
1346
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1347
    return dst;
1348
}
1349

    
1350
#if 0
1351
/**
1352
 * @param src the data which should be escaped
1353
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1354
 * @param length the length of the src data
1355
 * @param dst_length the length of the dst array
1356
 * @returns length of escaped data in bytes or -1 if an error occured
1357
 */
1358
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1359
    int i, escape_count, si, di;
1360
    uint8_t *temp;
1361
    
1362
    assert(length>=0);
1363
    assert(dst_length>0);
1364
    
1365
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1366

1367
    if(length==0) return 1;
1368

1369
    escape_count= 0;
1370
    for(i=0; i<length; i+=2){
1371
        if(src[i]) continue;
1372
        if(i>0 && src[i-1]==0) 
1373
            i--;
1374
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1375
            escape_count++;
1376
            i+=2;
1377
        }
1378
    }
1379
    
1380
    if(escape_count==0){ 
1381
        if(dst+1 != src)
1382
            memcpy(dst+1, src, length);
1383
        return length + 1;
1384
    }
1385
    
1386
    if(length + escape_count + 1> dst_length)
1387
        return -1;
1388

1389
    //this should be damn rare (hopefully)
1390

1391
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1392
    temp= h->rbsp_buffer;
1393
//printf("encoding esc\n");
1394
    
1395
    si= 0;
1396
    di= 0;
1397
    while(si < length){
1398
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1399
            temp[di++]= 0; si++;
1400
            temp[di++]= 0; si++;
1401
            temp[di++]= 3; 
1402
            temp[di++]= src[si++];
1403
        }
1404
        else
1405
            temp[di++]= src[si++];
1406
    }
1407
    memcpy(dst+1, temp, length+escape_count);
1408
    
1409
    assert(di == length+escape_count);
1410
    
1411
    return di + 1;
1412
}
1413

1414
/**
1415
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1416
 */
1417
static void encode_rbsp_trailing(PutBitContext *pb){
1418
    int length;
1419
    put_bits(pb, 1, 1);
1420
    length= (-put_bits_count(pb))&7;
1421
    if(length) put_bits(pb, length, 0);
1422
}
1423
#endif
1424

    
1425
/**
1426
 * identifies the exact end of the bitstream
1427
 * @return the length of the trailing, or 0 if damaged
1428
 */
1429
static int decode_rbsp_trailing(uint8_t *src){
1430
    int v= *src;
1431
    int r;
1432

    
1433
    tprintf("rbsp trailing %X\n", v);
1434

    
1435
    for(r=1; r<9; r++){
1436
        if(v&1) return r;
1437
        v>>=1;
1438
    }
1439
    return 0;
1440
}
1441

    
1442
/**
1443
 * idct tranforms the 16 dc values and dequantize them.
1444
 * @param qp quantization parameter
1445
 */
1446
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1447
    const int qmul= dequant_coeff[qp][0];
1448
#define stride 16
1449
    int i;
1450
    int temp[16]; //FIXME check if this is a good idea
1451
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1452
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1453

    
1454
//memset(block, 64, 2*256);
1455
//return;
1456
    for(i=0; i<4; i++){
1457
        const int offset= y_offset[i];
1458
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1459
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1460
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1461
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1462

    
1463
        temp[4*i+0]= z0+z3;
1464
        temp[4*i+1]= z1+z2;
1465
        temp[4*i+2]= z1-z2;
1466
        temp[4*i+3]= z0-z3;
1467
    }
1468

    
1469
    for(i=0; i<4; i++){
1470
        const int offset= x_offset[i];
1471
        const int z0= temp[4*0+i] + temp[4*2+i];
1472
        const int z1= temp[4*0+i] - temp[4*2+i];
1473
        const int z2= temp[4*1+i] - temp[4*3+i];
1474
        const int z3= temp[4*1+i] + temp[4*3+i];
1475

    
1476
        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1477
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1478
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1479
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1480
    }
1481
}
1482

    
1483
#if 0
1484
/**
1485
 * dct tranforms the 16 dc values.
1486
 * @param qp quantization parameter ??? FIXME
1487
 */
1488
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1489
//    const int qmul= dequant_coeff[qp][0];
1490
    int i;
1491
    int temp[16]; //FIXME check if this is a good idea
1492
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1493
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1494

1495
    for(i=0; i<4; i++){
1496
        const int offset= y_offset[i];
1497
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1498
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1499
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1500
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1501

1502
        temp[4*i+0]= z0+z3;
1503
        temp[4*i+1]= z1+z2;
1504
        temp[4*i+2]= z1-z2;
1505
        temp[4*i+3]= z0-z3;
1506
    }
1507

1508
    for(i=0; i<4; i++){
1509
        const int offset= x_offset[i];
1510
        const int z0= temp[4*0+i] + temp[4*2+i];
1511
        const int z1= temp[4*0+i] - temp[4*2+i];
1512
        const int z2= temp[4*1+i] - temp[4*3+i];
1513
        const int z3= temp[4*1+i] + temp[4*3+i];
1514

1515
        block[stride*0 +offset]= (z0 + z3)>>1;
1516
        block[stride*2 +offset]= (z1 + z2)>>1;
1517
        block[stride*8 +offset]= (z1 - z2)>>1;
1518
        block[stride*10+offset]= (z0 - z3)>>1;
1519
    }
1520
}
1521
#endif
1522

    
1523
#undef xStride
1524
#undef stride
1525

    
1526
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1527
    const int qmul= dequant_coeff[qp][0];
1528
    const int stride= 16*2;
1529
    const int xStride= 16;
1530
    int a,b,c,d,e;
1531

    
1532
    a= block[stride*0 + xStride*0];
1533
    b= block[stride*0 + xStride*1];
1534
    c= block[stride*1 + xStride*0];
1535
    d= block[stride*1 + xStride*1];
1536

    
1537
    e= a-b;
1538
    a= a+b;
1539
    b= c-d;
1540
    c= c+d;
1541

    
1542
    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1543
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1544
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1545
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1546
}
1547

    
1548
#if 0
1549
static void chroma_dc_dct_c(DCTELEM *block){
1550
    const int stride= 16*2;
1551
    const int xStride= 16;
1552
    int a,b,c,d,e;
1553

1554
    a= block[stride*0 + xStride*0];
1555
    b= block[stride*0 + xStride*1];
1556
    c= block[stride*1 + xStride*0];
1557
    d= block[stride*1 + xStride*1];
1558

1559
    e= a-b;
1560
    a= a+b;
1561
    b= c-d;
1562
    c= c+d;
1563

1564
    block[stride*0 + xStride*0]= (a+c);
1565
    block[stride*0 + xStride*1]= (e+b);
1566
    block[stride*1 + xStride*0]= (a-c);
1567
    block[stride*1 + xStride*1]= (e-b);
1568
}
1569
#endif
1570

    
1571
/**
1572
 * gets the chroma qp.
1573
 */
1574
static inline int get_chroma_qp(H264Context *h, int qscale){
1575
    
1576
    return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1577
}
1578

    
1579

    
1580
#if 0
1581
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1582
    int i;
1583
    //FIXME try int temp instead of block
1584
    
1585
    for(i=0; i<4; i++){
1586
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1587
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1588
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1589
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1590
        const int z0= d0 + d3;
1591
        const int z3= d0 - d3;
1592
        const int z1= d1 + d2;
1593
        const int z2= d1 - d2;
1594
        
1595
        block[0 + 4*i]=   z0 +   z1;
1596
        block[1 + 4*i]= 2*z3 +   z2;
1597
        block[2 + 4*i]=   z0 -   z1;
1598
        block[3 + 4*i]=   z3 - 2*z2;
1599
    }    
1600

1601
    for(i=0; i<4; i++){
1602
        const int z0= block[0*4 + i] + block[3*4 + i];
1603
        const int z3= block[0*4 + i] - block[3*4 + i];
1604
        const int z1= block[1*4 + i] + block[2*4 + i];
1605
        const int z2= block[1*4 + i] - block[2*4 + i];
1606
        
1607
        block[0*4 + i]=   z0 +   z1;
1608
        block[1*4 + i]= 2*z3 +   z2;
1609
        block[2*4 + i]=   z0 -   z1;
1610
        block[3*4 + i]=   z3 - 2*z2;
1611
    }
1612
}
1613
#endif
1614

    
1615
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1616
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1617
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1618
    int i;
1619
    const int * const quant_table= quant_coeff[qscale];
1620
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1621
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1622
    const unsigned int threshold2= (threshold1<<1);
1623
    int last_non_zero;
1624

    
1625
    if(seperate_dc){
1626
        if(qscale<=18){
1627
            //avoid overflows
1628
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1629
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1630
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1631

    
1632
            int level= block[0]*quant_coeff[qscale+18][0];
1633
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1634
                if(level>0){
1635
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1636
                    block[0]= level;
1637
                }else{
1638
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1639
                    block[0]= -level;
1640
                }
1641
//                last_non_zero = i;
1642
            }else{
1643
                block[0]=0;
1644
            }
1645
        }else{
1646
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1647
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1648
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1649

    
1650
            int level= block[0]*quant_table[0];
1651
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1652
                if(level>0){
1653
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1654
                    block[0]= level;
1655
                }else{
1656
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1657
                    block[0]= -level;
1658
                }
1659
//                last_non_zero = i;
1660
            }else{
1661
                block[0]=0;
1662
            }
1663
        }
1664
        last_non_zero= 0;
1665
        i=1;
1666
    }else{
1667
        last_non_zero= -1;
1668
        i=0;
1669
    }
1670

    
1671
    for(; i<16; i++){
1672
        const int j= scantable[i];
1673
        int level= block[j]*quant_table[j];
1674

    
1675
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1676
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1677
        if(((unsigned)(level+threshold1))>threshold2){
1678
            if(level>0){
1679
                level= (bias + level)>>QUANT_SHIFT;
1680
                block[j]= level;
1681
            }else{
1682
                level= (bias - level)>>QUANT_SHIFT;
1683
                block[j]= -level;
1684
            }
1685
            last_non_zero = i;
1686
        }else{
1687
            block[j]=0;
1688
        }
1689
    }
1690

    
1691
    return last_non_zero;
1692
}
1693

    
1694
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1695
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1696
    ((uint32_t*)(src+0*stride))[0]= a;
1697
    ((uint32_t*)(src+1*stride))[0]= a;
1698
    ((uint32_t*)(src+2*stride))[0]= a;
1699
    ((uint32_t*)(src+3*stride))[0]= a;
1700
}
1701

    
1702
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1703
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1704
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1705
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1706
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1707
}
1708

    
1709
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1710
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1711
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1712
    
1713
    ((uint32_t*)(src+0*stride))[0]= 
1714
    ((uint32_t*)(src+1*stride))[0]= 
1715
    ((uint32_t*)(src+2*stride))[0]= 
1716
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1717
}
1718

    
1719
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1720
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1721
    
1722
    ((uint32_t*)(src+0*stride))[0]= 
1723
    ((uint32_t*)(src+1*stride))[0]= 
1724
    ((uint32_t*)(src+2*stride))[0]= 
1725
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1726
}
1727

    
1728
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1729
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1730
    
1731
    ((uint32_t*)(src+0*stride))[0]= 
1732
    ((uint32_t*)(src+1*stride))[0]= 
1733
    ((uint32_t*)(src+2*stride))[0]= 
1734
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1735
}
1736

    
1737
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1738
    ((uint32_t*)(src+0*stride))[0]= 
1739
    ((uint32_t*)(src+1*stride))[0]= 
1740
    ((uint32_t*)(src+2*stride))[0]= 
1741
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1742
}
1743

    
1744

    
1745
#define LOAD_TOP_RIGHT_EDGE\
1746
    const int t4= topright[0];\
1747
    const int t5= topright[1];\
1748
    const int t6= topright[2];\
1749
    const int t7= topright[3];\
1750

    
1751
#define LOAD_LEFT_EDGE\
1752
    const int l0= src[-1+0*stride];\
1753
    const int l1= src[-1+1*stride];\
1754
    const int l2= src[-1+2*stride];\
1755
    const int l3= src[-1+3*stride];\
1756

    
1757
#define LOAD_TOP_EDGE\
1758
    const int t0= src[ 0-1*stride];\
1759
    const int t1= src[ 1-1*stride];\
1760
    const int t2= src[ 2-1*stride];\
1761
    const int t3= src[ 3-1*stride];\
1762

