Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 7bc9090a

History | View | Annotate | Download (149 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
#undef NDEBUG
35
#include <assert.h>
36

    
37
#define interlaced_dct interlaced_dct_is_a_bad_name
38
#define mb_intra mb_intra_isnt_initalized_see_mb_type
39

    
40
#define LUMA_DC_BLOCK_INDEX   25
41
#define CHROMA_DC_BLOCK_INDEX 26
42

    
43
#define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
44
#define COEFF_TOKEN_VLC_BITS           8
45
#define TOTAL_ZEROS_VLC_BITS           9
46
#define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
47
#define RUN_VLC_BITS                   3
48
#define RUN7_VLC_BITS                  6
49

    
50
#define MAX_SPS_COUNT 32
51
#define MAX_PPS_COUNT 256
52

    
53
#define MAX_MMCO_COUNT 66
54

    
55
/**
56
 * Sequence parameter set
57
 */
58
typedef struct SPS{
59
    
60
    int profile_idc;
61
    int level_idc;
62
    int multiple_slice_groups;         ///< more_than_one_slice_group_allowed_flag
63
    int arbitrary_slice_order;         ///< arbitrary_slice_order_allowed_flag
64
    int redundant_slices;              ///< redundant_slices_allowed_flag
65
    int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
66
    int poc_type;                      ///< pic_order_cnt_type
67
    int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
68
    int delta_pic_order_always_zero_flag;
69
    int offset_for_non_ref_pic;
70
    int offset_for_top_to_bottom_field;
71
    int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
72
    int ref_frame_count;               ///< num_ref_frames
73
    int required_frame_num_update_behaviour_flag;
74
    int mb_width;                      ///< frame_width_in_mbs_minus1 + 1
75
    int mb_height;                     ///< frame_height_in_mbs_minus1 + 1
76
    int frame_mbs_only_flag;
77
    int mb_aff;                        ///<mb_adaptive_frame_field_flag
78
    int direct_8x8_inference_flag;
79
    int vui_parameters_present_flag;
80
    int sar_width;
81
    int sar_height;
82
    short offset_for_ref_frame[256]; //FIXME dyn aloc?
83
}SPS;
84

    
85
/**
86
 * Picture parameter set
87
 */
88
typedef struct PPS{
89
    int sps_id;
90
    int cabac;                  ///< entropy_coding_mode_flag
91
    int pic_order_present;      ///< pic_order_present_flag
92
    int slice_group_count;      ///< num_slice_groups_minus1 + 1
93
    int mb_slice_group_map_type;
94
    int ref_count[2];           ///< num_ref_idx_l0/1_active_minus1 + 1
95
    int weighted_pred;          ///< weighted_pred_flag
96
    int weighted_bipred_idc;
97
    int init_qp;                ///< pic_init_qp_minus26 + 26
98
    int init_qs;                ///< pic_init_qs_minus26 + 26
99
    int chroma_qp_index_offset;
100
    int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
101
    int constrained_intra_pred; ///< constrained_intra_pred_flag
102
    int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
103
    int crop;                   ///< frame_cropping_flag
104
    int crop_left;              ///< frame_cropping_rect_left_offset
105
    int crop_right;             ///< frame_cropping_rect_right_offset
106
    int crop_top;               ///< frame_cropping_rect_top_offset
107
    int crop_bottom;            ///< frame_cropping_rect_bottom_offset
108
}PPS;
109

    
110
/**
111
 * Memory management control operation opcode.
112
 */
113
typedef enum MMCOOpcode{
114
    MMCO_END=0,
115
    MMCO_SHORT2UNUSED,
116
    MMCO_LONG2UNUSED,
117
    MMCO_SHORT2LONG,
118
    MMCO_SET_MAX_LONG,
119
    MMCO_RESET, 
120
    MMCO_LONG,
121
} MMCOOpcode;
122

    
123
/**
124
 * Memory management control operation.
125
 */
126
typedef struct MMCO{
127
    MMCOOpcode opcode;
128
    int short_frame_num;
129
    int long_index;
130
} MMCO;
131

    
132
/**
133
 * H264Context
134
 */
135
typedef struct H264Context{
136
    MpegEncContext s;
137
    int nal_ref_idc;        
138
    int nal_unit_type;
139
#define NAL_SLICE                1
140
#define NAL_DPA                        2
141
#define NAL_DPB                        3
142
#define NAL_DPC                        4
143
#define NAL_IDR_SLICE                5
144
#define NAL_SEI                        6
145
#define NAL_SPS                        7
146
#define NAL_PPS                        8
147
#define NAL_PICTURE_DELIMITER        9
148
#define NAL_FILTER_DATA                10
149
    uint8_t *rbsp_buffer;
150
    int rbsp_buffer_size;
151

    
152
    int chroma_qp; //QPc
153

    
154
    int prev_mb_skiped; //FIXME remove (IMHO not used)
155

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

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

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

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

    
198
    SPS sps_buffer[MAX_SPS_COUNT];
199
    SPS sps; ///< current sps
200
    
201
    PPS pps_buffer[MAX_PPS_COUNT];
202
    /**
203
     * current pps
204
     */
205
    PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
206

    
207
    int slice_num;
208
    uint8_t *slice_table_base;
209
    uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
210
    int slice_type;
211
    int slice_type_fixed;
212
    
213
    //interlacing specific flags
214
    int mb_field_decoding_flag;
215
    
216
    int sub_mb_type[4];
217
    
218
    //POC stuff
219
    int poc_lsb;
220
    int poc_msb;
221
    int delta_poc_bottom;
222
    int delta_poc[2];
223
    int frame_num;
224
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
225
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
226
    int frame_num_offset;         ///< for POC type 2
227
    int prev_frame_num_offset;    ///< for POC type 2
228
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
229

    
230
    /**
231
     * frame_num for frames or 2*frame_num for field pics.
232
     */
233
    int curr_pic_num;
234
    
235
    /**
236
     * max_frame_num or 2*max_frame_num for field pics.
237
     */
238
    int max_pic_num;
239

    
240
    //Weighted pred stuff
241
    int luma_log2_weight_denom;
242
    int chroma_log2_weight_denom;
243
    int luma_weight[2][16];
244
    int luma_offset[2][16];
245
    int chroma_weight[2][16][2];
246
    int chroma_offset[2][16][2];
247
   
248
    //deblock
249
    int disable_deblocking_filter_idc;
250
    int slice_alpha_c0_offset_div2;
251
    int slice_beta_offset_div2;
252
     
253
    int redundant_pic_count;
254
    
255
    int direct_spatial_mv_pred;
256

    
257
    /**
258
     * num_ref_idx_l0/1_active_minus1 + 1
259
     */
260
    int ref_count[2];// FIXME split for AFF
261
    Picture *short_ref[16];
262
    Picture *long_ref[16];
263
    Picture default_ref_list[2][32];
264
    Picture ref_list[2][32]; //FIXME size?
265
    Picture field_ref_list[2][32]; //FIXME size?
266
    
267
    /**
268
     * memory management control operations buffer.
269
     */
270
    MMCO mmco[MAX_MMCO_COUNT];
271
    int mmco_index;
272
    
273
    int long_ref_count;  ///< number of actual long term references
274
    int short_ref_count; ///< number of actual short term references
275
    
276
    //data partitioning
277
    GetBitContext intra_gb;
278
    GetBitContext inter_gb;
279
    GetBitContext *intra_gb_ptr;
280
    GetBitContext *inter_gb_ptr;
281
    
282
    DCTELEM mb[16*24] __align8;
283
}H264Context;
284

    
285
static VLC coeff_token_vlc[4];
286
static VLC chroma_dc_coeff_token_vlc;
287

    
288
static VLC total_zeros_vlc[15];
289
static VLC chroma_dc_total_zeros_vlc[3];
290

    
291
static VLC run_vlc[6];
292
static VLC run7_vlc;
293

    
294
/**
295
 * fill a rectangle.
296
 * @param h height of the recatangle, should be a constant
297
 * @param w width of the recatangle, should be a constant
298
 * @param size the size of val (1 or 4), should be a constant
299
 */
300
static inline void fill_rectangle(void *p, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
301
    assert(size==1 || size==4);
302
    
303
    w      *= size;
304
    stride *= size;
305
    
306
//FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
307
    if(w==2 && h==2){
308
        *(uint16_t*)(p + 0)=
309
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
310
    }else if(w==2 && h==4){
311
        *(uint16_t*)(p + 0*stride)=
312
        *(uint16_t*)(p + 1*stride)=
313
        *(uint16_t*)(p + 2*stride)=
314
        *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
315
    }else if(w==4 && h==2){
316
        *(uint32_t*)(p + 0*stride)=
317
        *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
318
    }else if(w==4 && h==4){
319
        *(uint32_t*)(p + 0*stride)=
320
        *(uint32_t*)(p + 1*stride)=
321
        *(uint32_t*)(p + 2*stride)=
322
        *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
323
    }else if(w==8 && h==1){
324
        *(uint32_t*)(p + 0)=
325
        *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
326
    }else if(w==8 && h==2){
327
        *(uint32_t*)(p + 0 + 0*stride)=
328
        *(uint32_t*)(p + 4 + 0*stride)=
329
        *(uint32_t*)(p + 0 + 1*stride)=
330
        *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
331
    }else if(w==8 && h==4){
332
        *(uint64_t*)(p + 0*stride)=
333
        *(uint64_t*)(p + 1*stride)=
334
        *(uint64_t*)(p + 2*stride)=
335
        *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
336
    }else if(w==16 && h==2){
337
        *(uint64_t*)(p + 0+0*stride)=
338
        *(uint64_t*)(p + 8+0*stride)=
339
        *(uint64_t*)(p + 0+1*stride)=
340
        *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
341
    }else if(w==16 && h==4){
342
        *(uint64_t*)(p + 0+0*stride)=
343
        *(uint64_t*)(p + 8+0*stride)=
344
        *(uint64_t*)(p + 0+1*stride)=
345
        *(uint64_t*)(p + 8+1*stride)=
346
        *(uint64_t*)(p + 0+2*stride)=
347
        *(uint64_t*)(p + 8+2*stride)=
348
        *(uint64_t*)(p + 0+3*stride)=
349
        *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
350
    }else
351
        assert(0);
352
}
353

    
354
static inline void fill_caches(H264Context *h, int mb_type){
355
    MpegEncContext * const s = &h->s;
356
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
357
    int topleft_xy, top_xy, topright_xy, left_xy[2];
358
    int topleft_type, top_type, topright_type, left_type[2];
359
    int left_block[4];
360
    int i;
361

    
362
    //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it 
363
    
364
    if(h->sps.mb_aff){
365
    //FIXME
366
    }else{
367
        topleft_xy = mb_xy-1 - s->mb_stride;
368
        top_xy     = mb_xy   - s->mb_stride;
369
        topright_xy= mb_xy+1 - s->mb_stride;
370
        left_xy[0]   = mb_xy-1;
371
        left_xy[1]   = mb_xy-1;
372
        left_block[0]= 0;
373
        left_block[1]= 1;
374
        left_block[2]= 2;
375
        left_block[3]= 3;
376
    }
377

    
378
    topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
379
    top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
380
    topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
381
    left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
382
    left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
383

    
384
    if(IS_INTRA(mb_type)){
385
        h->topleft_samples_available= 
386
        h->top_samples_available= 
387
        h->left_samples_available= 0xFFFF;
388
        h->topright_samples_available= 0xEEEA;
389

    
390
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
391
            h->topleft_samples_available= 0xB3FF;
392
            h->top_samples_available= 0x33FF;
393
            h->topright_samples_available= 0x26EA;
394
        }
395
        for(i=0; i<2; i++){
396
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
397
                h->topleft_samples_available&= 0xDF5F;
398
                h->left_samples_available&= 0x5F5F;
399
            }
400
        }
401
        
402
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
403
            h->topleft_samples_available&= 0x7FFF;
404
        
405
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
406
            h->topright_samples_available&= 0xFBFF;
407
    
408
        if(IS_INTRA4x4(mb_type)){
409
            if(IS_INTRA4x4(top_type)){
410
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
411
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
412
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
413
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
414
            }else{
415
                int pred;
416
                if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
417
                    pred= 2;
418
                else{
419
                    pred= -1;
420
                }
421
                h->intra4x4_pred_mode_cache[4+8*0]=
422
                h->intra4x4_pred_mode_cache[5+8*0]=
423
                h->intra4x4_pred_mode_cache[6+8*0]=
424
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
425
            }
426
            for(i=0; i<2; i++){
427
                if(IS_INTRA4x4(left_type[i])){
428
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
429
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
430
                }else{
431
                    int pred;
432
                    if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
433
                        pred= 2;
434
                    else{
435
                        pred= -1;
436
                    }
437
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
438
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
439
                }
440
            }
441
        }
442
    }
443
    
444
    
445
/*
446
0 . T T. T T T T 
447
1 L . .L . . . . 
448
2 L . .L . . . . 
449
3 . T TL . . . . 
450
4 L . .L . . . . 
451
5 L . .. . . . . 
452
*/
453
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
454
    if(top_type){
455
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
456
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
457
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
458
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
459
    
460
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
461
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
462
    
463
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
464
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
465
    }else{
466
        h->non_zero_count_cache[4+8*0]=      
467
        h->non_zero_count_cache[5+8*0]=
468
        h->non_zero_count_cache[6+8*0]=
469
        h->non_zero_count_cache[7+8*0]=
470
    
471
        h->non_zero_count_cache[1+8*0]=
472
        h->non_zero_count_cache[2+8*0]=
473
    
474
        h->non_zero_count_cache[1+8*3]=
475
        h->non_zero_count_cache[2+8*3]= 64;
476
    }
477
    
478
    if(left_type[0]){
479
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
480
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
481
        h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
482
        h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
483
    }else{
484
        h->non_zero_count_cache[3+8*1]= 
485
        h->non_zero_count_cache[3+8*2]= 
486
        h->non_zero_count_cache[0+8*1]= 
487
        h->non_zero_count_cache[0+8*4]= 64;
488
    }
489
    
490
    if(left_type[1]){
491
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
492
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
493
        h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
494
        h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
495
    }else{
496
        h->non_zero_count_cache[3+8*3]= 
497
        h->non_zero_count_cache[3+8*4]= 
498
        h->non_zero_count_cache[0+8*2]= 
499
        h->non_zero_count_cache[0+8*5]= 64;
500
    }
501
    
502
#if 1
503
    if(IS_INTER(mb_type)){
504
        int list;
505
        for(list=0; list<2; list++){
506
            if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
507
                /*if(!h->mv_cache_clean[list]){
508
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
509
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
510
                    h->mv_cache_clean[list]= 1;
511
                }*/
512
                continue; //FIXME direct mode ...
513
            }
514
            h->mv_cache_clean[list]= 0;
515
            
516
            if(IS_INTER(topleft_type)){
517
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
518
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
519
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
520
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
521
            }else{
522
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
523
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
524
            }
525
            
526
            if(IS_INTER(top_type)){
527
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
528
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
529
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
530
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
531
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
532
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
533
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
534
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
535
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
536
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
537
            }else{
538
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
539
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
540
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
541
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
542
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
543
            }
544

    
545
            if(IS_INTER(topright_type)){
546
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
547
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
548
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
549
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
550
            }else{
551
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
552
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
553
            }
554
            
555
            //FIXME unify cleanup or sth
556
            if(IS_INTER(left_type[0])){
557
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
558
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
559
                *(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]];
560
                *(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]];
561
                h->ref_cache[list][scan8[0] - 1 + 0*8]= 
562
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
563
            }else{
564
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
565
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
566
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
567
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
568
            }
569
            
570
            if(IS_INTER(left_type[1])){
571
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
572
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
573
                *(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]];
574
                *(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]];
575
                h->ref_cache[list][scan8[0] - 1 + 2*8]= 
576
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
577
            }else{
578
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
579
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
580
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
581
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
582
            }
583

    
584
            h->ref_cache[list][scan8[5 ]+1] = 
585
            h->ref_cache[list][scan8[7 ]+1] = 
586
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewher else)
587
            h->ref_cache[list][scan8[4 ]] = 
588
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
589
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
590
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
591
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
592
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
593
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
594
        }
595
//FIXME
596

    
597
    }
598
#endif
599
}
600

    
601
static inline void write_back_intra_pred_mode(H264Context *h){
602
    MpegEncContext * const s = &h->s;
603
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
604

    
605
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
606
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
607
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
608
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
609
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
610
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
611
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
612
}
613

    
614
/**
615
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
616
 */
617
static inline int check_intra4x4_pred_mode(H264Context *h){
618
    MpegEncContext * const s = &h->s;
619
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
620
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
621
    int i;
622
    
623
    if(!(h->top_samples_available&0x8000)){
624
        for(i=0; i<4; i++){
625
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
626
            if(status<0){
627
                fprintf(stderr, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
628
                return -1;
629
            } else if(status){
630
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
631
            }
632
        }
633
    }
634
    
635
    if(!(h->left_samples_available&0x8000)){
636
        for(i=0; i<4; i++){
637
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
638
            if(status<0){
639
                fprintf(stderr, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
640
                return -1;
641
            } else if(status){
642
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
643
            }
644
        }
645
    }
646

    
647
    return 0;
648
} //FIXME cleanup like next
649

    
650
/**
651
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
652
 */
653
static inline int check_intra_pred_mode(H264Context *h, int mode){
654
    MpegEncContext * const s = &h->s;
655
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
656
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
657
    
658
    if(!(h->top_samples_available&0x8000)){
659
        mode= top[ mode ];
660
        if(mode<0){
661
            fprintf(stderr, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
662
            return -1;
663
        }
664
    }
665
    
666
    if(!(h->left_samples_available&0x8000)){
667
        mode= left[ mode ];
668
        if(mode<0){
669
            fprintf(stderr, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
670
            return -1;
671
        } 
672
    }
673

    
674
    return mode;
675
}
676

    
677
/**
678
 * gets the predicted intra4x4 prediction mode.
679
 */
680
static inline int pred_intra_mode(H264Context *h, int n){
681
    const int index8= scan8[n];
682
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
683
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
684
    const int min= FFMIN(left, top);
685

    
686
    tprintf("mode:%d %d min:%d\n", left ,top, min);
687

    
688
    if(min<0) return DC_PRED;
689
    else      return min;
690
}
691

    
692
static inline void write_back_non_zero_count(H264Context *h){
693
    MpegEncContext * const s = &h->s;
694
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
695

    
696
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
697
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
698
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
699
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
700
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
701
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
702
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
703
    
704
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
705
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
706
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
707

    
708
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
709
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
710
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
711
}
712

    
713
/**
714
 * gets the predicted number of non zero coefficients.
715
 * @param n block index
716
 */
717
static inline int pred_non_zero_count(H264Context *h, int n){
718
    const int index8= scan8[n];
719
    const int left= h->non_zero_count_cache[index8 - 1];
720
    const int top = h->non_zero_count_cache[index8 - 8];
721
    int i= left + top;
722
    
723
    if(i<64) i= (i+1)>>1;
724

    
725
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
726

    
727
    return i&31;
728
}
729

    
730
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
731
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
732

    
733
    if(topright_ref != PART_NOT_AVAILABLE){
734
        *C= h->mv_cache[list][ i - 8 + part_width ];
735
        return topright_ref;
736
    }else{
737
        tprintf("topright MV not available\n");
738

    
739
        *C= h->mv_cache[list][ i - 8 - 1 ];
740
        return h->ref_cache[list][ i - 8 - 1 ];
741
    }
742
}
743

    
744
/**
745
 * gets the predicted MV.
746
 * @param n the block index
747
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
748
 * @param mx the x component of the predicted motion vector
749
 * @param my the y component of the predicted motion vector
750
 */
751
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
752
    MpegEncContext * const s = &h->s;
753
    const int index8= scan8[n];
754
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
755
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
756
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
757
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
758
    const int16_t * C;
759
    int diagonal_ref, match_count;
760

    
761
    assert(part_width==1 || part_width==2 || part_width==4);
762

    
763
/* mv_cache
764
  B . . A T T T T 
765
  U . . L . . , .
766
  U . . L . . . .
767
  U . . L . . , .
768
  . . . L . . . .
769
*/
770

    
771
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
772
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
773
    
774
    if(match_count > 1){ //most common
775
        *mx= mid_pred(A[0], B[0], C[0]);
776
        *my= mid_pred(A[1], B[1], C[1]);
777
    }else if(match_count==1){
778
        if(left_ref==ref){
779
            *mx= A[0];
780
            *my= A[1];        
781
        }else if(top_ref==ref){
782
            *mx= B[0];
783
            *my= B[1];        
784
        }else{
785
            *mx= C[0];
786
            *my= C[1];        
787
        }
788
    }else{
789
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
790
            *mx= A[0];
791
            *my= A[1];        
792
        }else{
793
            *mx= mid_pred(A[0], B[0], C[0]);
794
            *my= mid_pred(A[1], B[1], C[1]);
795
        }
796
    }
797
        
798
    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, s->mb_x, s->mb_y, n, list);
799
}
800

    
801
/**
802
 * gets the directionally predicted 16x8 MV.
803
 * @param n the block index
804
 * @param mx the x component of the predicted motion vector
805
 * @param my the y component of the predicted motion vector
806
 */
807
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
808
    MpegEncContext * const s = &h->s;
809
    if(n==0){
810
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
811
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
812

    
813
        tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", top_ref, B[0], B[1], s->mb_x, s->mb_y, n, list);
814
        
815
        if(top_ref == ref){
816
            *mx= B[0];
817
            *my= B[1];
818
            return;
819
        }
820
    }else{
821
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
822
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
823
        