    
1763
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1764
    const int lt= src[-1-1*stride];
1765
    LOAD_TOP_EDGE
1766
    LOAD_LEFT_EDGE
1767

    
1768
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1769
    src[0+2*stride]=
1770
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1771
    src[0+1*stride]=
1772
    src[1+2*stride]=
1773
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1774
    src[0+0*stride]=
1775
    src[1+1*stride]=
1776
    src[2+2*stride]=
1777
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1778
    src[1+0*stride]=
1779
    src[2+1*stride]=
1780
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1781
    src[2+0*stride]=
1782
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1783
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1784
}
1785

    
1786
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1787
    LOAD_TOP_EDGE    
1788
    LOAD_TOP_RIGHT_EDGE    
1789
//    LOAD_LEFT_EDGE    
1790

    
1791
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1792
    src[1+0*stride]=
1793
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1794
    src[2+0*stride]=
1795
    src[1+1*stride]=
1796
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1797
    src[3+0*stride]=
1798
    src[2+1*stride]=
1799
    src[1+2*stride]=
1800
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1801
    src[3+1*stride]=
1802
    src[2+2*stride]=
1803
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1804
    src[3+2*stride]=
1805
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1806
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1807
}
1808

    
1809
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1810
    const int lt= src[-1-1*stride];
1811
    LOAD_TOP_EDGE    
1812
    LOAD_LEFT_EDGE    
1813
    const __attribute__((unused)) int unu= l3;
1814

    
1815
    src[0+0*stride]=
1816
    src[1+2*stride]=(lt + t0 + 1)>>1;
1817
    src[1+0*stride]=
1818
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1819
    src[2+0*stride]=
1820
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1821
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1822
    src[0+1*stride]=
1823
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1824
    src[1+1*stride]=
1825
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1826
    src[2+1*stride]=
1827
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1828
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1829
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1830
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1831
}
1832

    
1833
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1834
    LOAD_TOP_EDGE    
1835
    LOAD_TOP_RIGHT_EDGE    
1836
    const __attribute__((unused)) int unu= t7;
1837

    
1838
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1839
    src[1+0*stride]=
1840
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1841
    src[2+0*stride]=
1842
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1843
    src[3+0*stride]=
1844
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1845
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1846
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1847
    src[1+1*stride]=
1848
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1849
    src[2+1*stride]=
1850
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1851
    src[3+1*stride]=
1852
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1853
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1854
}
1855

    
1856
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1857
    LOAD_LEFT_EDGE    
1858

    
1859
    src[0+0*stride]=(l0 + l1 + 1)>>1;
1860
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1861
    src[2+0*stride]=
1862
    src[0+1*stride]=(l1 + l2 + 1)>>1;
1863
    src[3+0*stride]=
1864
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1865
    src[2+1*stride]=
1866
    src[0+2*stride]=(l2 + l3 + 1)>>1;
1867
    src[3+1*stride]=
1868
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1869
    src[3+2*stride]=
1870
    src[1+3*stride]=
1871
    src[0+3*stride]=
1872
    src[2+2*stride]=
1873
    src[2+3*stride]=
1874
    src[3+3*stride]=l3;
1875
}
1876
    
1877
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1878
    const int lt= src[-1-1*stride];
1879
    LOAD_TOP_EDGE    
1880
    LOAD_LEFT_EDGE    
1881
    const __attribute__((unused)) int unu= t3;
1882

    
1883
    src[0+0*stride]=
1884
    src[2+1*stride]=(lt + l0 + 1)>>1;
1885
    src[1+0*stride]=
1886
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1887
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1888
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1889
    src[0+1*stride]=
1890
    src[2+2*stride]=(l0 + l1 + 1)>>1;
1891
    src[1+1*stride]=
1892
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1893
    src[0+2*stride]=
1894
    src[2+3*stride]=(l1 + l2+ 1)>>1;
1895
    src[1+2*stride]=
1896
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1897
    src[0+3*stride]=(l2 + l3 + 1)>>1;
1898
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1899
}
1900

    
1901
static void pred16x16_vertical_c(uint8_t *src, int stride){
1902
    int i;
1903
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1904
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1905
    const uint32_t c= ((uint32_t*)(src-stride))[2];
1906
    const uint32_t d= ((uint32_t*)(src-stride))[3];
1907
    
1908
    for(i=0; i<16; i++){
1909
        ((uint32_t*)(src+i*stride))[0]= a;
1910
        ((uint32_t*)(src+i*stride))[1]= b;
1911
        ((uint32_t*)(src+i*stride))[2]= c;
1912
        ((uint32_t*)(src+i*stride))[3]= d;
1913
    }
1914
}
1915

    
1916
static void pred16x16_horizontal_c(uint8_t *src, int stride){
1917
    int i;
1918

    
1919
    for(i=0; i<16; i++){
1920
        ((uint32_t*)(src+i*stride))[0]=
1921
        ((uint32_t*)(src+i*stride))[1]=
1922
        ((uint32_t*)(src+i*stride))[2]=
1923
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1924
    }
1925
}
1926

    
1927
static void pred16x16_dc_c(uint8_t *src, int stride){
1928
    int i, dc=0;
1929

    
1930
    for(i=0;i<16; i++){
1931
        dc+= src[-1+i*stride];
1932
    }
1933
    
1934
    for(i=0;i<16; i++){
1935
        dc+= src[i-stride];
1936
    }
1937

    
1938
    dc= 0x01010101*((dc + 16)>>5);
1939

    
1940
    for(i=0; i<16; i++){
1941
        ((uint32_t*)(src+i*stride))[0]=
1942
        ((uint32_t*)(src+i*stride))[1]=
1943
        ((uint32_t*)(src+i*stride))[2]=
1944
        ((uint32_t*)(src+i*stride))[3]= dc;
1945
    }
1946
}
1947

    
1948
static void pred16x16_left_dc_c(uint8_t *src, int stride){
1949
    int i, dc=0;
1950

    
1951
    for(i=0;i<16; i++){
1952
        dc+= src[-1+i*stride];
1953
    }
1954
    
1955
    dc= 0x01010101*((dc + 8)>>4);
1956

    
1957
    for(i=0; i<16; i++){
1958
        ((uint32_t*)(src+i*stride))[0]=
1959
        ((uint32_t*)(src+i*stride))[1]=
1960
        ((uint32_t*)(src+i*stride))[2]=
1961
        ((uint32_t*)(src+i*stride))[3]= dc;
1962
    }
1963
}
1964

    
1965
static void pred16x16_top_dc_c(uint8_t *src, int stride){
1966
    int i, dc=0;
1967

    
1968
    for(i=0;i<16; i++){
1969
        dc+= src[i-stride];
1970
    }
1971
    dc= 0x01010101*((dc + 8)>>4);
1972

    
1973
    for(i=0; i<16; i++){
1974
        ((uint32_t*)(src+i*stride))[0]=
1975
        ((uint32_t*)(src+i*stride))[1]=
1976
        ((uint32_t*)(src+i*stride))[2]=
1977
        ((uint32_t*)(src+i*stride))[3]= dc;
1978
    }
1979
}
1980

    
1981
static void pred16x16_128_dc_c(uint8_t *src, int stride){
1982
    int i;
1983

    
1984
    for(i=0; i<16; i++){
1985
        ((uint32_t*)(src+i*stride))[0]=
1986
        ((uint32_t*)(src+i*stride))[1]=
1987
        ((uint32_t*)(src+i*stride))[2]=
1988
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1989
    }
1990
}
1991

    
1992
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1993
  int i, j, k;
1994
  int a;
1995
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
1996
  const uint8_t * const src0 = src+7-stride;
1997
  const uint8_t *src1 = src+8*stride-1;
1998
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
1999
  int H = src0[1] - src0[-1];
2000
  int V = src1[0] - src2[ 0];
2001
  for(k=2; k<=8; ++k) {
2002
    src1 += stride; src2 -= stride;
2003
    H += k*(src0[k] - src0[-k]);
2004
    V += k*(src1[0] - src2[ 0]);
2005
  }
2006
  if(svq3){
2007
    H = ( 5*(H/4) ) / 16;
2008
    V = ( 5*(V/4) ) / 16;
2009

    
2010
    /* required for 100% accuracy */
2011
    i = H; H = V; V = i;
2012
  }else{
2013
    H = ( 5*H+32 ) >> 6;
2014
    V = ( 5*V+32 ) >> 6;
2015
  }
2016

    
2017
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2018
  for(j=16; j>0; --j) {
2019
    int b = a;
2020
    a += V;
2021
    for(i=-16; i<0; i+=4) {
2022
      src[16+i] = cm[ (b    ) >> 5 ];
2023
      src[17+i] = cm[ (b+  H) >> 5 ];
2024
      src[18+i] = cm[ (b+2*H) >> 5 ];
2025
      src[19+i] = cm[ (b+3*H) >> 5 ];
2026
      b += 4*H;
2027
    }
2028
    src += stride;
2029
  }
2030
}
2031

    
2032
static void pred16x16_plane_c(uint8_t *src, int stride){
2033
    pred16x16_plane_compat_c(src, stride, 0);
2034
}
2035

    
2036
static void pred8x8_vertical_c(uint8_t *src, int stride){
2037
    int i;
2038
    const uint32_t a= ((uint32_t*)(src-stride))[0];
2039
    const uint32_t b= ((uint32_t*)(src-stride))[1];
2040
    
2041
    for(i=0; i<8; i++){
2042
        ((uint32_t*)(src+i*stride))[0]= a;
2043
        ((uint32_t*)(src+i*stride))[1]= b;
2044
    }
2045
}
2046

    
2047
static void pred8x8_horizontal_c(uint8_t *src, int stride){
2048
    int i;
2049

    
2050
    for(i=0; i<8; i++){
2051
        ((uint32_t*)(src+i*stride))[0]=
2052
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2053
    }
2054
}
2055

    
2056
static void pred8x8_128_dc_c(uint8_t *src, int stride){
2057
    int i;
2058

    
2059
    for(i=0; i<4; i++){
2060
        ((uint32_t*)(src+i*stride))[0]= 
2061
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2062
    }
2063
    for(i=4; i<8; i++){
2064
        ((uint32_t*)(src+i*stride))[0]= 
2065
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2066
    }
2067
}
2068

    
2069
static void pred8x8_left_dc_c(uint8_t *src, int stride){
2070
    int i;
2071
    int dc0, dc2;
2072

    
2073
    dc0=dc2=0;
2074
    for(i=0;i<4; i++){
2075
        dc0+= src[-1+i*stride];
2076
        dc2+= src[-1+(i+4)*stride];
2077
    }
2078
    dc0= 0x01010101*((dc0 + 2)>>2);
2079
    dc2= 0x01010101*((dc2 + 2)>>2);
2080

    
2081
    for(i=0; i<4; i++){
2082
        ((uint32_t*)(src+i*stride))[0]=
2083
        ((uint32_t*)(src+i*stride))[1]= dc0;
2084
    }
2085
    for(i=4; i<8; i++){
2086
        ((uint32_t*)(src+i*stride))[0]=
2087
        ((uint32_t*)(src+i*stride))[1]= dc2;
2088
    }
2089
}
2090

    
2091
static void pred8x8_top_dc_c(uint8_t *src, int stride){
2092
    int i;
2093
    int dc0, dc1;
2094

    
2095
    dc0=dc1=0;
2096
    for(i=0;i<4; i++){
2097
        dc0+= src[i-stride];
2098
        dc1+= src[4+i-stride];
2099
    }
2100
    dc0= 0x01010101*((dc0 + 2)>>2);
2101
    dc1= 0x01010101*((dc1 + 2)>>2);
2102

    
2103
    for(i=0; i<4; i++){
2104
        ((uint32_t*)(src+i*stride))[0]= dc0;
2105
        ((uint32_t*)(src+i*stride))[1]= dc1;
2106
    }
2107
    for(i=4; i<8; i++){
2108
        ((uint32_t*)(src+i*stride))[0]= dc0;
2109
        ((uint32_t*)(src+i*stride))[1]= dc1;
2110
    }
2111
}
2112

    
2113

    
2114
static void pred8x8_dc_c(uint8_t *src, int stride){
2115
    int i;
2116
    int dc0, dc1, dc2, dc3;
2117

    
2118
    dc0=dc1=dc2=0;
2119
    for(i=0;i<4; i++){
2120
        dc0+= src[-1+i*stride] + src[i-stride];
2121
        dc1+= src[4+i-stride];
2122
        dc2+= src[-1+(i+4)*stride];
2123
    }
2124
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2125
    dc0= 0x01010101*((dc0 + 4)>>3);
2126
    dc1= 0x01010101*((dc1 + 2)>>2);
2127
    dc2= 0x01010101*((dc2 + 2)>>2);
2128