824
        tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], s->mb_x, s->mb_y, n, list);
825

    
826
        if(left_ref == ref){
827
            *mx= A[0];
828
            *my= A[1];
829
            return;
830
        }
831
    }
832

    
833
    //RARE
834
    pred_motion(h, n, 4, list, ref, mx, my);
835
}
836

    
837
/**
838
 * gets the directionally predicted 8x16 MV.
839
 * @param n the block index
840
 * @param mx the x component of the predicted motion vector
841
 * @param my the y component of the predicted motion vector
842
 */
843
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
844
    MpegEncContext * const s = &h->s;
845
    if(n==0){
846
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
847
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
848
        
849
        tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], s->mb_x, s->mb_y, n, list);
850

    
851
        if(left_ref == ref){
852
            *mx= A[0];
853
            *my= A[1];
854
            return;
855
        }
856
    }else{
857
        const int16_t * C;
858
        int diagonal_ref;
859

    
860
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
861
        
862
        tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", diagonal_ref, C[0], C[1], s->mb_x, s->mb_y, n, list);
863

    
864
        if(diagonal_ref == ref){ 
865
            *mx= C[0];
866
            *my= C[1];
867
            return;
868
        }
869
    }
870

    
871
    //RARE
872
    pred_motion(h, n, 2, list, ref, mx, my);
873
}
874

    
875
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
876
    MpegEncContext * const s = &h->s;
877
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
878
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
879

    
880
    tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, s->mb_x, s->mb_y);
881

    
882
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
883
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
884
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
885
       
886
        *mx = *my = 0;
887
        return;
888
    }
889
        
890
    pred_motion(h, 0, 4, 0, 0, mx, my);
891

    
892
    return;
893
}
894

    
895
static inline void write_back_motion(H264Context *h, int mb_type){
896
    MpegEncContext * const s = &h->s;
897
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
898
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
899
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
900
    int list;
901

    
902
    for(list=0; list<2; list++){
903
        int y;
904
        if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
905
            if(1){ //FIXME skip or never read if mb_type doesnt use it
906
                for(y=0; y<4; y++){
907
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
908
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
909
                }
910
                for(y=0; y<2; y++){
911
                    *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
912
                }
913
            }
914
            continue; //FIXME direct mode ...
915
        }
916
        
917
        for(y=0; y<4; y++){
918
            *(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];
919
            *(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];
920
        }
921
        for(y=0; y<2; y++){
922
            s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
923
            s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
924
        }
925
    }
926
}
927

    
928
/**
929
 * Decodes a network abstraction layer unit.
930
 * @param consumed is the number of bytes used as input
931
 * @param length is the length of the array
932
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
933
 * @returns decoded bytes, might be src+1 if no escapes 
934
 */
935
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
936
    int i, si, di;
937
    uint8_t *dst;
938

    
939
//    src[0]&0x80;                //forbidden bit
940
    h->nal_ref_idc= src[0]>>5;
941
    h->nal_unit_type= src[0]&0x1F;
942

    
943
    src++; length--;
944
#if 0    
945
    for(i=0; i<length; i++)
946
        printf("%2X ", src[i]);
947
#endif
948
    for(i=0; i+1<length; i+=2){
949
        if(src[i]) continue;
950
        if(i>0 && src[i-1]==0) i--;
951
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
952
            if(src[i+2]!=3){
953
                /* startcode, so we must be past the end */
954
                length=i;
955
            }
956
            break;
957
        }
958
    }
959

    
960
    if(i>=length-1){ //no escaped 0
961
        *dst_length= length;
962
        *consumed= length+1; //+1 for the header
963
        return src; 
964
    }
965

    
966
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
967
    dst= h->rbsp_buffer;
968

    
969
//printf("deoding esc\n");
970
    si=di=0;
971
    while(si<length){ 
972
        //remove escapes (very rare 1:2^22)
973
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
974
            if(src[si+2]==3){ //escape
975
                dst[di++]= 0;
976
                dst[di++]= 0;
977
                si+=3;
978
            }else //next start code
979
                break;
980
        }
981

    
982
        dst[di++]= src[si++];
983
    }
984

    
985
    *dst_length= di;
986
    *consumed= si + 1;//+1 for the header
987
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
988
    return dst;
989
}
990

    
991
/**
992
 * @param src the data which should be escaped
993
 * @param dst the target buffer, dst+1 == src is allowed as a special case
994
 * @param length the length of the src data
995
 * @param dst_length the length of the dst array
996
 * @returns length of escaped data in bytes or -1 if an error occured
997
 */
998
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
999
    int i, escape_count, si, di;
1000
    uint8_t *temp;
1001
    
1002
    assert(length>=0);
1003
    assert(dst_length>0);
1004
    
1005
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1006

    
1007
    if(length==0) return 1;
1008

    
1009
    escape_count= 0;
1010
    for(i=0; i<length; i+=2){
1011
        if(src[i]) continue;
1012
        if(i>0 && src[i-1]==0) 
1013
            i--;
1014
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1015
            escape_count++;
1016
            i+=2;
1017
        }
1018
    }
1019
    
1020
    if(escape_count==0){ 
1021
        if(dst+1 != src)
1022
            memcpy(dst+1, src, length);
1023
        return length + 1;
1024
    }
1025
    
1026
    if(length + escape_count + 1> dst_length)
1027
        return -1;
1028

    
1029
    //this should be damn rare (hopefully)
1030

    
1031
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1032
    temp= h->rbsp_buffer;
1033
//printf("encoding esc\n");
1034
    
1035
    si= 0;
1036
    di= 0;
1037
    while(si < length){
1038
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1039
            temp[di++]= 0; si++;
1040
            temp[di++]= 0; si++;
1041
            temp[di++]= 3; 
1042
            temp[di++]= src[si++];
1043
        }
1044
        else
1045
            temp[di++]= src[si++];
1046
    }
1047
    memcpy(dst+1, temp, length+escape_count);
1048
    
1049
    assert(di == length+escape_count);
1050
    
1051
    return di + 1;
1052
}
1053

    
1054
/**
1055
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1056
 */
1057
static void encode_rbsp_trailing(PutBitContext *pb){
1058
    int length;
1059
    put_bits(pb, 1, 1);
1060
    length= (-get_bit_count(pb))&7;
1061
    if(length) put_bits(pb, length, 0);
1062
}
1063

    
1064
/**
1065
 * identifies the exact end of the bitstream
1066
 * @return the length of the trailing, or 0 if damaged
1067
 */
1068
static int decode_rbsp_trailing(uint8_t *src){
1069
    int v= *src;
1070
    int r;
1071

    
1072
    tprintf("rbsp trailing %X\n", v);
1073

    
1074
    for(r=1; r<9; r++){
1075
        if(v&1) return r;
1076
        v>>=1;
1077
    }
1078
    return 0;
1079
}
1080

    
1081
/**
1082
 * idct tranforms the 16 dc values and dequantize them.
1083
 * @param qp quantization parameter
1084
 */
1085
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1086
    const int qmul= dequant_coeff[qp][0];
1087
#define stride 16
1088
    int i;
1089
    int temp[16]; //FIXME check if this is a good idea
1090
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1091
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1092

    
1093
//memset(block, 64, 2*256);
1094
//return;
1095
    for(i=0; i<4; i++){
1096
        const int offset= y_offset[i];
1097
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1098
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1099
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1100
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1101

    
1102
        temp[4*i+0]= z0+z3;
1103
        temp[4*i+1]= z1+z2;
1104
        temp[4*i+2]= z1-z2;
1105
        temp[4*i+3]= z0-z3;
1106
    }
1107

    
1108
    for(i=0; i<4; i++){
1109
        const int offset= x_offset[i];
1110
        const int z0= temp[4*0+i] + temp[4*2+i];
1111
        const int z1= temp[4*0+i] - temp[4*2+i];
1112
        const int z2= temp[4*1+i] - temp[4*3+i];
1113
        const int z3= temp[4*1+i] + temp[4*3+i];
1114

    
1115
        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1116
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1117
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1118
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1119
    }
1120
}
1121

    
1122
/**
1123
 * dct tranforms the 16 dc values.
1124
 * @param qp quantization parameter ??? FIXME
1125
 */
1126
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1127
//    const int qmul= dequant_coeff[qp][0];
1128
    int i;
1129
    int temp[16]; //FIXME check if this is a good idea
1130
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1131
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1132

    
1133
    for(i=0; i<4; i++){
1134
        const int offset= y_offset[i];
1135
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1136
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1137
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1138
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1139

    
1140
        temp[4*i+0]= z0+z3;
1141
        temp[4*i+1]= z1+z2;
1142
        temp[4*i+2]= z1-z2;
1143
        temp[4*i+3]= z0-z3;
1144
    }
1145

    
1146
    for(i=0; i<4; i++){
1147
        const int offset= x_offset[i];
1148
        const int z0= temp[4*0+i] + temp[4*2+i];
1149
        const int z1= temp[4*0+i] - temp[4*2+i];
1150
        const int z2= temp[4*1+i] - temp[4*3+i];
1151
        const int z3= temp[4*1+i] + temp[4*3+i];
1152

    
1153
        block[stride*0 +offset]= (z0 + z3)>>1;
1154
        block[stride*2 +offset]= (z1 + z2)>>1;
1155
        block[stride*8 +offset]= (z1 - z2)>>1;
1156
        block[stride*10+offset]= (z0 - z3)>>1;
1157
    }
1158
}
1159
#undef xStride
1160
#undef stride
1161

    
1162
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1163
    const int qmul= dequant_coeff[qp][0];
1164
    const int stride= 16*2;
1165
    const int xStride= 16;
1166
    int a,b,c,d,e;
1167

    
1168
    a= block[stride*0 + xStride*0];
1169
    b= block[stride*0 + xStride*1];
1170
    c= block[stride*1 + xStride*0];
1171
    d= block[stride*1 + xStride*1];
1172

    
1173
    e= a-b;
1174
    a= a+b;
1175
    b= c-d;
1176
    c= c+d;
1177

    
1178
    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1179
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1180
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1181
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1182
}
1183

    
1184
static void chroma_dc_dct_c(DCTELEM *block){
1185
    const int stride= 16*2;
1186
    const int xStride= 16;
1187
    int a,b,c,d,e;
1188

    
1189
    a= block[stride*0 + xStride*0];
1190
    b= block[stride*0 + xStride*1];
1191
    c= block[stride*1 + xStride*0];
1192
    d= block[stride*1 + xStride*1];
1193

    
1194
    e= a-b;
1195
    a= a+b;
1196
    b= c-d;
1197
    c= c+d;
1198

    
1199
    block[stride*0 + xStride*0]= (a+c);
1200
    block[stride*0 + xStride*1]= (e+b);
1201
    block[stride*1 + xStride*0]= (a-c);
1202
    block[stride*1 + xStride*1]= (e-b);
1203
}
1204