    
2129
    for(i=0; i<4; i++){
2130
        ((uint32_t*)(src+i*stride))[0]= dc0;
2131
        ((uint32_t*)(src+i*stride))[1]= dc1;
2132
    }
2133
    for(i=4; i<8; i++){
2134
        ((uint32_t*)(src+i*stride))[0]= dc2;
2135
        ((uint32_t*)(src+i*stride))[1]= dc3;
2136
    }
2137
}
2138

    
2139
static void pred8x8_plane_c(uint8_t *src, int stride){
2140
  int j, k;
2141
  int a;
2142
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
2143
  const uint8_t * const src0 = src+3-stride;
2144
  const uint8_t *src1 = src+4*stride-1;
2145
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
2146
  int H = src0[1] - src0[-1];
2147
  int V = src1[0] - src2[ 0];
2148
  for(k=2; k<=4; ++k) {
2149
    src1 += stride; src2 -= stride;
2150
    H += k*(src0[k] - src0[-k]);
2151
    V += k*(src1[0] - src2[ 0]);
2152
  }
2153
  H = ( 17*H+16 ) >> 5;
2154
  V = ( 17*V+16 ) >> 5;
2155

    
2156
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2157
  for(j=8; j>0; --j) {
2158
    int b = a;
2159
    a += V;
2160
    src[0] = cm[ (b    ) >> 5 ];
2161
    src[1] = cm[ (b+  H) >> 5 ];
2162
    src[2] = cm[ (b+2*H) >> 5 ];
2163
    src[3] = cm[ (b+3*H) >> 5 ];
2164
    src[4] = cm[ (b+4*H) >> 5 ];
2165
    src[5] = cm[ (b+5*H) >> 5 ];
2166
    src[6] = cm[ (b+6*H) >> 5 ];
2167
    src[7] = cm[ (b+7*H) >> 5 ];
2168
    src += stride;
2169
  }
2170
}
2171

    
2172
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2173
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2174
                           int src_x_offset, int src_y_offset,
2175
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2176
    MpegEncContext * const s = &h->s;
2177
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2178
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2179
    const int luma_xy= (mx&3) + ((my&3)<<2);
2180
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2181
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2182
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2183
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2184
    int extra_height= extra_width;
2185
    int emu=0;
2186
    const int full_mx= mx>>2;
2187
    const int full_my= my>>2;
2188
    
2189
    assert(pic->data[0]);
2190
    
2191
    if(mx&7) extra_width -= 3;
2192
    if(my&7) extra_height -= 3;
2193
    
2194
    if(   full_mx < 0-extra_width 
2195
       || full_my < 0-extra_height 
2196
       || full_mx + 16/*FIXME*/ > s->width + extra_width 
2197
       || full_my + 16/*FIXME*/ > s->height + extra_height){
2198
        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);
2199
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2200
        emu=1;
2201
    }
2202
    
2203
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2204
    if(!square){
2205
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2206
    }
2207
    
2208
    if(s->flags&CODEC_FLAG_GRAY) return;
2209
    
2210
    if(emu){
2211
        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);
2212
            src_cb= s->edge_emu_buffer;
2213
    }
2214
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2215

    
2216
    if(emu){
2217
        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);
2218
            src_cr= s->edge_emu_buffer;
2219
    }
2220
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2221
}
2222

    
2223
static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2224
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2225
                           int x_offset, int y_offset,
2226
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2227
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2228
                           int list0, int list1){
2229
    MpegEncContext * const s = &h->s;
2230
    qpel_mc_func *qpix_op=  qpix_put;
2231
    h264_chroma_mc_func chroma_op= chroma_put;
2232
    
2233
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2234
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2235
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2236
    x_offset += 8*s->mb_x;
2237
    y_offset += 8*s->mb_y;
2238
    
2239
    if(list0){
2240
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2241
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2242
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2243
                           qpix_op, chroma_op);
2244

    
2245
        qpix_op=  qpix_avg;
2246
        chroma_op= chroma_avg;
2247
    }
2248

    
2249
    if(list1){
2250
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2251
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2252
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
2253
                           qpix_op, chroma_op);
2254
    }
2255
}
2256

    
2257
static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2258
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2259
                           int x_offset, int y_offset,
2260
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2261
                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2262
                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2263
                           int list0, int list1){
2264
    MpegEncContext * const s = &h->s;
2265

    
2266
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
2267
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
2268
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
2269
    x_offset += 8*s->mb_x;
2270
    y_offset += 8*s->mb_y;
2271
    
2272
    if(list0 && list1){
2273
        /* don't optimize for luma-only case, since B-frames usually
2274
         * use implicit weights => chroma too. */
2275
        uint8_t *tmp_cb = s->obmc_scratchpad;
2276
        uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2277
        uint8_t *tmp_y  = tmp_cr + 8*s->uvlinesize;
2278
        int refn0 = h->ref_cache[0][ scan8[n] ];
2279
        int refn1 = h->ref_cache[1][ scan8[n] ];
2280

    
2281
        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2282
                    dest_y, dest_cb, dest_cr,
2283
                    x_offset, y_offset, qpix_put, chroma_put);
2284
        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2285
                    tmp_y, tmp_cb, tmp_cr,
2286
                    x_offset, y_offset, qpix_put, chroma_put);
2287

    
2288
        if(h->use_weight == 2){
2289
            int weight0 = h->implicit_weight[refn0][refn1];
2290
            int weight1 = 64 - weight0;
2291
            luma_weight_avg(  dest_y,  tmp_y,  s->  linesize, 5, weight0, weight1, 0, 0);
2292
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2293
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2294
        }else{
2295
            luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2296
                            h->luma_weight[0][refn0], h->luma_weight[1][refn1], 
2297
                            h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2298
            chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2299
                            h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0], 
2300
                            h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2301
            chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2302
                            h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1], 
2303
                            h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2304
        }
2305
    }else{
2306
        int list = list1 ? 1 : 0;
2307
        int refn = h->ref_cache[list][ scan8[n] ];
2308
        Picture *ref= &h->ref_list[list][refn];
2309
        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2310
                    dest_y, dest_cb, dest_cr, x_offset, y_offset,
2311
                    qpix_put, chroma_put);
2312

    
2313
        luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2314
                       h->luma_weight[list][refn], h->luma_offset[list][refn]);
2315
        if(h->use_weight_chroma){
2316
            chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2317
                             h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2318
            chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2319
                             h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2320
        }
2321
    }
2322
}
2323

    
2324
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2325
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2326
                           int x_offset, int y_offset,
2327
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2328
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2329
                           h264_weight_func *weight_op, h264_biweight_func *weight_avg, 
2330
                           int list0, int list1){
2331
    if((h->use_weight==2 && list0 && list1
2332
        && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2333
       || h->use_weight==1)
2334
        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2335
                         x_offset, y_offset, qpix_put, chroma_put,
2336
                         weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2337
    else
2338
        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2339
                    x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2340
}
2341

    
2342
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2343
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2344
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2345
                      h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2346
    MpegEncContext * const s = &h->s;
2347
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2348
    const int mb_type= s->current_picture.mb_type[mb_xy];
2349
    
2350
    assert(IS_INTER(mb_type));
2351
    
2352
    if(IS_16X16(mb_type)){
2353
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2354
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2355
                &weight_op[0], &weight_avg[0],
2356
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2357
    }else if(IS_16X8(mb_type)){
2358
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2359
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2360
                &weight_op[1], &weight_avg[1],
2361
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2362
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2363
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2364
                &weight_op[1], &weight_avg[1],
2365
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2366
    }else if(IS_8X16(mb_type)){
2367
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2368
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2369
                &weight_op[2], &weight_avg[2],
2370
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2371
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2372
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2373
                &weight_op[2], &weight_avg[2],
2374
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2375
    }else{
2376
        int i;
2377
        
2378
        assert(IS_8X8(mb_type));
2379

    
2380
        for(i=0; i<4; i++){
2381
            const int sub_mb_type= h->sub_mb_type[i];
2382
            const int n= 4*i;
2383
            int x_offset= (i&1)<<2;
2384
            int y_offset= (i&2)<<1;
2385

    
2386
            if(IS_SUB_8X8(sub_mb_type)){
2387
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2388
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2389
                    &weight_op[3], &weight_avg[3],
2390
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2391
            }else if(IS_SUB_8X4(sub_mb_type)){
2392
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2393
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2394
                    &weight_op[4], &weight_avg[4],
2395
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2396
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2397
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2398
                    &weight_op[4], &weight_avg[4],
2399
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2400
            }else if(IS_SUB_4X8(sub_mb_type)){
2401
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2402
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2403
                    &weight_op[5], &weight_avg[5],
2404
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2405
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2406
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2407
                    &weight_op[5], &weight_avg[5],
2408
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2409
            }else{
2410
                int j;
2411
                assert(IS_SUB_4X4(sub_mb_type));
2412
                for(j=0; j<4; j++){
2413
                    int sub_x_offset= x_offset + 2*(j&1);
2414
                    int sub_y_offset= y_offset +   (j&2);
2415
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2416
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2417
                        &weight_op[6], &weight_avg[6],
2418
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2419
                }
2420
            }
2421
        }
2422
    }
2423
}
2424

    
2425
static void decode_init_vlc(H264Context *h){
2426
    static int done = 0;
2427

    
2428
    if (!done) {
2429
        int i;
2430
        done = 1;
2431

    
2432
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2433
                 &chroma_dc_coeff_token_len [0], 1, 1,
2434
                 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2435

    
2436
        for(i=0; i<4; i++){
2437
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2438
                     &coeff_token_len [i][0], 1, 1,
2439
                     &coeff_token_bits[i][0], 1, 1, 1);
2440
        }
2441

    
2442
        for(i=0; i<3; i++){
2443
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2444
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2445
                     &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2446
        }
2447
        for(i=0; i<15; i++){
2448
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2449
                     &total_zeros_len [i][0], 1, 1,
2450
                     &total_zeros_bits[i][0], 1, 1, 1);
2451
        }
2452

    
2453
        for(i=0; i<6; i++){
2454
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2455
                     &run_len [i][0], 1, 1,
2456
                     &run_bits[i][0], 1, 1, 1);
2457
        }
2458
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2459
                 &run_len [6][0], 1, 1,
2460
                 &run_bits[6][0], 1, 1, 1);
2461
    }
2462
}
2463

    
2464
/**
2465
 * Sets the intra prediction function pointers.
2466
 */
2467
static void init_pred_ptrs(H264Context *h){
2468
//    MpegEncContext * const s = &h->s;
2469

    
2470
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2471
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2472
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2473
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2474
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2475
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2476
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2477
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2478
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2479
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2480
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2481
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2482

    
2483
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2484
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2485
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2486
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2487
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2488
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2489
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2490

    
2491
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2492
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2493
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2494
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2495
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2496
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2497
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2498
}
2499

    
2500
static void free_tables(H264Context *h){
2501
    av_freep(&h->intra4x4_pred_mode);
2502
    av_freep(&h->chroma_pred_mode_table);
2503
    av_freep(&h->cbp_table);
2504
    av_freep(&h->mvd_table[0]);
2505
    av_freep(&h->mvd_table[1]);
2506
    av_freep(&h->direct_table);
2507
    av_freep(&h->non_zero_count);
2508
    av_freep(&h->slice_table_base);
2509
    av_freep(&h->top_border);
2510
    h->slice_table= NULL;
2511

    
2512
    av_freep(&h->mb2b_xy);
2513
    av_freep(&h->mb2b8_xy);
2514

    
2515
    av_freep(&h->s.obmc_scratchpad);
2516
}
2517

    
2518
/**
2519
 * allocates tables.
2520
 * needs widzh/height
2521
 */
2522
static int alloc_tables(H264Context *h){
2523
    MpegEncContext * const s = &h->s;
2524
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2525
    int x,y;
2526

    
2527
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2528

    
2529
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2530
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2531
    CHECKED_ALLOCZ(h->top_border       , s->mb_width * (16+8+8) * sizeof(uint8_t))
2532
    CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2533

    
2534
    if( h->pps.cabac ) {
2535
        CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2536
        CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2537
        CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2538
        CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2539
    }
2540

    
2541
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2542
    h->slice_table= h->slice_table_base + s->mb_stride + 1;
2543

    
2544
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint16_t));
2545
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2546
    for(y=0; y<s->mb_height; y++){
2547
        for(x=0; x<s->mb_width; x++){
2548
            const int mb_xy= x + y*s->mb_stride;
2549
            const int b_xy = 4*x + 4*y*h->b_stride;
2550
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2551
        
2552
            h->mb2b_xy [mb_xy]= b_xy;
2553
            h->mb2b8_xy[mb_xy]= b8_xy;
2554
        }
2555
    }
2556

    
2557
    s->obmc_scratchpad = NULL;
2558

    
2559
    return 0;
2560
fail:
2561
    free_tables(h);
2562
    return -1;
2563
}
2564

    
2565
static void common_init(H264Context *h){
2566
    MpegEncContext * const s = &h->s;
2567

    
2568
    s->width = s->avctx->width;
2569
    s->height = s->avctx->height;
2570
    s->codec_id= s->avctx->codec->id;
2571
    
2572
    init_pred_ptrs(h);
2573

    
2574
    s->unrestricted_mv=1;
2575
    s->decode=1; //FIXME
2576
}
2577

    
2578
static int decode_init(AVCodecContext *avctx){
2579
    H264Context *h= avctx->priv_data;
2580
    MpegEncContext * const s = &h->s;
2581

    
2582
    MPV_decode_defaults(s);
2583
    
2584
    s->avctx = avctx;
2585
    common_init(h);
2586

    
2587
    s->out_format = FMT_H264;
2588
    s->workaround_bugs= avctx->workaround_bugs;
2589

    
2590
    // set defaults
2591
//    s->decode_mb= ff_h263_decode_mb;
2592
    s->low_delay= 1;
2593
    avctx->pix_fmt= PIX_FMT_YUV420P;
2594

    
2595
    decode_init_vlc(h);
2596
    
2597
    if(avctx->codec_tag != 0x31637661 && avctx->codec_tag != 0x31435641) // avc1
2598
        h->is_avc = 0;
2599
    else {
2600
        if((avctx->extradata_size == 0) || (avctx->extradata == NULL)) {
2601
            av_log(avctx, AV_LOG_ERROR, "AVC codec requires avcC data\n");
2602
            return -1;
2603
        }
2604
        h->is_avc = 1;
2605
        h->got_avcC = 0;
2606
    }
2607

    
2608
    return 0;
2609
}
2610

    
2611
static void frame_start(H264Context *h){
2612
    MpegEncContext * const s = &h->s;
2613
    int i;
2614

    
2615
    MPV_frame_start(s, s->avctx);
2616
    ff_er_frame_start(s);
2617

    
2618
    assert(s->linesize && s->uvlinesize);
2619

    
2620
    for(i=0; i<16; i++){
2621
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2622
        h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2623
    }
2624
    for(i=0; i<4; i++){
2625
        h->block_offset[16+i]=
2626
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2627
    }
2628

    
2629
    /* can't be in alloc_tables because linesize isn't known there.
2630
     * FIXME: redo bipred weight to not require extra buffer? */
2631
    if(!s->obmc_scratchpad)
2632
        s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
2633

    
2634
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2635
}
2636

    
2637
static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2638
    MpegEncContext * const s = &h->s;
2639
    int i;
2640
    
2641
    src_y  -=   linesize;
2642
    src_cb -= uvlinesize;
2643
    src_cr -= uvlinesize;
2644

    
2645
    h->left_border[0]= h->top_border[s->mb_x][15];
2646
    for(i=1; i<17; i++){
2647
        h->left_border[i]= src_y[15+i*  linesize];
2648
    }
2649
    
2650
    *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y +  16*linesize);
2651
    *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2652

    
2653
    if(!(s->flags&CODEC_FLAG_GRAY)){
2654
        h->left_border[17  ]= h->top_border[s->mb_x][16+7];
2655
        h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2656
        for(i=1; i<9; i++){
2657
            h->left_border[i+17  ]= src_cb[7+i*uvlinesize];
2658
            h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2659
        }
2660
        *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2661
        *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2662
    }
2663
}
2664

    
2665
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){
2666
    MpegEncContext * const s = &h->s;
2667
    int temp8, i;
2668
    uint64_t temp64;
2669
    int deblock_left = (s->mb_x > 0);
2670
    int deblock_top  = (s->mb_y > 0);
2671

    
2672
    src_y  -=   linesize + 1;
2673
    src_cb -= uvlinesize + 1;
2674
    src_cr -= uvlinesize + 1;
2675

    
2676
#define XCHG(a,b,t,xchg)\
2677
t= a;\
2678
if(xchg)\
2679
    a= b;\
2680
b= t;
2681

    
2682
    if(deblock_left){
2683
        for(i = !deblock_top; i<17; i++){
2684
            XCHG(h->left_border[i     ], src_y [i*  linesize], temp8, xchg);
2685
        }
2686
    }
2687

    
2688
    if(deblock_top){
2689
        XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2690
        XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2691
    }
2692

    
2693
    if(!(s->flags&CODEC_FLAG_GRAY)){
2694
        if(deblock_left){
2695
            for(i = !deblock_top; i<9; i++){
2696
                XCHG(h->left_border[i+17  ], src_cb[i*uvlinesize], temp8, xchg);
2697
                XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2698
            }
2699
        }
2700
        if(deblock_top){
2701
            XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2702
            XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2703
        }
2704
    }
2705
}
2706

    
2707
static void hl_decode_mb(H264Context *h){
2708
    MpegEncContext * const s = &h->s;
2709
    const int mb_x= s->mb_x;
2710
    const int mb_y= s->mb_y;
2711
    const int mb_xy= mb_x + mb_y*s->mb_stride;
2712
    const int mb_type= s->current_picture.mb_type[mb_xy];
2713
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2714
    int linesize, uvlinesize /*dct_offset*/;
2715
    int i;
2716

    
2717
    if(!s->decode)
2718
        return;
2719

    
2720
    if(s->mb_skiped){
2721
    }
2722

    
2723
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2724
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2725
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2726

    
2727
    if (h->mb_field_decoding_flag) {
2728
        linesize = s->linesize * 2;
2729
        uvlinesize = s->uvlinesize * 2;
2730
        if(mb_y&1){ //FIXME move out of this func?
2731
            dest_y -= s->linesize*15;
2732
            dest_cb-= s->linesize*7;
2733
            dest_cr-= s->linesize*7;
2734
        }
2735
    } else {
2736
        linesize = s->linesize;
2737
        uvlinesize = s->uvlinesize;
2738
//        dct_offset = s->linesize * 16;
2739
    }
2740

    
2741
    if(IS_INTRA(mb_type)){
2742
        if(h->deblocking_filter)
2743
            xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2744

    
2745
        if(!(s->flags&CODEC_FLAG_GRAY)){
2746
            h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2747
            h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2748
        }
2749

    
2750
        if(IS_INTRA4x4(mb_type)){
2751
            if(!s->encoding){
2752
                for(i=0; i<16; i++){
2753
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2754
                    uint8_t *topright;
2755
                    const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2756
                    int tr;
2757

    
2758
                    if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2759
                        const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2760
                        assert(mb_y || linesize <= h->block_offset[i]);
2761
                        if(!topright_avail){
2762
                            tr= ptr[3 - linesize]*0x01010101;
2763
                            topright= (uint8_t*) &tr;
2764
                        }else if(i==5 && h->deblocking_filter){
2765
                            tr= *(uint32_t*)h->top_border[mb_x+1];
2766
                            topright= (uint8_t*) &tr;
2767
                        }else
2768
                            topright= ptr + 4 - linesize;
2769
                    }else
2770
                        topright= NULL;
2771

    
2772
                    h->pred4x4[ dir ](ptr, topright, linesize);
2773
                    if(h->non_zero_count_cache[ scan8[i] ]){
2774
                        if(s->codec_id == CODEC_ID_H264)
2775
                            s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2776
                        else
2777
                            svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2778
                    }
2779
                }
2780
            }
2781
        }else{
2782
            h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2783
            if(s->codec_id == CODEC_ID_H264)
2784
                h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2785
            else
2786
                svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2787
        }
2788
        if(h->deblocking_filter)
2789
            xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2790
    }else if(s->codec_id == CODEC_ID_H264){
2791
        hl_motion(h, dest_y, dest_cb, dest_cr,
2792
                  s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
2793
                  s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
2794
                  s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2795
    }
2796

    
2797

    
2798
    if(!IS_INTRA4x4(mb_type)){
2799
        if(s->codec_id == CODEC_ID_H264){
2800
            for(i=0; i<16; i++){
2801
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2802
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2803
                    s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2804
                }
2805
            }
2806
        }else{
2807
            for(i=0; i<16; i++){
2808
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2809
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2810
                    svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2811
                }
2812
            }
2813
        }
2814
    }
2815

    
2816
    if(!(s->flags&CODEC_FLAG_GRAY)){
2817
        chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2818
        chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2819
        if(s->codec_id == CODEC_ID_H264){
2820
            for(i=16; i<16+4; i++){
2821
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2822
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2823
                    s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2824
                }
2825
            }
2826
            for(i=20; i<20+4; i++){
2827
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2828
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2829
                    s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2830
                }
2831
            }
2832
        }else{
2833
            for(i=16; i<16+4; i++){
2834
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2835
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2836
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2837
                }
2838
            }
2839
            for(i=20; i<20+4; i++){
2840
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2841
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2842
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2843
                }
2844
            }
2845
        }
2846
    }
2847
    if(h->deblocking_filter) {
2848
        backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2849
        fill_caches(h, mb_type, 1); //FIXME dont fill stuff which isnt used by filter_mb
2850
        filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2851
    }
2852
}
2853

    
2854
/**
2855
 * fills the default_ref_list.
2856
 */
2857
static int fill_default_ref_list(H264Context *h){
2858
    MpegEncContext * const s = &h->s;
2859
    int i;
2860
    int smallest_poc_greater_than_current = -1;
2861
    Picture sorted_short_ref[16];
2862
    
2863
    if(h->slice_type==B_TYPE){
2864
        int out_i;
2865
        int limit= -1;
2866

    
2867
        /* sort frame according to poc in B slice */
2868
        for(out_i=0; out_i<h->short_ref_count; out_i++){
2869
            int best_i=-1;
2870
            int best_poc=INT_MAX;
2871

    
2872
            for(i=0; i<h->short_ref_count; i++){
2873
                const int poc= h->short_ref[i]->poc;
2874
                if(poc > limit && poc < best_poc){
2875
                    best_poc= poc;
2876
                    best_i= i;
2877
                }
2878
            }
2879
            
2880
            assert(best_i != -1);
2881
            
2882
            limit= best_poc;
2883
            sorted_short_ref[out_i]= *h->short_ref[best_i];
2884
            tprintf("sorted poc: %d->%d poc:%d fn:%d\n", best_i, out_i, sorted_short_ref[out_i].poc, sorted_short_ref[out_i].frame_num);
2885
            if (-1 == smallest_poc_greater_than_current) {
2886
                if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
2887
                    smallest_poc_greater_than_current = out_i;
2888
                }
2889
            }
2890
        }
2891
    }
2892

    
2893
    if(s->picture_structure == PICT_FRAME){
2894
        if(h->slice_type==B_TYPE){
2895
            int list;
2896
            tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
2897

    
2898
            // find the largest poc
2899
            for(list=0; list<2; list++){
2900
                int index = 0;
2901
                int j= -99;
2902
                int step= list ? -1 : 1;
2903

    
2904
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
2905
                    while(j<0 || j>= h->short_ref_count){
2906
                        step = -step;
2907
                        j= smallest_poc_greater_than_current + (step>>1);
2908
                    }
2909
                    if(sorted_short_ref[j].reference != 3) continue;
2910
                    h->default_ref_list[list][index  ]= sorted_short_ref[j];
2911
                    h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
2912
                }
2913

    
2914
                for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
2915
                    if(h->long_ref[i] == NULL) continue;
2916
                    if(h->long_ref[i]->reference != 3) continue;
2917

    
2918
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
2919
                    h->default_ref_list[ list ][index++].pic_id= i;;
2920
                }
2921
                
2922
                if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
2923
                    // swap the two first elements of L1 when
2924
                    // L0 and L1 are identical
2925
                    Picture temp= h->default_ref_list[1][0];
2926
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
2927
                    h->default_ref_list[1][0] = temp;
2928
                }
2929