    
1205
/**
1206
 * gets the chroma qp.
1207
 */
1208
static inline int get_chroma_qp(H264Context *h, int qscale){
1209
    
1210
    return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1211
}
1212

    
1213

    
1214
/**
1215
 *
1216
 */
1217
static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1218
    int i;
1219
    uint8_t *cm = cropTbl + MAX_NEG_CROP;
1220

    
1221
    block[0] += 32;
1222
#if 1
1223
    for(i=0; i<4; i++){
1224
        const int z0=  block[i + 4*0]     +  block[i + 4*2];
1225
        const int z1=  block[i + 4*0]     -  block[i + 4*2];
1226
        const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
1227
        const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
1228

    
1229
        block[i + 4*0]= z0 + z3;
1230
        block[i + 4*1]= z1 + z2;
1231
        block[i + 4*2]= z1 - z2;
1232
        block[i + 4*3]= z0 - z3;
1233
    }
1234

    
1235
    for(i=0; i<4; i++){
1236
        const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
1237
        const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
1238
        const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
1239
        const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
1240

    
1241
        dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ];
1242
        dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ];
1243
        dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ];
1244
        dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ];
1245
    }
1246
#else
1247
    for(i=0; i<4; i++){
1248
        const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
1249
        const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
1250
        const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
1251
        const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
1252

    
1253
        block[0 + 4*i]= z0 + z3;
1254
        block[1 + 4*i]= z1 + z2;
1255
        block[2 + 4*i]= z1 - z2;
1256
        block[3 + 4*i]= z0 - z3;
1257
    }
1258

    
1259
    for(i=0; i<4; i++){
1260
        const int z0=  block[i + 4*0]     +  block[i + 4*2];
1261
        const int z1=  block[i + 4*0]     -  block[i + 4*2];
1262
        const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
1263
        const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
1264

    
1265
        dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1266
        dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1267
        dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1268
        dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1269
    }
1270
#endif
1271
}
1272

    
1273
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1274
    int i;
1275
    //FIXME try int temp instead of block
1276
    
1277
    for(i=0; i<4; i++){
1278
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1279
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1280
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1281
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1282
        const int z0= d0 + d3;
1283
        const int z3= d0 - d3;
1284
        const int z1= d1 + d2;
1285
        const int z2= d1 - d2;
1286
        
1287
        block[0 + 4*i]=   z0 +   z1;
1288
        block[1 + 4*i]= 2*z3 +   z2;
1289
        block[2 + 4*i]=   z0 -   z1;
1290
        block[3 + 4*i]=   z3 - 2*z2;
1291
    }    
1292

    
1293
    for(i=0; i<4; i++){
1294
        const int z0= block[0*4 + i] + block[3*4 + i];
1295
        const int z3= block[0*4 + i] - block[3*4 + i];
1296
        const int z1= block[1*4 + i] + block[2*4 + i];
1297
        const int z2= block[1*4 + i] - block[2*4 + i];
1298
        
1299
        block[0*4 + i]=   z0 +   z1;
1300
        block[1*4 + i]= 2*z3 +   z2;
1301
        block[2*4 + i]=   z0 -   z1;
1302
        block[3*4 + i]=   z3 - 2*z2;
1303
    }
1304
}
1305

    
1306
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1307
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1308
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1309
    int i;
1310
    const int * const quant_table= quant_coeff[qscale];
1311
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1312
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1313
    const unsigned int threshold2= (threshold1<<1);
1314
    int last_non_zero;
1315

    
1316
    if(seperate_dc){
1317
        if(qscale<=18){
1318
            //avoid overflows
1319
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1320
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1321
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1322

    
1323
            int level= block[0]*quant_coeff[qscale+18][0];
1324
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1325
                if(level>0){
1326
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1327
                    block[0]= level;
1328
                }else{
1329
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1330
                    block[0]= -level;
1331
                }
1332
//                last_non_zero = i;
1333
            }else{
1334
                block[0]=0;
1335
            }
1336
        }else{
1337
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1338
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1339
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1340

    
1341
            int level= block[0]*quant_table[0];
1342
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1343
                if(level>0){
1344
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1345
                    block[0]= level;
1346
                }else{
1347
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1348
                    block[0]= -level;
1349
                }
1350
//                last_non_zero = i;
1351
            }else{
1352
                block[0]=0;
1353
            }
1354
        }
1355
        last_non_zero= 0;
1356
        i=1;
1357
    }else{
1358
        last_non_zero= -1;
1359
        i=0;
1360
    }
1361

    
1362
    for(; i<16; i++){
1363
        const int j= scantable[i];
1364
        int level= block[j]*quant_table[j];
1365

    
1366
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1367
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1368
        if(((unsigned)(level+threshold1))>threshold2){
1369
            if(level>0){
1370
                level= (bias + level)>>QUANT_SHIFT;
1371
                block[j]= level;
1372
            }else{
1373
                level= (bias - level)>>QUANT_SHIFT;
1374
                block[j]= -level;
1375
            }
1376
            last_non_zero = i;
1377
        }else{
1378
            block[j]=0;
1379
        }
1380
    }
1381

    
1382
    return last_non_zero;
1383
}
1384

    
1385
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1386
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1387
    ((uint32_t*)(src+0*stride))[0]= a;
1388
    ((uint32_t*)(src+1*stride))[0]= a;
1389
    ((uint32_t*)(src+2*stride))[0]= a;
1390
    ((uint32_t*)(src+3*stride))[0]= a;
1391
}
1392

    
1393
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1394
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1395
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1396
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1397
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1398
}
1399

    
1400
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1401
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1402
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1403
    
1404
    ((uint32_t*)(src+0*stride))[0]= 
1405
    ((uint32_t*)(src+1*stride))[0]= 
1406
    ((uint32_t*)(src+2*stride))[0]= 
1407
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1408
}
1409

    
1410
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1411
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1412
    
1413
    ((uint32_t*)(src+0*stride))[0]= 
1414
    ((uint32_t*)(src+1*stride))[0]= 
1415
    ((uint32_t*)(src+2*stride))[0]= 
1416
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1417
}
1418

    
1419
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1420
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1421
    
1422
    ((uint32_t*)(src+0*stride))[0]= 
1423
    ((uint32_t*)(src+1*stride))[0]= 
1424
    ((uint32_t*)(src+2*stride))[0]= 
1425
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1426
}
1427

    
1428
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1429
    ((uint32_t*)(src+0*stride))[0]= 
1430
    ((uint32_t*)(src+1*stride))[0]= 
1431
    ((uint32_t*)(src+2*stride))[0]= 
1432
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1433
}
1434

    
1435

    
1436
#define LOAD_TOP_RIGHT_EDGE\
1437
    const int t4= topright[0];\
1438
    const int t5= topright[1];\
1439
    const int t6= topright[2];\
1440
    const int t7= topright[3];\
1441

    
1442
#define LOAD_LEFT_EDGE\
1443
    const int l0= src[-1+0*stride];\
1444
    const int l1= src[-1+1*stride];\
1445
    const int l2= src[-1+2*stride];\
1446
    const int l3= src[-1+3*stride];\
1447

    
1448
#define LOAD_TOP_EDGE\
1449
    const int t0= src[ 0-1*stride];\
1450
    const int t1= src[ 1-1*stride];\
1451
    const int t2= src[ 2-1*stride];\
1452
    const int t3= src[ 3-1*stride];\
1453

    
1454
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1455
    const int lt= src[-1-1*stride];
1456
    LOAD_TOP_EDGE
1457
    LOAD_LEFT_EDGE
1458

    
1459
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1460
    src[0+2*stride]=
1461
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1462
    src[0+1*stride]=
1463
    src[1+2*stride]=
1464
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1465
    src[0+0*stride]=
1466
    src[1+1*stride]=
1467
    src[2+2*stride]=
1468
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1469
    src[1+0*stride]=
1470
    src[2+1*stride]=
1471
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1472
    src[2+0*stride]=
1473
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1474
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1475
};
1476

    
1477
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1478
    LOAD_TOP_EDGE    
1479
    LOAD_TOP_RIGHT_EDGE    
1480
//    LOAD_LEFT_EDGE    
1481

    
1482
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1483
    src[1+0*stride]=
1484
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1485
    src[2+0*stride]=
1486
    src[1+1*stride]=
1487
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1488
    src[3+0*stride]=
1489
    src[2+1*stride]=
1490
    src[1+2*stride]=
1491
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1492
    src[3+1*stride]=
1493
    src[2+2*stride]=
1494
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1495
    src[3+2*stride]=
1496
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1497
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1498
};
1499

    
1500
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1501
    const int lt= src[-1-1*stride];
1502
    LOAD_TOP_EDGE    
1503
    LOAD_LEFT_EDGE    
1504
    const __attribute__((unused)) int unu= l3;
1505

    
1506
    src[0+0*stride]=
1507
    src[1+2*stride]=(lt + t0 + 1)>>1;
1508
    src[1+0*stride]=
1509
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1510
    src[2+0*stride]=
1511
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1512
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1513
    src[0+1*stride]=
1514
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1515
    src[1+1*stride]=
1516
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1517
    src[2+1*stride]=
1518
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1519
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1520
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1521
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1522
};
1523

    
1524
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1525
    LOAD_TOP_EDGE    
1526
    LOAD_TOP_RIGHT_EDGE    
1527
    const __attribute__((unused)) int unu= t7;
1528

    
1529
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1530
    src[1+0*stride]=
1531
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1532
    src[2+0*stride]=
1533
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1534
    src[3+0*stride]=
1535
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1536
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1537
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1538
    src[1+1*stride]=
1539
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1540
    src[2+1*stride]=
1541
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1542
    src[3+1*stride]=
1543
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1544
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1545
};
1546

    
1547
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1548
    LOAD_LEFT_EDGE    
1549

    
1550
    src[0+0*stride]=(l0 + l1 + 1)>>1;
1551
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1552
    src[2+0*stride]=
1553
    src[0+1*stride]=(l1 + l2 + 1)>>1;
1554
    src[3+0*stride]=
1555
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1556
    src[2+1*stride]=
1557
    src[0+2*stride]=(l2 + l3 + 1)>>1;
1558
    src[3+1*stride]=
1559
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1560
    src[3+2*stride]=
1561
    src[1+3*stride]=
1562
    src[0+3*stride]=
1563
    src[2+2*stride]=
1564
    src[2+3*stride]=
1565
    src[3+3*stride]=l3;
1566
};
1567
    
1568
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1569
    const int lt= src[-1-1*stride];
1570
    LOAD_TOP_EDGE    
1571
    LOAD_LEFT_EDGE    
1572
    const __attribute__((unused)) int unu= t3;
1573