    
2930
                if(index < h->ref_count[ list ])
2931
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2932
            }
2933
        }else{
2934
            int index=0;
2935
            for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2936
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2937
                h->default_ref_list[0][index  ]= *h->short_ref[i];
2938
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2939
            }
2940
            for(i = 0; i < 16 && index < h->ref_count[0]; i++){
2941
                if(h->long_ref[i] == NULL) continue;
2942
                if(h->long_ref[i]->reference != 3) continue;
2943
                h->default_ref_list[0][index  ]= *h->long_ref[i];
2944
                h->default_ref_list[0][index++].pic_id= i;;
2945
            }
2946
            if(index < h->ref_count[0])
2947
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2948
        }
2949
    }else{ //FIELD
2950
        if(h->slice_type==B_TYPE){
2951
        }else{
2952
            //FIXME second field balh
2953
        }
2954
    }
2955
#ifdef TRACE
2956
    for (i=0; i<h->ref_count[0]; i++) {
2957
        tprintf("List0: %s fn:%d 0x%p\n", (h->default_ref_list[0][i].long_ref ? "LT" : "ST"), h->default_ref_list[0][i].pic_id, h->default_ref_list[0][i].data[0]);
2958
    }
2959
    if(h->slice_type==B_TYPE){
2960
        for (i=0; i<h->ref_count[1]; i++) {
2961
            tprintf("List1: %s fn:%d 0x%p\n", (h->default_ref_list[1][i].long_ref ? "LT" : "ST"), h->default_ref_list[1][i].pic_id, h->default_ref_list[0][i].data[0]);
2962
        }
2963
    }
2964
#endif
2965
    return 0;
2966
}
2967

    
2968
static void print_short_term(H264Context *h);
2969
static void print_long_term(H264Context *h);
2970

    
2971
static int decode_ref_pic_list_reordering(H264Context *h){
2972
    MpegEncContext * const s = &h->s;
2973
    int list;
2974
    
2975
    print_short_term(h);
2976
    print_long_term(h);
2977
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2978
    
2979
    for(list=0; list<2; list++){
2980
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2981

    
2982
        if(get_bits1(&s->gb)){
2983
            int pred= h->curr_pic_num;
2984
            int index;
2985

    
2986
            for(index=0; ; index++){
2987
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2988
                int pic_id;
2989
                int i;
2990
                
2991
                if(reordering_of_pic_nums_idc==3) 
2992
                    break;
2993
                
2994
                if(index >= h->ref_count[list]){
2995
                    av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2996
                    return -1;
2997
                }
2998
                
2999
                if(reordering_of_pic_nums_idc<3){
3000
                    if(reordering_of_pic_nums_idc<2){
3001
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3002

    
3003
                        if(abs_diff_pic_num >= h->max_pic_num){
3004
                            av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3005
                            return -1;
3006
                        }
3007

    
3008
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3009
                        else                                pred+= abs_diff_pic_num;
3010
                        pred &= h->max_pic_num - 1;
3011
                    
3012
                        for(i= h->ref_count[list]-1; i>=index; i--){
3013
                            if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
3014
                                break;
3015
                        }
3016
                    }else{
3017
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3018

    
3019
                        for(i= h->ref_count[list]-1; i>=index; i--){
3020
                            if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
3021
                                break;
3022
                        }
3023
                    }
3024

    
3025
                    if(i < index){
3026
                        av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3027
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3028
                    }else if(i > index){
3029
                        Picture tmp= h->ref_list[list][i];
3030
                        for(; i>index; i--){
3031
                            h->ref_list[list][i]= h->ref_list[list][i-1];
3032
                        }
3033
                        h->ref_list[list][index]= tmp;
3034
                    }
3035
                }else{
3036
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3037
                    return -1;
3038
                }
3039
            }
3040
        }
3041

    
3042
        if(h->slice_type!=B_TYPE) break;
3043
    }
3044
    
3045
    if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3046
        direct_dist_scale_factor(h);
3047
    return 0;    
3048
}
3049

    
3050
static int pred_weight_table(H264Context *h){
3051
    MpegEncContext * const s = &h->s;
3052
    int list, i;
3053
    int luma_def, chroma_def;
3054
    
3055
    h->use_weight= 0;
3056
    h->use_weight_chroma= 0;
3057
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3058
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3059
    luma_def = 1<<h->luma_log2_weight_denom;
3060
    chroma_def = 1<<h->chroma_log2_weight_denom;
3061

    
3062
    for(list=0; list<2; list++){
3063
        for(i=0; i<h->ref_count[list]; i++){
3064
            int luma_weight_flag, chroma_weight_flag;
3065
            
3066
            luma_weight_flag= get_bits1(&s->gb);
3067
            if(luma_weight_flag){
3068
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
3069
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
3070
                if(   h->luma_weight[list][i] != luma_def
3071
                   || h->luma_offset[list][i] != 0)
3072
                    h->use_weight= 1;
3073
            }else{
3074
                h->luma_weight[list][i]= luma_def;
3075
                h->luma_offset[list][i]= 0;
3076
            }
3077

    
3078
            chroma_weight_flag= get_bits1(&s->gb);
3079
            if(chroma_weight_flag){
3080
                int j;
3081
                for(j=0; j<2; j++){
3082
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3083
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3084
                    if(   h->chroma_weight[list][i][j] != chroma_def
3085
                       || h->chroma_offset[list][i][j] != 0)
3086
                        h->use_weight_chroma= 1;
3087
                }
3088
            }else{
3089
                int j;
3090
                for(j=0; j<2; j++){
3091
                    h->chroma_weight[list][i][j]= chroma_def;
3092
                    h->chroma_offset[list][i][j]= 0;
3093
                }
3094
            }
3095
        }
3096
        if(h->slice_type != B_TYPE) break;
3097
    }
3098
    h->use_weight= h->use_weight || h->use_weight_chroma;
3099
    return 0;
3100
}
3101

    
3102
static void implicit_weight_table(H264Context *h){
3103
    MpegEncContext * const s = &h->s;
3104
    int ref0, ref1;
3105
    int cur_poc = s->current_picture_ptr->poc;
3106

    
3107
    if(   h->ref_count[0] == 1 && h->ref_count[1] == 1
3108
       && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3109
        h->use_weight= 0;
3110
        h->use_weight_chroma= 0;
3111
        return;
3112
    }
3113

    
3114
    h->use_weight= 2;
3115
    h->use_weight_chroma= 2;
3116
    h->luma_log2_weight_denom= 5;
3117
    h->chroma_log2_weight_denom= 5;
3118

    
3119
    /* FIXME: MBAFF */
3120
    for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3121
        int poc0 = h->ref_list[0][ref0].poc;
3122
        for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3123
            int poc1 = h->ref_list[0][ref1].poc;
3124
            int td = clip(poc1 - poc0, -128, 127);
3125
            if(td){
3126
                int tb = clip(cur_poc - poc0, -128, 127);
3127
                int tx = (16384 + (ABS(td) >> 1)) / td;
3128
                int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3129
                if(dist_scale_factor < -64 || dist_scale_factor > 128)
3130
                    h->implicit_weight[ref0][ref1] = 32;
3131
                else
3132
                    h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3133
            }else
3134
                h->implicit_weight[ref0][ref1] = 32;
3135
        }
3136
    }
3137
}
3138

    
3139
/**
3140
 * instantaneous decoder refresh.
3141
 */
3142
static void idr(H264Context *h){
3143
    int i,j;
3144

    
3145
#define CHECK_DELAY(pic) \
3146
    for(j = 0; h->delayed_pic[j]; j++) \
3147
        if(pic == h->delayed_pic[j]){ \
3148
            pic->reference=1; \
3149
            break; \
3150
        }
3151

    
3152
    for(i=0; i<h->long_ref_count; i++){
3153
        h->long_ref[i]->reference=0;
3154
        CHECK_DELAY(h->long_ref[i]);
3155
        h->long_ref[i]= NULL;
3156
    }
3157
    h->long_ref_count=0;
3158

    
3159
    for(i=0; i<h->short_ref_count; i++){
3160
        h->short_ref[i]->reference=0;
3161
        CHECK_DELAY(h->short_ref[i]);
3162
        h->short_ref[i]= NULL;
3163
    }
3164
    h->short_ref_count=0;
3165
}
3166
#undef CHECK_DELAY
3167

    
3168
/**
3169
 *
3170
 * @return the removed picture or NULL if an error occures
3171
 */
3172
static Picture * remove_short(H264Context *h, int frame_num){
3173
    MpegEncContext * const s = &h->s;
3174
    int i;
3175
    
3176
    if(s->avctx->debug&FF_DEBUG_MMCO)
3177
        av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3178
    
3179
    for(i=0; i<h->short_ref_count; i++){
3180
        Picture *pic= h->short_ref[i];
3181
        if(s->avctx->debug&FF_DEBUG_MMCO)
3182
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3183
        if(pic->frame_num == frame_num){
3184
            h->short_ref[i]= NULL;
3185
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3186
            h->short_ref_count--;
3187
            return pic;
3188
        }
3189
    }
3190
    return NULL;
3191
}
3192

    
3193
/**
3194
 *
3195
 * @return the removed picture or NULL if an error occures
3196
 */
3197
static Picture * remove_long(H264Context *h, int i){
3198
    Picture *pic;
3199

    
3200
    pic= h->long_ref[i];
3201
    h->long_ref[i]= NULL;
3202
    if(pic) h->long_ref_count--;
3203

    
3204
    return pic;
3205
}
3206

    
3207
/**
3208
 * print short term list
3209
 */
3210
static void print_short_term(H264Context *h) {
3211
    uint32_t i;
3212
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3213
        av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3214
        for(i=0; i<h->short_ref_count; i++){
3215
            Picture *pic= h->short_ref[i];
3216
            av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3217
        }
3218
    }
3219
}
3220

    
3221
/**
3222
 * print long term list
3223
 */
3224
static void print_long_term(H264Context *h) {
3225
    uint32_t i;
3226
    if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3227
        av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3228
        for(i = 0; i < 16; i++){
3229
            Picture *pic= h->long_ref[i];
3230
            if (pic) {
3231
                av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3232
            }
3233
        }
3234
    }
3235
}
3236

    
3237
/**
3238
 * Executes the reference picture marking (memory management control operations).
3239
 */
3240
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3241
    MpegEncContext * const s = &h->s;
3242
    int i, j;
3243
    int current_is_long=0;
3244
    Picture *pic;
3245
    
3246
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3247
        av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3248
        
3249
    for(i=0; i<mmco_count; i++){
3250
        if(s->avctx->debug&FF_DEBUG_MMCO)
3251
            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);
3252

    
3253
        switch(mmco[i].opcode){
3254
        case MMCO_SHORT2UNUSED:
3255
            pic= remove_short(h, mmco[i].short_frame_num);
3256
            if(pic==NULL) return -1;
3257
            pic->reference= 0;
3258
            break;
3259
        case MMCO_SHORT2LONG:
3260
            pic= remove_long(h, mmco[i].long_index);
3261
            if(pic) pic->reference=0;
3262
            
3263
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3264
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
3265
            break;
3266
        case MMCO_LONG2UNUSED:
3267
            pic= remove_long(h, mmco[i].long_index);
3268
            if(pic==NULL) return -1;
3269
            pic->reference= 0;
3270
            break;
3271
        case MMCO_LONG:
3272
            pic= remove_long(h, mmco[i].long_index);
3273
            if(pic) pic->reference=0;
3274
            
3275
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3276
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
3277
            h->long_ref_count++;
3278
            
3279
            current_is_long=1;
3280
            break;
3281
        case MMCO_SET_MAX_LONG:
3282
            assert(mmco[i].long_index <= 16);
3283
            // just remove the long term which index is greater than new max
3284
            for(j = mmco[i].long_index; j<16; j++){
3285
                pic = remove_long(h, j);
3286
                if (pic) pic->reference=0;
3287
            }
3288
            break;
3289
        case MMCO_RESET:
3290
            while(h->short_ref_count){
3291
                pic= remove_short(h, h->short_ref[0]->frame_num);
3292
                pic->reference=0;
3293
            }
3294
            for(j = 0; j < 16; j++) {
3295
                pic= remove_long(h, j);
3296
                if(pic) pic->reference=0;
3297
            }
3298
            break;
3299
        default: assert(0);
3300
        }
3301
    }
3302
    
3303
    if(!current_is_long){
3304
        pic= remove_short(h, s->current_picture_ptr->frame_num);
3305
        if(pic){
3306
            pic->reference=0;
3307
            av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3308
        }
3309
        