    
1574
    src[0+0*stride]=
1575
    src[2+1*stride]=(lt + l0 + 1)>>1;
1576
    src[1+0*stride]=
1577
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1578
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1579
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1580
    src[0+1*stride]=
1581
    src[2+2*stride]=(l0 + l1 + 1)>>1;
1582
    src[1+1*stride]=
1583
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1584
    src[0+2*stride]=
1585
    src[2+3*stride]=(l1 + l2+ 1)>>1;
1586
    src[1+2*stride]=
1587
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1588
    src[0+3*stride]=(l2 + l3 + 1)>>1;
1589
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1590
};
1591

    
1592
static void pred16x16_vertical_c(uint8_t *src, int stride){
1593
    int i;
1594
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1595
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1596
    const uint32_t c= ((uint32_t*)(src-stride))[2];
1597
    const uint32_t d= ((uint32_t*)(src-stride))[3];
1598
    
1599
    for(i=0; i<16; i++){
1600
        ((uint32_t*)(src+i*stride))[0]= a;
1601
        ((uint32_t*)(src+i*stride))[1]= b;
1602
        ((uint32_t*)(src+i*stride))[2]= c;
1603
        ((uint32_t*)(src+i*stride))[3]= d;
1604
    }
1605
}
1606

    
1607
static void pred16x16_horizontal_c(uint8_t *src, int stride){
1608
    int i;
1609

    
1610
    for(i=0; i<16; i++){
1611
        ((uint32_t*)(src+i*stride))[0]=
1612
        ((uint32_t*)(src+i*stride))[1]=
1613
        ((uint32_t*)(src+i*stride))[2]=
1614
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1615
    }
1616
}
1617

    
1618
static void pred16x16_dc_c(uint8_t *src, int stride){
1619
    int i, dc=0;
1620

    
1621
    for(i=0;i<16; i++){
1622
        dc+= src[-1+i*stride];
1623
    }
1624
    
1625
    for(i=0;i<16; i++){
1626
        dc+= src[i-stride];
1627
    }
1628

    
1629
    dc= 0x01010101*((dc + 16)>>5);
1630

    
1631
    for(i=0; i<16; i++){
1632
        ((uint32_t*)(src+i*stride))[0]=
1633
        ((uint32_t*)(src+i*stride))[1]=
1634
        ((uint32_t*)(src+i*stride))[2]=
1635
        ((uint32_t*)(src+i*stride))[3]= dc;
1636
    }
1637
}
1638

    
1639
static void pred16x16_left_dc_c(uint8_t *src, int stride){
1640
    int i, dc=0;
1641

    
1642
    for(i=0;i<16; i++){
1643
        dc+= src[-1+i*stride];
1644
    }
1645
    
1646
    dc= 0x01010101*((dc + 8)>>4);
1647

    
1648
    for(i=0; i<16; i++){
1649
        ((uint32_t*)(src+i*stride))[0]=
1650
        ((uint32_t*)(src+i*stride))[1]=
1651
        ((uint32_t*)(src+i*stride))[2]=
1652
        ((uint32_t*)(src+i*stride))[3]= dc;
1653
    }
1654
}
1655

    
1656
static void pred16x16_top_dc_c(uint8_t *src, int stride){
1657
    int i, dc=0;
1658

    
1659
    for(i=0;i<16; i++){
1660
        dc+= src[i-stride];
1661
    }
1662
    dc= 0x01010101*((dc + 8)>>4);
1663

    
1664
    for(i=0; i<16; i++){
1665
        ((uint32_t*)(src+i*stride))[0]=
1666
        ((uint32_t*)(src+i*stride))[1]=
1667
        ((uint32_t*)(src+i*stride))[2]=
1668
        ((uint32_t*)(src+i*stride))[3]= dc;
1669
    }
1670
}
1671

    
1672
static void pred16x16_128_dc_c(uint8_t *src, int stride){
1673
    int i;
1674

    
1675
    for(i=0; i<16; i++){
1676
        ((uint32_t*)(src+i*stride))[0]=
1677
        ((uint32_t*)(src+i*stride))[1]=
1678
        ((uint32_t*)(src+i*stride))[2]=
1679
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1680
    }
1681
}
1682

    
1683
static void pred16x16_plane_c(uint8_t *src, int stride){
1684
    uint8_t *cm = cropTbl + MAX_NEG_CROP;
1685
    int i, dx, dy, dc;
1686
    int temp[16];
1687
    
1688
    dc= 16*(src[15-stride] + src[-1+15*stride]);
1689
    
1690
    dx=dy=0;
1691
    for(i=1; i<9; i++){
1692
        dx += i*(src[7+i-stride] - src[7-i-stride]);
1693
        dy += i*(src[-1+(7+i)*stride] - src[-1+(7-i)*stride]);
1694
    }
1695
    dx= (5*dx+32)>>6;
1696
    dy= (5*dy+32)>>6;
1697
    
1698
    dc += 16;
1699

    
1700
    //FIXME modifiy dc,dx,dy to avoid -7
1701
    
1702
    for(i=0; i<16; i++)
1703
        temp[i]= dx*(i-7) + dc;
1704
    
1705
    if(   (dc - ABS(dx)*8 - ABS(dy)*8)>>5 < 0
1706
       || (dc + ABS(dx)*8 + ABS(dy)*8)>>5 > 255){
1707
    
1708
        for(i=0; i<16; i++){
1709
            int j;
1710
            for(j=0; j<16; j++)
1711
                src[j + i*stride]= cm[ (temp[j] + dy*(i-7))>>5 ];
1712
        }
1713
    }else{
1714
        for(i=0; i<16; i++){
1715
            int j;
1716
            for(j=0; j<16; j++)
1717
                src[j + i*stride]= (temp[j] + dy*(i-7))>>5;
1718
        }
1719
    }
1720
}
1721

    
1722
static void pred8x8_vertical_c(uint8_t *src, int stride){
1723
    int i;
1724
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1725
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1726
    
1727
    for(i=0; i<8; i++){
1728
        ((uint32_t*)(src+i*stride))[0]= a;
1729
        ((uint32_t*)(src+i*stride))[1]= b;
1730
    }
1731
}
1732

    
1733
static void pred8x8_horizontal_c(uint8_t *src, int stride){
1734
    int i;
1735

    
1736
    for(i=0; i<8; i++){
1737
        ((uint32_t*)(src+i*stride))[0]=
1738
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1739
    }
1740
}
1741

    
1742
static void pred8x8_128_dc_c(uint8_t *src, int stride){
1743
    int i;
1744

    
1745
    for(i=0; i<4; i++){
1746
        ((uint32_t*)(src+i*stride))[0]= 
1747
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1748
    }
1749
    for(i=4; i<8; i++){
1750
        ((uint32_t*)(src+i*stride))[0]= 
1751
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1752
    }
1753
}
1754

    
1755
static void pred8x8_left_dc_c(uint8_t *src, int stride){
1756
    int i;
1757
    int dc0, dc2;
1758

    
1759
    dc0=dc2=0;
1760
    for(i=0;i<4; i++){
1761
        dc0+= src[-1+i*stride];
1762
        dc2+= src[-1+(i+4)*stride];
1763
    }
1764
    dc0= 0x01010101*((dc0 + 2)>>2);
1765
    dc2= 0x01010101*((dc2 + 2)>>2);
1766

    
1767
    for(i=0; i<4; i++){
1768
        ((uint32_t*)(src+i*stride))[0]=
1769
        ((uint32_t*)(src+i*stride))[1]= dc0;
1770
    }
1771
    for(i=4; i<8; i++){
1772
        ((uint32_t*)(src+i*stride))[0]=
1773
        ((uint32_t*)(src+i*stride))[1]= dc2;
1774
    }
1775
}
1776

    
1777
static void pred8x8_top_dc_c(uint8_t *src, int stride){
1778
    int i;
1779
    int dc0, dc1;
1780

    
1781
    dc0=dc1=0;
1782
    for(i=0;i<4; i++){
1783
        dc0+= src[i-stride];
1784
        dc1+= src[4+i-stride];
1785
    }
1786
    dc0= 0x01010101*((dc0 + 2)>>2);
1787
    dc1= 0x01010101*((dc1 + 2)>>2);
1788

    
1789
    for(i=0; i<4; i++){
1790
        ((uint32_t*)(src+i*stride))[0]= dc0;
1791
        ((uint32_t*)(src+i*stride))[1]= dc1;
1792
    }
1793
    for(i=4; i<8; i++){
1794
        ((uint32_t*)(src+i*stride))[0]= dc0;
1795
        ((uint32_t*)(src+i*stride))[1]= dc1;
1796
    }
1797
}
1798

    
1799

    
1800
static void pred8x8_dc_c(uint8_t *src, int stride){
1801
    int i;
1802
    int dc0, dc1, dc2, dc3;
1803

    
1804
    dc0=dc1=dc2=0;
1805
    for(i=0;i<4; i++){
1806
        dc0+= src[-1+i*stride] + src[i-stride];
1807
        dc1+= src[4+i-stride];
1808
        dc2+= src[-1+(i+4)*stride];
1809
    }
1810
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1811
    dc0= 0x01010101*((dc0 + 4)>>3);
1812
    dc1= 0x01010101*((dc1 + 2)>>2);
1813
    dc2= 0x01010101*((dc2 + 2)>>2);
1814

    
1815
    for(i=0; i<4; i++){
1816
        ((uint32_t*)(src+i*stride))[0]= dc0;
1817
        ((uint32_t*)(src+i*stride))[1]= dc1;
1818
    }
1819
    for(i=4; i<8; i++){
1820
        ((uint32_t*)(src+i*stride))[0]= dc2;
1821
        ((uint32_t*)(src+i*stride))[1]= dc3;
1822
    }
1823
}
1824

    
1825
static void pred8x8_plane_c(uint8_t *src, int stride){
1826
    uint8_t *cm = cropTbl + MAX_NEG_CROP;
1827
    int i, dx, dy, dc;
1828
    int temp[8];
1829

    
1830
    dc= 16*(src[7-stride] + src[-1+7*stride]);
1831
    
1832
    dx=dy=0;
1833
    for(i=1; i<5; i++){
1834
        dx += i*(src[3+i-stride] - src[3-i-stride]);
1835
        dy += i*(src[-1+(3+i)*stride] - src[-1+(3-i)*stride]);
1836
    }
1837
    dx= (17*dx+16)>>5;
1838
    dy= (17*dy+16)>>5;
1839
    
1840
    dc += 16;
1841
    
1842
    //FIXME modifiy dc,dx,dy to avoid -3
1843
    
1844
    for(i=0; i<8; i++)
1845
        temp[i]= dx*(i-3) + dc;
1846
    
1847
    if(   (dc - ABS(dx)*4 - ABS(dy)*4)>>5 < 0
1848
       || (dc + ABS(dx)*4 + ABS(dy)*4)>>5 > 255){
1849
    
1850
        for(i=0; i<8; i++){
1851
            int j;
1852
            for(j=0; j<8; j++)
1853
                src[j + i*stride]= cm[ (temp[j] + dy*(i-3))>>5 ];
1854
        }
1855
    }else{
1856
        for(i=0; i<8; i++){
1857
            int j;
1858
            for(j=0; j<8; j++)
1859
                src[j + i*stride]= (temp[j] + dy*(i-3))>>5;
1860
        }
1861
    }
1862
}
1863

    
1864
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1865
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1866
                           int src_x_offset, int src_y_offset,
1867
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1868
    MpegEncContext * const s = &h->s;
1869
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1870
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1871
    const int luma_xy= (mx&3) + ((my&3)<<2);
1872
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1873
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1874
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1875
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1876
    int extra_height= extra_width;
1877
    int emu=0;
1878
    const int full_mx= mx>>2;
1879
    const int full_my= my>>2;
1880
    
1881
    assert(pic->data[0]);
1882
    
1883
    if(mx&7) extra_width -= 3;
1884
    if(my&7) extra_height -= 3;
1885
    
1886
    if(   full_mx < 0-extra_width 
1887
       || full_my < 0-extra_height 
1888
       || full_mx + 16/*FIXME*/ > s->width + extra_width 
1889
       || full_my + 16/*FIXME*/ > s->height + extra_height){
1890
        ff_emulated_edge_mc(s, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, s->width, s->height);
1891
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1892
        emu=1;
1893
    }
1894
    
1895
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1896
    if(!square){
1897
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1898
    }
1899
    
1900
    if(s->flags&CODEC_FLAG_GRAY) return;
1901
    
1902
    if(emu){
1903
        ff_emulated_edge_mc(s, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
1904
            src_cb= s->edge_emu_buffer;
1905
    }
1906
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1907

    
1908
    if(emu){
1909
        ff_emulated_edge_mc(s, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
1910
            src_cr= s->edge_emu_buffer;
1911
    }
1912
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1913
}
1914

    
1915
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1916
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1917
                           int x_offset, int y_offset,
1918
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1919
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1920
                           int list0, int list1){
1921
    MpegEncContext * const s = &h->s;
1922
    qpel_mc_func *qpix_op=  qpix_put;
1923
    h264_chroma_mc_func chroma_op= chroma_put;
1924
    
1925
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
1926
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
1927
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
1928
    x_offset += 8*s->mb_x;
1929
    y_offset += 8*s->mb_y;
1930
    
1931
    if(list0){
1932
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1933
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1934
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1935
                           qpix_op, chroma_op);
1936

    
1937
        qpix_op=  qpix_avg;
1938
        chroma_op= chroma_avg;
1939
    }
1940

    
1941
    if(list1){
1942
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1943
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1944
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1945
                           qpix_op, chroma_op);
1946
    }
1947
}
1948

    
1949
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1950
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1951
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
1952
    MpegEncContext * const s = &h->s;
1953
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1954
    const int mb_type= s->current_picture.mb_type[mb_xy];
1955
    
1956
    assert(IS_INTER(mb_type));
1957
    
1958
    if(IS_16X16(mb_type)){
1959
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1960
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1961
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1962
    }else if(IS_16X8(mb_type)){
1963
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1964
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1965
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1966
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1967
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1968
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1969
    }else if(IS_8X16(mb_type)){
1970
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
1971
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1972
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1973
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
1974
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1975
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1976
    }else{
1977
        int i;
1978
        
1979
        assert(IS_8X8(mb_type));
1980

    
1981
        for(i=0; i<4; i++){
1982
            const int sub_mb_type= h->sub_mb_type[i];
1983
            const int n= 4*i;
1984
            int x_offset= (i&1)<<2;
1985
            int y_offset= (i&2)<<1;
1986

    
1987
            if(IS_SUB_8X8(sub_mb_type)){
1988
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1989
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1990
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1991
            }else if(IS_SUB_8X4(sub_mb_type)){
1992
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1993
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1994
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1995
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1996
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1997
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1998
            }else if(IS_SUB_4X8(sub_mb_type)){
1999
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2000
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2001
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2002
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2003
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2004
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2005
            }else{
2006
                int j;
2007
                assert(IS_SUB_4X4(sub_mb_type));
2008
                for(j=0; j<4; j++){
2009
                    int sub_x_offset= x_offset + 2*(j&1);
2010
                    int sub_y_offset= y_offset +   (j&2);
2011
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2012
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2013
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2014
                }
2015
            }
2016
        }
2017
    }
2018
}
2019

    
2020
static void decode_init_vlc(H264Context *h){
2021
    static int done = 0;
2022

    
2023
    if (!done) {
2024
        int i;
2025
        done = 1;
2026

    
2027
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2028
                 &chroma_dc_coeff_token_len [0], 1, 1,
2029
                 &chroma_dc_coeff_token_bits[0], 1, 1);
2030

    
2031
        for(i=0; i<4; i++){
2032
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2033
                     &coeff_token_len [i][0], 1, 1,
2034
                     &coeff_token_bits[i][0], 1, 1);
2035
        }
2036

    
2037
        for(i=0; i<3; i++){
2038
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2039
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2040
                     &chroma_dc_total_zeros_bits[i][0], 1, 1);
2041
        }
2042
        for(i=0; i<15; i++){
2043
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2044
                     &total_zeros_len [i][0], 1, 1,
2045
                     &total_zeros_bits[i][0], 1, 1);
2046
        }
2047

    
2048
        for(i=0; i<6; i++){
2049
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2050
                     &run_len [i][0], 1, 1,
2051
                     &run_bits[i][0], 1, 1);
2052
        }
2053
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2054
                 &run_len [6][0], 1, 1,
2055
                 &run_bits[6][0], 1, 1);
2056
    }
2057
}
2058

    
2059
/**
2060
 * Sets the intra prediction function pointers.
2061
 */
2062
static void init_pred_ptrs(H264Context *h){
2063
//    MpegEncContext * const s = &h->s;
2064

    
2065
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2066
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2067
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2068
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2069
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2070
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2071
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2072
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2073
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2074
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2075
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2076
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2077

    
2078
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2079
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2080
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2081
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2082
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2083
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2084
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2085

    
2086
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2087
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2088
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2089
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2090
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2091
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2092
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2093
}
2094

    
2095
//FIXME factorize
2096
#define CHECKED_ALLOCZ(p, size)\
2097
{\
2098
    p= av_mallocz(size);\
2099
    if(p==NULL){\
2100
        perror("malloc");\
2101
        goto fail;\
2102
    }\
2103
}
2104

    
2105
static void free_tables(H264Context *h){
2106
    MpegEncContext * const s = &h->s;
2107

    
2108
    av_freep(&h->intra4x4_pred_mode);
2109
    av_freep(&h->non_zero_count);
2110
    av_freep(&h->slice_table_base);
2111
    h->slice_table= NULL;
2112
    
2113
    av_freep(&h->mb2b_xy);
2114
    av_freep(&h->mb2b8_xy);
2115
}
2116

    
2117
/**
2118
 * allocates tables.
2119
 * needs widzh/height
2120
 */
2121
static int alloc_tables(H264Context *h){
2122
    MpegEncContext * const s = &h->s;
2123
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2124
    int x,y;
2125

    
2126
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2127
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2128
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2129

    
2130
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2131
    h->slice_table= h->slice_table_base + s->mb_stride + 1;
2132

    
2133
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint16_t));
2134
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2135
    for(y=0; y<s->mb_height; y++){
2136
        for(x=0; x<s->mb_width; x++){
2137
            const int mb_xy= x + y*s->mb_stride;
2138
            const int b_xy = 4*x + 4*y*h->b_stride;
2139
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2140
        
2141
            h->mb2b_xy [mb_xy]= b_xy;
2142
            h->mb2b8_xy[mb_xy]= b8_xy;
2143
        }
2144
    }
2145
    
2146
    return 0;
2147
fail:
2148
    free_tables(h);
2149
    return -1;
2150
}
2151

    
2152
static void common_init(H264Context *h){
2153
    MpegEncContext * const s = &h->s;
2154
    int i;
2155

    
2156
    s->width = s->avctx->width;
2157
    s->height = s->avctx->height;
2158
    s->codec_id= s->avctx->codec->id;
2159
    
2160
    init_pred_ptrs(h);
2161

    
2162
    s->decode=1; //FIXME
2163
}
2164

    
2165
static int decode_init(AVCodecContext *avctx){
2166
    H264Context *h= avctx->priv_data;
2167
    MpegEncContext * const s = &h->s;
2168

    
2169
    s->avctx = avctx;
2170
    common_init(h);
2171

    
2172
    s->out_format = FMT_H264;
2173
    s->workaround_bugs= avctx->workaround_bugs;
2174

    
2175
    // set defaults
2176
    s->progressive_sequence=1;
2177
//    s->decode_mb= ff_h263_decode_mb;
2178
    s->low_delay= 1;
2179
    avctx->pix_fmt= PIX_FMT_YUV420P;
2180

    
2181
    decode_init_vlc(h);
2182
    
2183
    return 0;
2184
}
2185

    
2186
static void frame_start(H264Context *h){
2187
    MpegEncContext * const s = &h->s;
2188
    int i;
2189

    
2190
    MPV_frame_start(s, s->avctx);
2191
    ff_er_frame_start(s);
2192
    h->mmco_index=0;
2193

    
2194
    assert(s->linesize && s->uvlinesize);
2195

    
2196
    for(i=0; i<16; i++){
2197
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2198
        h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2199
    }
2200
    for(i=0; i<4; i++){
2201
        h->block_offset[16+i]=
2202
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2203
    }
2204

    
2205
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2206
}
2207

    
2208
static void hl_decode_mb(H264Context *h){
2209
    MpegEncContext * const s = &h->s;
2210
    const int mb_x= s->mb_x;
2211
    const int mb_y= s->mb_y;
2212
    const int mb_xy= mb_x + mb_y*s->mb_stride;
2213
    const int mb_type= s->current_picture.mb_type[mb_xy];
2214
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2215
    int linesize, uvlinesize /*dct_offset*/;
2216
    int i;
2217

    
2218
    if(!s->decode)
2219
        return;
2220

    
2221
    if(s->mb_skiped){
2222
    }
2223

    
2224
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2225
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2226
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2227

    
2228
    if (h->mb_field_decoding_flag) {
2229
        linesize = s->linesize * 2;
2230
        uvlinesize = s->uvlinesize * 2;
2231
        if(mb_y&1){ //FIXME move out of this func?
2232
            dest_y -= s->linesize*15;
2233
            dest_cb-= s->linesize*7;
2234
            dest_cr-= s->linesize*7;
2235
        }
2236
    } else {
2237
        linesize = s->linesize;
2238
        uvlinesize = s->uvlinesize;
2239
//        dct_offset = s->linesize * 16;
2240
    }
2241

    
2242
    if(IS_INTRA(mb_type)){
2243
        if(!(s->flags&CODEC_FLAG_GRAY)){
2244
            h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2245
            h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2246
        }
2247

    
2248
        if(IS_INTRA4x4(mb_type)){
2249
            if(!s->encoding){
2250
                for(i=0; i<16; i++){
2251
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2252
                    uint8_t *topright= ptr + 4 - linesize;
2253
                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2254
                    const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2255
                    int tr;
2256