3310
        if(h->short_ref_count)
3311
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3312

    
3313
        h->short_ref[0]= s->current_picture_ptr;
3314
        h->short_ref[0]->long_ref=0;
3315
        h->short_ref_count++;
3316
    }
3317
    
3318
    print_short_term(h);
3319
    print_long_term(h);
3320
    return 0; 
3321
}
3322

    
3323
static int decode_ref_pic_marking(H264Context *h){
3324
    MpegEncContext * const s = &h->s;
3325
    int i;
3326
    
3327
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3328
        s->broken_link= get_bits1(&s->gb) -1;
3329
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3330
        if(h->mmco[0].long_index == -1)
3331
            h->mmco_index= 0;
3332
        else{
3333
            h->mmco[0].opcode= MMCO_LONG;
3334
            h->mmco_index= 1;
3335
        } 
3336
    }else{
3337
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3338
            for(i= 0; i<MAX_MMCO_COUNT; i++) { 
3339
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3340

    
3341
                h->mmco[i].opcode= opcode;
3342
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3343
                    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
3344
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3345
                        fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3346
                        return -1;
3347
                    }*/
3348
                }
3349
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3350
                    h->mmco[i].long_index= get_ue_golomb(&s->gb);
3351
                    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){
3352
                        av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3353
                        return -1;
3354
                    }
3355
                }
3356
                    
3357
                if(opcode > MMCO_LONG){
3358
                    av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3359
                    return -1;
3360
                }
3361
                if(opcode == MMCO_END)
3362
                    break;
3363
            }
3364
            h->mmco_index= i;
3365
        }else{
3366
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3367

    
3368
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3369
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3370
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3371
                h->mmco_index= 1;
3372
            }else
3373
                h->mmco_index= 0;
3374
        }
3375
    }
3376
    
3377
    return 0; 
3378
}
3379

    
3380
static int init_poc(H264Context *h){
3381
    MpegEncContext * const s = &h->s;
3382
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3383
    int field_poc[2];
3384

    
3385
    if(h->nal_unit_type == NAL_IDR_SLICE){
3386
        h->frame_num_offset= 0;
3387
    }else{
3388
        if(h->frame_num < h->prev_frame_num)
3389
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3390
        else
3391
            h->frame_num_offset= h->prev_frame_num_offset;
3392
    }
3393

    
3394
    if(h->sps.poc_type==0){
3395
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3396

    
3397
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3398
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3399
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3400
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3401
        else
3402
            h->poc_msb = h->prev_poc_msb;
3403
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3404
        field_poc[0] = 
3405
        field_poc[1] = h->poc_msb + h->poc_lsb;
3406
        if(s->picture_structure == PICT_FRAME) 
3407
            field_poc[1] += h->delta_poc_bottom;
3408
    }else if(h->sps.poc_type==1){
3409
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3410
        int i;
3411

    
3412
        if(h->sps.poc_cycle_length != 0)
3413
            abs_frame_num = h->frame_num_offset + h->frame_num;
3414
        else
3415
            abs_frame_num = 0;
3416

    
3417
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
3418
            abs_frame_num--;
3419
            
3420
        expected_delta_per_poc_cycle = 0;
3421
        for(i=0; i < h->sps.poc_cycle_length; i++)
3422
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3423

    
3424
        if(abs_frame_num > 0){
3425
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3426
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3427

    
3428
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3429
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
3430
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3431
        } else
3432
            expectedpoc = 0;
3433

    
3434
        if(h->nal_ref_idc == 0) 
3435
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3436
        
3437
        field_poc[0] = expectedpoc + h->delta_poc[0];
3438
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3439

    
3440
        if(s->picture_structure == PICT_FRAME)
3441
            field_poc[1] += h->delta_poc[1];
3442
    }else{
3443
        int poc;
3444
        if(h->nal_unit_type == NAL_IDR_SLICE){
3445
            poc= 0;
3446
        }else{
3447
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3448
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3449
        }
3450
        field_poc[0]= poc;
3451
        field_poc[1]= poc;
3452
    }
3453
    
3454
    if(s->picture_structure != PICT_BOTTOM_FIELD)
3455
        s->current_picture_ptr->field_poc[0]= field_poc[0];
3456
    if(s->picture_structure != PICT_TOP_FIELD)
3457
        s->current_picture_ptr->field_poc[1]= field_poc[1];
3458
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3459
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3460

    
3461
    return 0;
3462
}
3463

    
3464
/**
3465
 * decodes a slice header.
3466
 * this will allso call MPV_common_init() and frame_start() as needed
3467
 */
3468
static int decode_slice_header(H264Context *h){
3469
    MpegEncContext * const s = &h->s;
3470
    int first_mb_in_slice, pps_id;
3471
    int num_ref_idx_active_override_flag;
3472
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3473

    
3474
    s->current_picture.reference= h->nal_ref_idc != 0;
3475

    
3476
    first_mb_in_slice= get_ue_golomb(&s->gb);
3477

    
3478
    h->slice_type= get_ue_golomb(&s->gb);
3479
    if(h->slice_type > 9){
3480
        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);
3481
        return -1;
3482
    }
3483
    if(h->slice_type > 4){
3484
        h->slice_type -= 5;
3485
        h->slice_type_fixed=1;
3486
    }else
3487
        h->slice_type_fixed=0;
3488
    
3489
    h->slice_type= slice_type_map[ h->slice_type ];
3490
    
3491
    s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
3492
        
3493
    pps_id= get_ue_golomb(&s->gb);
3494
    if(pps_id>255){
3495
        av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3496
        return -1;
3497
    }
3498
    h->pps= h->pps_buffer[pps_id];
3499
    if(h->pps.slice_group_count == 0){
3500
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3501
        return -1;
3502
    }
3503

    
3504
    h->sps= h->sps_buffer[ h->pps.sps_id ];
3505
    if(h->sps.log2_max_frame_num == 0){
3506
        av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3507
        return -1;
3508
    }
3509
    
3510
    s->mb_width= h->sps.mb_width;
3511
    s->mb_height= h->sps.mb_height;
3512
    
3513
    h->b_stride=  s->mb_width*4 + 1;
3514
    h->b8_stride= s->mb_width*2 + 1;
3515

    
3516
    s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3517
    s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
3518
    
3519
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3520
    if(h->sps.frame_mbs_only_flag)
3521
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
3522
    else
3523
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
3524
    
3525
    if (s->context_initialized 
3526
        && (   s->width != s->avctx->width || s->height != s->avctx->height)) {
3527
        free_tables(h);
3528
        MPV_common_end(s);
3529
    }
3530
    if (!s->context_initialized) {
3531
        if (MPV_common_init(s) < 0)
3532
            return -1;
3533

    
3534
        alloc_tables(h);
3535

    
3536
        s->avctx->width = s->width;
3537
        s->avctx->height = s->height;
3538
        s->avctx->sample_aspect_ratio= h->sps.sar;
3539
        if(!s->avctx->sample_aspect_ratio.den)
3540
            s->avctx->sample_aspect_ratio.den = 1;
3541

    
3542
        if(h->sps.timing_info_present_flag && h->sps.fixed_frame_rate_flag){
3543
            s->avctx->frame_rate = h->sps.time_scale;
3544
            s->avctx->frame_rate_base = h->sps.num_units_in_tick;
3545
        }
3546
    }
3547

    
3548
    if(h->slice_num == 0){
3549
        frame_start(h);
3550
    }
3551

    
3552
    s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3553
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3554

    
3555
    if(h->sps.frame_mbs_only_flag){
3556
        s->picture_structure= PICT_FRAME;
3557
    }else{
3558
        if(get_bits1(&s->gb)) //field_pic_flag
3559
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3560
        else
3561
            s->picture_structure= PICT_FRAME;
3562
    }
3563

    
3564
    if(s->picture_structure==PICT_FRAME){
3565
        h->curr_pic_num=   h->frame_num;
3566
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3567
    }else{
3568
        h->curr_pic_num= 2*h->frame_num;
3569
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3570
    }
3571
        
3572
    if(h->nal_unit_type == NAL_IDR_SLICE){
3573
        get_ue_golomb(&s->gb); /* idr_pic_id */
3574
    }
3575
   
3576
    if(h->sps.poc_type==0){
3577
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3578
        
3579
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3580
            h->delta_poc_bottom= get_se_golomb(&s->gb);
3581
        }
3582
    }
3583
    
3584
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3585
        h->delta_poc[0]= get_se_golomb(&s->gb);
3586
        
3587
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3588
            h->delta_poc[1]= get_se_golomb(&s->gb);
3589
    }
3590
    
3591
    init_poc(h);
3592
    
3593
    if(h->pps.redundant_pic_cnt_present){
3594
        h->redundant_pic_count= get_ue_golomb(&s->gb);
3595
    }
3596

    
3597
    //set defaults, might be overriden a few line later
3598
    h->ref_count[0]= h->pps.ref_count[0];
3599
    h->ref_count[1]= h->pps.ref_count[1];
3600

    
3601
    if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3602
        if(h->slice_type == B_TYPE){
3603
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
3604
        }
3605
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
3606
    
3607
        if(num_ref_idx_active_override_flag){
3608
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3609
            if(h->slice_type==B_TYPE)
3610
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3611

    
3612
            if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3613
                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3614
                return -1;
3615
            }
3616
        }
3617
    }
3618

    
3619
    if(h->slice_num == 0){
3620
        fill_default_ref_list(h);
3621
    }
3622

    
3623
    decode_ref_pic_list_reordering(h);
3624

    
3625
    if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) 
3626
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3627
        pred_weight_table(h);
3628
    else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
3629
        implicit_weight_table(h);
3630
    else
3631
        h->use_weight = 0;
3632
    
3633
    if(s->current_picture.reference)
3634
        decode_ref_pic_marking(h);
3635

    
3636
    if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3637
        h->cabac_init_idc = get_ue_golomb(&s->gb);
3638

    
3639
    h->last_qscale_diff = 0;
3640
    s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3641
    if(s->qscale<0 || s->qscale>51){
3642
        av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3643
        return -1;
3644
    }
3645
    h->chroma_qp = get_chroma_qp(h, s->qscale);
3646
    //FIXME qscale / qp ... stuff
3647
    if(h->slice_type == SP_TYPE){
3648
        get_bits1(&s->gb); /* sp_for_switch_flag */
3649
    }
3650
    if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3651
        get_se_golomb(&s->gb); /* slice_qs_delta */
3652
    }
3653

    
3654
    h->deblocking_filter = 1;
3655
    h->slice_alpha_c0_offset = 0;
3656
    h->slice_beta_offset = 0;
3657
    if( h->pps.deblocking_filter_parameters_present ) {
3658
        h->deblocking_filter= get_ue_golomb(&s->gb);
3659
        if(h->deblocking_filter < 2) 
3660
            h->deblocking_filter^= 1; // 1<->0
3661

    
3662
        if( h->deblocking_filter ) {
3663
            h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3664
            h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3665
        }
3666
    }
3667

    
3668
#if 0 //FMO
3669
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3670
        slice_group_change_cycle= get_bits(&s->gb, ?);
3671
#endif
3672

    
3673
    h->slice_num++;
3674

    
3675
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3676
        av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d weight:%d%s\n", 
3677
               h->slice_num, first_mb_in_slice, 
3678
               av_get_pict_type_char(h->slice_type),
3679
               pps_id, h->frame_num,
3680
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3681
               h->ref_count[0], h->ref_count[1],
3682
               s->qscale,
3683
               h->deblocking_filter,
3684
               h->use_weight,
3685
               h->use_weight==1 && h->use_weight_chroma ? "c" : ""
3686
               );
3687
    }
3688

    
3689
    return 0;
3690
}
3691

    
3692
/**
3693
 *
3694
 */
3695
static inline int get_level_prefix(GetBitContext *gb){
3696
    unsigned int buf;
3697
    int log;
3698
    
3699
    OPEN_READER(re, gb);
3700
    UPDATE_CACHE(re, gb);
3701
    buf=GET_CACHE(re, gb);
3702
    
3703
    log= 32 - av_log2(buf);
3704
#ifdef TRACE
3705
    print_bin(buf>>(32-log), log);
3706
    av_log(NULL, AV_LOG_DEBUG, "%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3707
#endif
3708