    
2257
                    if(!topright_avail){
2258
                        tr= ptr[3 - linesize]*0x01010101;
2259
                        topright= (uint8_t*) &tr;
2260
                    }
2261

    
2262
                    h->pred4x4[ dir ](ptr, topright, linesize);
2263
                    if(h->non_zero_count_cache[ scan8[i] ])
2264
                        h264_add_idct_c(ptr, h->mb + i*16, linesize);
2265
                }
2266
            }
2267
        }else{
2268
            h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2269
            h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2270
        }
2271
    }else{
2272
        hl_motion(h, dest_y, dest_cb, dest_cr,
2273
                  s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
2274
                  s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2275
    }
2276

    
2277

    
2278
    if(!IS_INTRA4x4(mb_type)){
2279
        for(i=0; i<16; i++){
2280
            if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2281
                uint8_t * const ptr= dest_y + h->block_offset[i];
2282
                h264_add_idct_c(ptr, h->mb + i*16, linesize);
2283
            }
2284
        }
2285
    }
2286

    
2287
    if(!(s->flags&CODEC_FLAG_GRAY)){
2288
        chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2289
        chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2290
        for(i=16; i<16+4; i++){
2291
            if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2292
                uint8_t * const ptr= dest_cb + h->block_offset[i];
2293
                h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2294
            }
2295
        }
2296
        for(i=20; i<20+4; i++){
2297
            if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2298
                uint8_t * const ptr= dest_cr + h->block_offset[i];
2299
                h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2300
            }
2301
        }
2302
    }
2303
}
2304

    
2305
static void decode_mb_cabac(H264Context *h){
2306
//    MpegEncContext * const s = &h->s;
2307
}
2308

    
2309
/**
2310
 * fills the default_ref_list.
2311
 */
2312
static int fill_default_ref_list(H264Context *h){
2313
    MpegEncContext * const s = &h->s;
2314
    int i;
2315
    Picture sorted_short_ref[16];
2316
    
2317
    if(h->slice_type==B_TYPE){
2318
        int out_i;
2319
        int limit= -1;
2320

    
2321
        for(out_i=0; out_i<h->short_ref_count; out_i++){
2322
            int best_i=-1;
2323
            int best_poc=-1;
2324

    
2325
            for(i=0; i<h->short_ref_count; i++){
2326
                const int poc= h->short_ref[i]->poc;
2327
                if(poc > limit && poc < best_poc){
2328
                    best_poc= poc;
2329
                    best_i= i;
2330
                }
2331
            }
2332
            
2333
            assert(best_i != -1);
2334
            
2335
            limit= best_poc;
2336
            sorted_short_ref[out_i]= *h->short_ref[best_i];
2337
        }
2338
    }
2339

    
2340
    if(s->picture_structure == PICT_FRAME){
2341
        if(h->slice_type==B_TYPE){
2342
            const int current_poc= s->current_picture_ptr->poc;
2343
            int list;
2344

    
2345
            for(list=0; list<2; list++){
2346
                int index=0;
2347

    
2348
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2349
                    const int i2= list ? h->short_ref_count - i - 1 : i;
2350
                    const int poc= sorted_short_ref[i2].poc;
2351
                    
2352
                    if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2353

    
2354
                    if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2355
                        h->default_ref_list[list][index  ]= sorted_short_ref[i2];
2356
                        h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2357
                    }
2358
                }
2359

    
2360
                for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2361
                    if(h->long_ref[i]->reference != 3) continue;
2362

    
2363
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
2364
                    h->default_ref_list[ list ][index++].pic_id= i;;
2365
                }
2366
                
2367
                if(h->long_ref_count > 1 && h->short_ref_count==0){
2368
                    Picture temp= h->default_ref_list[1][0];
2369
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
2370
                    h->default_ref_list[1][0] = temp;
2371
                }
2372

    
2373
                if(index < h->ref_count[ list ])
2374
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2375
            }
2376
        }else{
2377
            int index=0;
2378
            for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2379
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2380
                h->default_ref_list[0][index  ]= *h->short_ref[i];
2381
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2382
            }
2383
            for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2384
                if(h->long_ref[i]->reference != 3) continue;
2385
                h->default_ref_list[0][index  ]= *h->long_ref[i];
2386
                h->default_ref_list[0][index++].pic_id= i;;
2387
            }
2388
            if(index < h->ref_count[0])
2389
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2390
        }
2391
    }else{ //FIELD
2392
        if(h->slice_type==B_TYPE){
2393
        }else{
2394
            //FIXME second field balh
2395
        }
2396
    }
2397
    return 0;
2398
}
2399

    
2400
static int decode_ref_pic_list_reordering(H264Context *h){
2401
    MpegEncContext * const s = &h->s;
2402
    int list;
2403
    
2404
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2405
    
2406
    for(list=0; list<2; list++){
2407
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2408

    
2409
        if(get_bits1(&s->gb)){
2410
            int pred= h->curr_pic_num;
2411
            int index;
2412

    
2413
            for(index=0; ; index++){
2414
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2415
                int pic_id;
2416
                int i;
2417
                
2418
                
2419
                if(index >= h->ref_count[list]){
2420
                    fprintf(stderr, "reference count overflow\n");
2421
                    return -1;
2422
                }
2423
                
2424
                if(reordering_of_pic_nums_idc<3){
2425
                    if(reordering_of_pic_nums_idc<2){
2426
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2427

    
2428
                        if(abs_diff_pic_num >= h->max_pic_num){
2429
                            fprintf(stderr, "abs_diff_pic_num overflow\n");
2430
                            return -1;
2431
                        }
2432

    
2433
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2434
                        else                                pred+= abs_diff_pic_num;
2435
                        pred &= h->max_pic_num - 1;
2436
                    
2437
                        for(i= h->ref_count[list]-1; i>=index; i--){
2438
                            if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2439
                                break;
2440
                        }
2441
                    }else{
2442
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2443

    
2444
                        for(i= h->ref_count[list]-1; i>=index; i--){
2445
                            if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2446
                                break;
2447
                        }
2448
                    }
2449

    
2450
                    if(i < index){
2451
                        fprintf(stderr, "reference picture missing during reorder\n");
2452
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2453
                    }else if(i > index){
2454
                        Picture tmp= h->ref_list[list][i];
2455
                        for(; i>index; i--){
2456
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2457
                        }
2458
                        h->ref_list[list][index]= tmp;
2459
                    }
2460
                }else if(reordering_of_pic_nums_idc==3) 
2461
                    break;
2462
                else{
2463
                    fprintf(stderr, "illegal reordering_of_pic_nums_idc\n");
2464
                    return -1;
2465
                }
2466
            }
2467
        }
2468

    
2469
        if(h->slice_type!=B_TYPE) break;
2470
    }
2471
    return 0;    
2472
}
2473

    
2474
static int pred_weight_table(H264Context *h){
2475
    MpegEncContext * const s = &h->s;
2476
    int list, i;
2477
    
2478
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2479
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2480

    
2481
    for(list=0; list<2; list++){
2482
        for(i=0; i<h->ref_count[list]; i++){
2483
            int luma_weight_flag, chroma_weight_flag;
2484
            
2485
            luma_weight_flag= get_bits1(&s->gb);
2486
            if(luma_weight_flag){
2487
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
2488
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
2489
            }
2490

    
2491
            chroma_weight_flag= get_bits1(&s->gb);
2492
            if(chroma_weight_flag){
2493
                int j;
2494
                for(j=0; j<2; j++){
2495
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2496
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2497
                }
2498
            }
2499
        }
2500
        if(h->slice_type != B_TYPE) break;
2501
    }
2502
    return 0;
2503
}
2504

    
2505
/**
2506
 * instantaneos decoder refresh.
2507
 */
2508
static void idr(H264Context *h){
2509
    int i;
2510

    
2511
    for(i=0; i<h->long_ref_count; i++){
2512
        h->long_ref[i]->reference=0;
2513
        h->long_ref[i]= NULL;
2514
    }
2515
    h->long_ref_count=0;
2516

    
2517
    for(i=0; i<h->short_ref_count; i++){
2518
        h->short_ref[i]->reference=0;
2519
        h->short_ref[i]= NULL;
2520
    }
2521
    h->short_ref_count=0;
2522
}
2523

    
2524
/**
2525
 *
2526
 * @return the removed picture or NULL if an error occures
2527
 */
2528
static Picture * remove_short(H264Context *h, int frame_num){
2529
    MpegEncContext * const s = &h->s;
2530
    int i;
2531
    
2532
    if(s->avctx->debug&FF_DEBUG_MMCO)
2533
        printf("remove short %d count %d\n", frame_num, h->short_ref_count);
2534
    
2535
    for(i=0; i<h->short_ref_count; i++){
2536
        Picture *pic= h->short_ref[i];
2537
        if(s->avctx->debug&FF_DEBUG_MMCO)
2538
            printf("%d %d %X\n", i, pic->frame_num, (int)pic);
2539
        if(pic->frame_num == frame_num){
2540
            h->short_ref[i]= NULL;
2541
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2542
            h->short_ref_count--;
2543
            return pic;
2544
        }
2545
    }
2546
    return NULL;
2547
}
2548

    
2549
/**
2550
 *
2551
 * @return the removed picture or NULL if an error occures
2552
 */
2553
static Picture * remove_long(H264Context *h, int i){
2554
    Picture *pic;
2555

    
2556
    if(i >= h->long_ref_count) return NULL;
2557
    pic= h->long_ref[i];
2558
    if(pic==NULL) return NULL;
2559
    
2560
    h->long_ref[i]= NULL;
2561
    memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2562
    h->long_ref_count--;
2563

    
2564
    return pic;
2565
}
2566

    
2567
/**
2568
 * Executes the reference picture marking (memory management control operations).
2569
 */
2570
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2571
    MpegEncContext * const s = &h->s;
2572
    int i;
2573
    int current_is_long=0;
2574
    Picture *pic;
2575
    
2576
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2577
        printf("no mmco here\n");
2578
        
2579
    for(i=0; i<mmco_count; i++){
2580
        if(s->avctx->debug&FF_DEBUG_MMCO)
2581
            printf("mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
2582

    
2583
        switch(mmco[i].opcode){
2584
        case MMCO_SHORT2UNUSED:
2585
            pic= remove_short(h, mmco[i].short_frame_num);
2586
            if(pic==NULL) return -1;
2587
            pic->reference= 0;
2588
            break;
2589
        case MMCO_SHORT2LONG:
2590
            pic= remove_long(h, mmco[i].long_index);
2591
            if(pic) pic->reference=0;
2592
            
2593
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2594
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
2595
            break;
2596
        case MMCO_LONG2UNUSED:
2597
            pic= remove_long(h, mmco[i].long_index);
2598
            if(pic==NULL) return -1;
2599
            pic->reference= 0;
2600
            break;
2601
        case MMCO_LONG:
2602
            pic= remove_long(h, mmco[i].long_index);
2603
            if(pic) pic->reference=0;
2604
            