    
3709
    LAST_SKIP_BITS(re, gb, log);
3710
    CLOSE_READER(re, gb);
3711

    
3712
    return log-1;
3713
}
3714

    
3715
/**
3716
 * decodes a residual block.
3717
 * @param n block index
3718
 * @param scantable scantable
3719
 * @param max_coeff number of coefficients in the block
3720
 * @return <0 if an error occured
3721
 */
3722
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3723
    MpegEncContext * const s = &h->s;
3724
    const uint16_t *qmul= dequant_coeff[qp];
3725
    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};
3726
    int level[16], run[16];
3727
    int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3728

    
3729
    //FIXME put trailing_onex into the context
3730

    
3731
    if(n == CHROMA_DC_BLOCK_INDEX){
3732
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3733
        total_coeff= coeff_token>>2;
3734
    }else{    
3735
        if(n == LUMA_DC_BLOCK_INDEX){
3736
            total_coeff= pred_non_zero_count(h, 0);
3737
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3738
            total_coeff= coeff_token>>2;
3739
        }else{
3740
            total_coeff= pred_non_zero_count(h, n);
3741
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3742
            total_coeff= coeff_token>>2;
3743
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3744
        }
3745
    }
3746

    
3747
    //FIXME set last_non_zero?
3748

    
3749
    if(total_coeff==0)
3750
        return 0;
3751
        
3752
    trailing_ones= coeff_token&3;
3753
    tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3754
    assert(total_coeff<=16);
3755
    
3756
    for(i=0; i<trailing_ones; i++){
3757
        level[i]= 1 - 2*get_bits1(gb);
3758
    }
3759

    
3760
    suffix_length= total_coeff > 10 && trailing_ones < 3;
3761

    
3762
    for(; i<total_coeff; i++){
3763
        const int prefix= get_level_prefix(gb);
3764
        int level_code, mask;
3765

    
3766
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3767
            if(suffix_length)
3768
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3769
            else
3770
                level_code= (prefix<<suffix_length); //part
3771
        }else if(prefix==14){
3772
            if(suffix_length)
3773
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3774
            else
3775
                level_code= prefix + get_bits(gb, 4); //part
3776
        }else if(prefix==15){
3777
            level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3778
            if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3779
        }else{
3780
            av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3781
            return -1;
3782
        }
3783

    
3784
        if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3785

    
3786
        mask= -(level_code&1);
3787
        level[i]= (((2+level_code)>>1) ^ mask) - mask;
3788

    
3789
        if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3790

    
3791
#if 1
3792
        if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3793
#else        
3794
        if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3795
        /* ? == prefix > 2 or sth */
3796
#endif
3797
        tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3798
    }
3799

    
3800
    if(total_coeff == max_coeff)
3801
        zeros_left=0;
3802
    else{
3803
        if(n == CHROMA_DC_BLOCK_INDEX)
3804
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3805
        else
3806
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3807
    }
3808
    
3809
    for(i=0; i<total_coeff-1; i++){
3810
        if(zeros_left <=0)
3811
            break;
3812
        else if(zeros_left < 7){
3813
            run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3814
        }else{
3815
            run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3816
        }
3817
        zeros_left -= run[i];
3818
    }
3819

    
3820
    if(zeros_left<0){
3821
        av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3822
        return -1;
3823
    }
3824
    
3825
    for(; i<total_coeff-1; i++){
3826
        run[i]= 0;
3827
    }
3828

    
3829
    run[i]= zeros_left;
3830

    
3831
    coeff_num=-1;
3832
    if(n > 24){
3833
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3834
            int j;
3835

    
3836
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3837
            j= scantable[ coeff_num ];
3838

    
3839
            block[j]= level[i];
3840
        }
3841
    }else{
3842
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into  rundecode?
3843
            int j;
3844

    
3845
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3846
            j= scantable[ coeff_num ];
3847

    
3848
            block[j]= level[i] * qmul[j];
3849
//            printf("%d %d  ", block[j], qmul[j]);
3850
        }
3851
    }
3852
    return 0;
3853
}
3854

    
3855
/**
3856
 * decodes a P_SKIP or B_SKIP macroblock
3857
 */
3858
static void decode_mb_skip(H264Context *h){
3859
    MpegEncContext * const s = &h->s;
3860
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3861
    int mb_type;
3862
    
3863
    memset(h->non_zero_count[mb_xy], 0, 16);
3864
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3865

    
3866
    if( h->slice_type == B_TYPE )
3867
    {
3868
        // just for fill_caches. pred_direct_motion will set the real mb_type
3869
        mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
3870
        //FIXME mbaff
3871

    
3872
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
3873
        pred_direct_motion(h, &mb_type);
3874
        if(h->pps.cabac){
3875
            fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
3876
            fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
3877
        }
3878
    }
3879
    else
3880
    {
3881
        int mx, my;
3882
        mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
3883

    
3884
        if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3885
            h->mb_field_decoding_flag= get_bits1(&s->gb);
3886
        }
3887
        if(h->mb_field_decoding_flag)
3888
            mb_type|= MB_TYPE_INTERLACED;
3889
        
3890
        fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
3891
        pred_pskip_motion(h, &mx, &my);
3892
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3893
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3894
        if(h->pps.cabac)
3895
            fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
3896
    }
3897

    
3898
    write_back_motion(h, mb_type);
3899
    s->current_picture.mb_type[mb_xy]= mb_type|MB_TYPE_SKIP;
3900
    s->current_picture.qscale_table[mb_xy]= s->qscale;
3901
    h->slice_table[ mb_xy ]= h->slice_num;
3902
    h->prev_mb_skiped= 1;
3903
}
3904

    
3905
/**
3906
 * decodes a macroblock
3907
 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3908
 */
3909
static int decode_mb_cavlc(H264Context *h){
3910
    MpegEncContext * const s = &h->s;
3911
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3912
    int mb_type, partition_count, cbp;
3913

    
3914
    s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?    
3915

    
3916
    tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3917
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3918
                down the code */
3919
    if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3920
        if(s->mb_skip_run==-1)
3921
            s->mb_skip_run= get_ue_golomb(&s->gb);
3922
        
3923
        if (s->mb_skip_run--) {
3924
            decode_mb_skip(h);
3925
            return 0;
3926
        }
3927
    }
3928
    if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3929
        if((s->mb_y&1)==0)
3930
            h->mb_field_decoding_flag = get_bits1(&s->gb);
3931
    }else
3932
        h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3933
    
3934
    h->prev_mb_skiped= 0;
3935
    
3936
    mb_type= get_ue_golomb(&s->gb);
3937
    if(h->slice_type == B_TYPE){
3938
        if(mb_type < 23){
3939
            partition_count= b_mb_type_info[mb_type].partition_count;
3940
            mb_type=         b_mb_type_info[mb_type].type;
3941
        }else{
3942
            mb_type -= 23;
3943
            goto decode_intra_mb;
3944
        }
3945
    }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3946
        if(mb_type < 5){
3947
            partition_count= p_mb_type_info[mb_type].partition_count;
3948
            mb_type=         p_mb_type_info[mb_type].type;
3949
        }else{
3950
            mb_type -= 5;
3951
            goto decode_intra_mb;
3952
        }
3953
    }else{
3954
       assert(h->slice_type == I_TYPE);
3955
decode_intra_mb:
3956
        if(mb_type > 25){
3957
            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);
3958
            return -1;
3959
        }
3960
        partition_count=0;
3961
        cbp= i_mb_type_info[mb_type].cbp;
3962
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3963
        mb_type= i_mb_type_info[mb_type].type;
3964
    }
3965

    
3966
    if(h->mb_field_decoding_flag)
3967
        mb_type |= MB_TYPE_INTERLACED;
3968

    
3969
    s->current_picture.mb_type[mb_xy]= mb_type;
3970
    h->slice_table[ mb_xy ]= h->slice_num;
3971
    
3972
    if(IS_INTRA_PCM(mb_type)){
3973
        const uint8_t *ptr;
3974
        int x, y;
3975
        
3976
        // we assume these blocks are very rare so we dont optimize it
3977
        align_get_bits(&s->gb);
3978
        
3979
        ptr= s->gb.buffer + get_bits_count(&s->gb);
3980
    
3981
        for(y=0; y<16; y++){
3982
            const int index= 4*(y&3) + 64*(y>>2);
3983
            for(x=0; x<16; x++){
3984
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3985
            }
3986
        }
3987
        for(y=0; y<8; y++){
3988
            const int index= 256 + 4*(y&3) + 32*(y>>2);
3989
            for(x=0; x<8; x++){
3990
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3991
            }
3992
        }
3993
        for(y=0; y<8; y++){
3994
            const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3995
            for(x=0; x<8; x++){
3996
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3997
            }
3998
        }
3999
    
4000
        skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
4001
        
4002
        //FIXME deblock filter, non_zero_count_cache init ...
4003
        memset(h->non_zero_count[mb_xy], 16, 16);
4004
        s->current_picture.qscale_table[mb_xy]= s->qscale;
4005
        
4006
        return 0;
4007
    }
4008
        
4009
    fill_caches(h, mb_type, 0);
4010

    
4011
    //mb_pred
4012
    if(IS_INTRA(mb_type)){
4013
//            init_top_left_availability(h);
4014
            if(IS_INTRA4x4(mb_type)){
4015
                int i;
4016

    
4017
//                fill_intra4x4_pred_table(h);
4018
                for(i=0; i<16; i++){
4019
                    const int mode_coded= !get_bits1(&s->gb);
4020
                    const int predicted_mode=  pred_intra_mode(h, i);
4021
                    int mode;
4022

    
4023
                    if(mode_coded){
4024
                        const int rem_mode= get_bits(&s->gb, 3);
4025
                        if(rem_mode<predicted_mode)
4026
                            mode= rem_mode;
4027
                        else
4028
                            mode= rem_mode + 1;
4029
                    }else{
4030
                        mode= predicted_mode;
4031
                    }
4032
                    
4033
                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
4034
                }
4035
                write_back_intra_pred_mode(h);
4036
                if( check_intra4x4_pred_mode(h) < 0)
4037
                    return -1;
4038
            }else{
4039
                h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
4040
                if(h->intra16x16_pred_mode < 0)
4041
                    return -1;
4042
            }
4043
            h->chroma_pred_mode= get_ue_golomb(&s->gb);
4044

    
4045
            h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
4046
            if(h->chroma_pred_mode < 0)
4047
                return -1;
4048
    }else if(partition_count==4){
4049
        int i, j, sub_partition_count[4], list, ref[2][4];
4050
        
4051
        if(h->slice_type == B_TYPE){
4052
            for(i=0; i<4; i++){
4053
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4054
                if(h->sub_mb_type[i] >=13){
4055
                    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);
4056
                    return -1;
4057
                }
4058
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4059
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4060
            }
4061
            if(   IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4062
               || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3]))
4063
                pred_direct_motion(h, &mb_type);
4064
        }else{
4065
            assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
4066
            for(i=0; i<4; i++){
4067
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4068
                if(h->sub_mb_type[i] >=4){
4069
                    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);
4070
                    return -1;
4071
                }
4072
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4073
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4074
            }
4075
        }
4076
        
4077
        for(list=0; list<2; list++){
4078
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4079
            if(ref_count == 0) continue;
4080
            for(i=0; i<4; i++){
4081
                if(IS_DIRECT(h->sub_mb_type[i])) continue;
4082
                if(IS_DIR(h->sub_mb_type[i], 0, list)){
4083
                    ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
4084
                }else{
4085
                 //FIXME
4086
                    ref[list][i] = -1;
4087
                }
4088
            }
4089
        }
4090
        
4091
        for(list=0; list<2; list++){
4092
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4093
            if(ref_count == 0) continue;
4094

    
4095
            for(i=0; i<4; i++){
4096
                if(IS_DIRECT(h->sub_mb_type[i])) continue;
4097
                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4098
                h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4099

    
4100
                if(IS_DIR(h->sub_mb_type[i], 0, list)){
4101
                    const int sub_mb_type= h->sub_mb_type[i];
4102
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4103
                    for(j=0; j<sub_partition_count[i]; j++){
4104
                        int mx, my;
4105
                        const int index= 4*i + block_width*j;
4106
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4107
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
4108
                        mx += get_se_golomb(&s->gb);
4109
                        my += get_se_golomb(&s->gb);
4110
                        tprintf("final mv:%d %d\n", mx, my);
4111

    
4112
                        if(IS_SUB_8X8(sub_mb_type)){
4113
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= 
4114
                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4115
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= 
4116
                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4117
                        }else if(IS_SUB_8X4(sub_mb_type)){
4118
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4119
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4120
                        }else if(IS_SUB_4X8(sub_mb_type)){
4121
                            mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4122
                            mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4123
                        }else{
4124
                            assert(IS_SUB_4X4(sub_mb_type));
4125
                            mv_cache[ 0 ][0]= mx;
4126
                            mv_cache[ 0 ][1]= my;
4127
                        }
4128
                    }
4129
                }else{
4130
                    uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4131
                    p[0] = p[1]=
4132
                    p[8] = p[9]= 0;
4133
                }
4134
            }
4135
        }
4136
    }else if(IS_DIRECT(mb_type)){
4137
        pred_direct_motion(h, &mb_type);
4138
        s->current_picture.mb_type[mb_xy]= mb_type;
4139
    }else{
4140
        int list, mx, my, i;
4141
         //FIXME we should set ref_idx_l? to 0 if we use that later ...
4142
        if(IS_16X16(mb_type)){
4143
            for(list=0; list<2; list++){
4144
                if(h->ref_count[list]>0){
4145
                    if(IS_DIR(mb_type, 0, list)){
4146
                        const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4147
                        fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
4148
                    }
4149
                }
4150
            }
4151
            for(list=0; list<2; list++){
4152
                if(IS_DIR(mb_type, 0, list)){
4153
                    pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
4154
                    mx += get_se_golomb(&s->gb);
4155
                    my += get_se_golomb(&s->gb);
4156
                    tprintf("final mv:%d %d\n", mx, my);
4157

    
4158
                    fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
4159
                }
4160
            }
4161
        }
4162
        else if(IS_16X8(mb_type)){
4163
            for(list=0; list<2; list++){
4164
                if(h->ref_count[list]>0){
4165
                    for(i=0; i<2; i++){
4166
                        if(IS_DIR(mb_type, i, list)){
4167
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4168
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
4169
                        }else // needed only for mixed refs (e.g. B_L0_L1_16x8)
4170
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
4171
                    }
4172
                }
4173
            }
4174
            for(list=0; list<2; list++){
4175
                for(i=0; i<2; i++){
4176
                    if(IS_DIR(mb_type, i, list)){
4177
                        pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
4178
                        mx += get_se_golomb(&s->gb);
4179
                        my += get_se_golomb(&s->gb);
4180
                        tprintf("final mv:%d %d\n", mx, my);
4181

    
4182
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
4183
                    }else
4184
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
4185
                }
4186
            }
4187
        }else{
4188
            assert(IS_8X16(mb_type));
4189
            for(list=0; list<2; list++){
4190
                if(h->ref_count[list]>0){
4191
                    for(i=0; i<2; i++){
4192
                        if(IS_DIR(mb_type, i, list)){ //FIXME optimize
4193
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4194
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
4195
                        }else // needed only for mixed refs
4196
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
4197
                    }
4198
                }
4199
            }
4200
            for(list=0; list<2; list++){
4201
                for(i=0; i<2; i++){
4202
                    if(IS_DIR(mb_type, i, list)){
4203
                        pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
4204
                        mx += get_se_golomb(&s->gb);
4205
                        my += get_se_golomb(&s->gb);
4206
                        tprintf("final mv:%d %d\n", mx, my);
4207

    
4208
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
4209
                    }else
4210
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
4211
                }
4212
            }
4213
        }
4214
    }
4215
    
4216
    if(IS_INTER(mb_type))
4217
        write_back_motion(h, mb_type);
4218
    
4219
    if(!IS_INTRA16x16(mb_type)){
4220
        cbp= get_ue_golomb(&s->gb);
4221
        if(cbp > 47){
4222
            av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
4223
            return -1;
4224
        }
4225
        
4226
        if(IS_INTRA4x4(mb_type))
4227
            cbp= golomb_to_intra4x4_cbp[cbp];
4228
        else
4229
            cbp= golomb_to_inter_cbp[cbp];
4230
    }
4231

    
4232
    if(cbp || IS_INTRA16x16(mb_type)){
4233
        int i8x8, i4x4, chroma_idx;
4234
        int chroma_qp, dquant;
4235
        GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
4236
        const uint8_t *scan, *dc_scan;
4237
        
4238
//        fill_non_zero_count_cache(h);
4239

    
4240
        if(IS_INTERLACED(mb_type)){
4241
            scan= field_scan;
4242
            dc_scan= luma_dc_field_scan;
4243
        }else{
4244
            scan= zigzag_scan;
4245
            dc_scan= luma_dc_zigzag_scan;
4246
        }
4247

    
4248
        dquant= get_se_golomb(&s->gb);
4249

    
4250
        if( dquant > 25 || dquant < -26 ){
4251
            av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
4252
            return -1;
4253
        }
4254
        
4255
        s->qscale += dquant;
4256
        if(((unsigned)s->qscale) > 51){
4257
            if(s->qscale<0) s->qscale+= 52;
4258
            else            s->qscale-= 52;
4259
        }
4260
        
4261
        h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
4262
        if(IS_INTRA16x16(mb_type)){
4263
            if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
4264
                return -1; //FIXME continue if partotioned and other retirn -1 too
4265
            }
4266

    
4267
            assert((cbp&15) == 0 || (cbp&15) == 15);
4268

    
4269
            if(cbp&15){
4270
                for(i8x8=0; i8x8<4; i8x8++){
4271
                    for(i4x4=0; i4x4<4; i4x4++){
4272
                        const int index= i4x4 + 4*i8x8;
4273
                        if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
4274
                            return -1;
4275
                        }
4276
                    }
4277
                }
4278
            }else{
4279
                fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
4280
            }
4281
        }else{
4282
            for(i8x8=0; i8x8<4; i8x8++){
4283
                if(cbp & (1<<i8x8)){
4284
                    for(i4x4=0; i4x4<4; i4x4++){
4285
                        const int index= i4x4 + 4*i8x8;
4286
                        
4287
                        if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
4288
                            return -1;
4289
                        }
4290
                    }
4291
                }else{
4292
                    uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4293
                    nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
4294
                }
4295
            }
4296
        }
4297
        
4298
        if(cbp&0x30){
4299
            for(chroma_idx=0; chroma_idx<2; chroma_idx++)
4300
                if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
4301
                    return -1;
4302
                }
4303
        }
4304

    
4305
        if(cbp&0x20){
4306
            for(chroma_idx=0; chroma_idx<2; chroma_idx++){
4307
                for(i4x4=0; i4x4<4; i4x4++){
4308
                    const int index= 16 + 4*chroma_idx + i4x4;
4309
                    if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
4310
                        return -1;
4311
                    }
4312
                }
4313
            }
4314
        }else{
4315
            uint8_t * const nnz= &h->non_zero_count_cache[0];
4316
            nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4317
            nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4318
        }
4319
    }else{
4320
        uint8_t * const nnz= &h->non_zero_count_cache[0];
4321
        fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
4322
        nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4323
        nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4324
    }
4325
    s->current_picture.qscale_table[mb_xy]= s->qscale;
4326
    write_back_non_zero_count(h);
4327

    
4328
    return 0;
4329
}
4330

    
4331
static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {
4332
    uint8_t *state= &h->cabac_state[ctx_base];
4333
    int mb_type;
4334
    
4335
    if(intra_slice){
4336
        MpegEncContext * const s = &h->s;
4337
        const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4338
        int ctx=0;
4339
        if( s->mb_x > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-1] ) )
4340
            ctx++;
4341
        if( s->mb_y > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
4342
            ctx++;
4343
        if( get_cabac( &h->cabac, &state[ctx] ) == 0 )
4344
            return 0;   /* I4x4 */
4345
        state += 2;
4346
    }else{
4347
        if( get_cabac( &h->cabac, &state[0] ) == 0 )
4348
            return 0;   /* I4x4 */
4349
    }
4350

    
4351
    if( get_cabac_terminate( &h->cabac ) )
4352
        return 25;  /* PCM */
4353

    
4354
    mb_type = 1; /* I16x16 */
4355
    if( get_cabac( &h->cabac, &state[1] ) )
4356
        mb_type += 12;  /* cbp_luma != 0 */
4357

    
4358
    if( get_cabac( &h->cabac, &state[2] ) ) {
4359
        if( get_cabac( &h->cabac, &state[2+intra_slice] ) )
4360
            mb_type += 4 * 2;   /* cbp_chroma == 2 */
4361
        else
4362
            mb_type += 4 * 1;   /* cbp_chroma == 1 */
4363
    }
4364
    if( get_cabac( &h->cabac, &state[3+intra_slice] ) )
4365
        mb_type += 2;
4366
    if( get_cabac( &h->cabac, &state[3+2*intra_slice] ) )
4367
        mb_type += 1;
4368
    return mb_type;
4369
}
4370

    
4371
static int decode_cabac_mb_type( H264Context *h ) {
4372
    MpegEncContext * const s = &h->s;
4373

    
4374
    if( h->slice_type == I_TYPE ) {
4375
        return decode_cabac_intra_mb_type(h, 3, 1);
4376
    } else if( h->slice_type == P_TYPE ) {
4377
        if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
4378
            /* P-type */
4379
            if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
4380
                if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
4381
                    return 0; /* P_L0_D16x16; */
4382
                else
4383
                    return 3; /* P_8x8; */
4384
            } else {
4385
                if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
4386
                    return 2; /* P_L0_D8x16; */
4387
                else
4388
                    return 1; /* P_L0_D16x8; */
4389
            }
4390
        } else {
4391
            return decode_cabac_intra_mb_type(h, 17, 0) + 5;
4392
        }
4393
    } else if( h->slice_type == B_TYPE ) {
4394
        const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4395
        int ctx = 0;
4396
        int bits;
4397

    
4398
        if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mb_xy-1] )
4399
                      && !IS_DIRECT( s->current_picture.mb_type[mb_xy-1] ) )
4400
            ctx++;
4401
        if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mb_xy-s->mb_stride] )
4402
                      && !IS_DIRECT( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
4403
            ctx++;
4404

    
4405
        if( !get_cabac( &h->cabac, &h->cabac_state[27+ctx] ) )
4406
            return 0; /* B_Direct_16x16 */
4407

    
4408
        if( !get_cabac( &h->cabac, &h->cabac_state[27+3] ) ) {
4409
            return 1 + get_cabac( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */
4410
        }
4411

    
4412
        bits = get_cabac( &h->cabac, &h->cabac_state[27+4] ) << 3;
4413
        bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 2;
4414
        bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 1;
4415
        bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] );
4416
        if( bits < 8 )
4417
            return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
4418
        else if( bits == 13 ) {
4419
            return decode_cabac_intra_mb_type(h, 32, 0) + 23;
4420
        } else if( bits == 14 )
4421
            return 11; /* B_L1_L0_8x16 */
4422
        else if( bits == 15 )
4423
            return 22; /* B_8x8 */
4424

    
4425
        bits= ( bits<<1 ) | get_cabac( &h->cabac, &h->cabac_state[27+5] );
4426
        return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */
4427
    } else {
4428
        /* TODO SI/SP frames? */
4429
        return -1;
4430
    }
4431
}
4432

    
4433
static int decode_cabac_mb_skip( H264Context *h) {
4434
    MpegEncContext * const s = &h->s;
4435
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4436
    const int mba_xy = mb_xy - 1;
4437
    const int mbb_xy = mb_xy - s->mb_stride;
4438
    int ctx = 0;
4439

    
4440
    if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mba_xy] ) )
4441
        ctx++;
4442
    if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ) )
4443
        ctx++;
4444

    
4445
    if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
4446
        return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
4447
    else /* B-frame */
4448
        return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
4449
}
4450

    
4451
static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
4452
    int mode = 0;
4453

    
4454
    if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
4455
        return pred_mode;
4456

    
4457
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4458
        mode += 1;
4459
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4460
        mode += 2;
4461
    if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4462
        mode += 4;
4463
    if( mode >= pred_mode )
4464
        return mode + 1;
4465
    else
4466
        return mode;
4467
}
4468

    
4469
static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
4470
    MpegEncContext * const s = &h->s;
4471
    const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4472
    const int mba_xy = mb_xy - 1;
4473
    const int mbb_xy = mb_xy - s->mb_stride;
4474

    
4475
    int ctx = 0;
4476

    
4477
    /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
4478
    if( s->mb_x > 0 && h->chroma_pred_mode_table[mba_xy] != 0 )
4479
        ctx++;
4480

    
4481
    if( s->mb_y > 0 && h-&