2605
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2606
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
2607
            h->long_ref_count++;
2608
            
2609
            current_is_long=1;
2610
            break;
2611
        case MMCO_SET_MAX_LONG:
2612
            assert(mmco[i].long_index <= 16);
2613
            while(mmco[i].long_index < h->long_ref_count){
2614
                pic= remove_long(h, mmco[i].long_index);
2615
                pic->reference=0;
2616
            }
2617
            while(mmco[i].long_index > h->long_ref_count){
2618
                h->long_ref[ h->long_ref_count++ ]= NULL;
2619
            }
2620
            break;
2621
        case MMCO_RESET:
2622
            while(h->short_ref_count){
2623
                pic= remove_short(h, h->short_ref[0]->frame_num);
2624
                pic->reference=0;
2625
            }
2626
            while(h->long_ref_count){
2627
                pic= remove_long(h, h->long_ref_count-1);
2628
                pic->reference=0;
2629
            }
2630
            break;
2631
        default: assert(0);
2632
        }
2633
    }
2634
    
2635
    if(!current_is_long){
2636
        pic= remove_short(h, s->current_picture_ptr->frame_num);
2637
        if(pic){
2638
            pic->reference=0;
2639
            fprintf(stderr, "illegal short term buffer state detected\n");
2640
        }
2641
        
2642
        if(h->short_ref_count)
2643
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2644

    
2645
        h->short_ref[0]= s->current_picture_ptr;
2646
        h->short_ref[0]->long_ref=0;
2647
        h->short_ref_count++;
2648
    }
2649
    
2650
    return 0; 
2651
}
2652

    
2653
static int decode_ref_pic_marking(H264Context *h){
2654
    MpegEncContext * const s = &h->s;
2655
    int i;
2656
    
2657
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2658
        s->broken_link= get_bits1(&s->gb) -1;
2659
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2660
        if(h->mmco[0].long_index == -1)
2661
            h->mmco_index= 0;
2662
        else{
2663
            h->mmco[0].opcode= MMCO_LONG;
2664
            h->mmco_index= 1;
2665
        } 
2666
    }else{
2667
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2668
            for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) { 
2669
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2670

    
2671
                h->mmco[i].opcode= opcode;
2672
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2673
                    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
2674
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2675
                        fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2676
                        return -1;
2677
                    }*/
2678
                }
2679
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2680
                    h->mmco[i].long_index= get_ue_golomb(&s->gb);
2681
                    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){
2682
                        fprintf(stderr, "illegal long ref in memory management control operation %d\n", opcode);
2683
                        return -1;
2684
                    }
2685
                }
2686
                    
2687
                if(opcode > MMCO_LONG){
2688
                    fprintf(stderr, "illegal memory management control operation %d\n", opcode);
2689
                    return -1;
2690
                }
2691
            }
2692
            h->mmco_index= i;
2693
        }else{
2694
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2695

    
2696
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2697
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2698
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2699
                h->mmco_index= 1;
2700
            }else
2701
                h->mmco_index= 0;
2702
        }
2703
    }
2704
    
2705
    return 0; 
2706
}
2707

    
2708
static int init_poc(H264Context *h){
2709
    MpegEncContext * const s = &h->s;
2710
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2711
    int field_poc[2];
2712

    
2713
    if(h->nal_unit_type == NAL_IDR_SLICE){
2714
        h->frame_num_offset= 0;
2715
    }else{
2716
        if(h->frame_num < h->prev_frame_num)
2717
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2718
        else
2719
            h->frame_num_offset= h->prev_frame_num_offset;
2720
    }
2721

    
2722
    if(h->sps.poc_type==0){
2723
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2724

    
2725
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2726
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2727
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2728
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2729
        else
2730
            h->poc_msb = h->prev_poc_msb;
2731
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2732
        field_poc[0] = 
2733
        field_poc[1] = h->poc_msb + h->poc_lsb;
2734
        if(s->picture_structure == PICT_FRAME) 
2735
            field_poc[1] += h->delta_poc_bottom;
2736
    }else if(h->sps.poc_type==1){
2737
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2738
        int i;
2739

    
2740
        if(h->sps.poc_cycle_length != 0)
2741
            abs_frame_num = h->frame_num_offset + h->frame_num;
2742
        else
2743
            abs_frame_num = 0;
2744

    
2745
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
2746
            abs_frame_num--;
2747
            
2748
        expected_delta_per_poc_cycle = 0;
2749
        for(i=0; i < h->sps.poc_cycle_length; i++)
2750
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2751

    
2752
        if(abs_frame_num > 0){
2753
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2754
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2755

    
2756
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2757
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
2758
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2759
        } else
2760
            expectedpoc = 0;
2761

    
2762
        if(h->nal_ref_idc == 0) 
2763
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2764
        
2765
        field_poc[0] = expectedpoc + h->delta_poc[0];
2766
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2767

    
2768
        if(s->picture_structure == PICT_FRAME)
2769
            field_poc[1] += h->delta_poc[1];
2770
    }else{
2771
        int poc;
2772
        if(h->nal_unit_type == NAL_IDR_SLICE){
2773
            poc= 0;
2774
        }else{
2775
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2776
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2777
        }
2778
        field_poc[0]= poc;
2779
        field_poc[1]= poc;
2780
    }
2781
    
2782
    if(s->picture_structure != PICT_BOTTOM_FIELD)
2783
        s->current_picture_ptr->field_poc[0]= field_poc[0];
2784
    if(s->picture_structure != PICT_TOP_FIELD)
2785
        s->current_picture_ptr->field_poc[1]= field_poc[1];
2786
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2787
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2788

    
2789
    return 0;
2790
}
2791

    
2792
/**
2793
 * decodes a slice header.
2794
 * this will allso call MPV_common_init() and frame_start() as needed
2795
 */
2796
static int decode_slice_header(H264Context *h){
2797
    MpegEncContext * const s = &h->s;
2798
    int first_mb_in_slice, pps_id;
2799
    int num_ref_idx_active_override_flag;
2800
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2801
    float new_aspect;
2802

    
2803
    s->current_picture.reference= h->nal_ref_idc != 0;
2804

    
2805
    first_mb_in_slice= get_ue_golomb(&s->gb);
2806

    
2807
    h->slice_type= get_ue_golomb(&s->gb);
2808
    if(h->slice_type > 9){
2809
        fprintf(stderr, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
2810
    }
2811
    if(h->slice_type > 4){
2812
        h->slice_type -= 5;
2813
        h->slice_type_fixed=1;
2814
    }else
2815
        h->slice_type_fixed=0;
2816
    
2817
    h->slice_type= slice_type_map[ h->slice_type ];
2818
    
2819
    s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
2820
        
2821
    pps_id= get_ue_golomb(&s->gb);
2822
    if(pps_id>255){
2823
        fprintf(stderr, "pps_id out of range\n");
2824
        return -1;
2825
    }
2826
    h->pps= h->pps_buffer[pps_id];
2827
    if(h->pps.slice_group_count == 0){
2828
        fprintf(stderr, "non existing PPS referenced\n");
2829
        return -1;
2830
    }
2831

    
2832
    h->sps= h->sps_buffer[ h->pps.sps_id ];
2833
    if(h->sps.log2_max_frame_num == 0){
2834
        fprintf(stderr, "non existing SPS referenced\n");
2835
        return -1;
2836
    }
2837
    
2838
    s->mb_width= h->sps.mb_width;
2839
    s->mb_height= h->sps.mb_height;
2840
    
2841
    h->b_stride=  s->mb_width*4;
2842
    h->b8_stride= s->mb_width*2;
2843

    
2844
    s->mb_x = first_mb_in_slice % s->mb_width;
2845
    s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
2846
    
2847
    s->width = 16*s->mb_width - 2*(h->pps.crop_left + h->pps.crop_right );
2848
    if(h->sps.frame_mbs_only_flag)
2849
        s->height= 16*s->mb_height - 2*(h->pps.crop_top  + h->pps.crop_bottom);
2850
    else
2851
        s->height= 16*s->mb_height - 4*(h->pps.crop_top  + h->pps.crop_bottom); //FIXME recheck
2852
    
2853
    if(h->pps.crop_left || h->pps.crop_top){
2854
        fprintf(stderr, "insane croping not completly supported, this could look slightly wrong ...\n");
2855
    }
2856

    
2857
    if(s->aspected_height) //FIXME emms at end of slice ?
2858
        new_aspect= h->sps.sar_width*s->width / (float)(s->height*h->sps.sar_height);
2859
    else
2860
        new_aspect=0;
2861

    
2862
    if (s->context_initialized 
2863
        && (   s->width != s->avctx->width || s->height != s->avctx->height 
2864
            || ABS(new_aspect - s->avctx->aspect_ratio) > 0.001)) {
2865
        free_tables(h);
2866
        MPV_common_end(s);
2867
    }
2868
    if (!s->context_initialized) {
2869
        if (MPV_common_init(s) < 0)
2870
            return -1;
2871

    
2872
        alloc_tables(h);
2873

    
2874
        s->avctx->width = s->width;
2875
        s->avctx->height = s->height;
2876
        s->avctx->aspect_ratio= new_aspect;
2877
    }
2878

    
2879
    if(first_mb_in_slice == 0){
2880
        frame_start(h);
2881
    }
2882

    
2883
    s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
2884
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2885

    
2886
    if(h->sps.frame_mbs_only_flag){
2887
        s->picture_structure= PICT_FRAME;
2888
    }else{
2889
        if(get_bits1(&s->gb)) //field_pic_flag
2890
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2891
        else
2892
            s->picture_structure= PICT_FRAME;
2893
    }
2894

    
2895
    if(s->picture_structure==PICT_FRAME){
2896
        h->curr_pic_num=   h->frame_num;
2897
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2898
    }else{
2899
        h->curr_pic_num= 2*h->frame_num;
2900
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2901
    }
2902
        
2903
    if(h->nal_unit_type == NAL_IDR_SLICE){
2904
        int idr_pic_id= get_ue_golomb(&s->gb);
2905
    }
2906
   
2907
    if(h->sps.poc_type==0){
2908
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2909
        
2910
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2911
            h->delta_poc_bottom= get_se_golomb(&s->gb);
2912
        }
2913
    }
2914
    
2915
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2916
        h->delta_poc[0]= get_se_golomb(&s->gb);
2917
        
2918
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2919
            h->delta_poc[1]= get_se_golomb(&s->gb);
2920
    }
2921
    
2922
    init_poc(h);
2923
    
2924
    if(h->pps.redundant_pic_cnt_present){
2925
        h->redundant_pic_count= get_ue_golomb(&s->gb);
2926
    }
2927

    
2928
    //set defaults, might be overriden a few line later
2929
    h->ref_count[0]= h->pps.ref_count[0];
2930
    h->ref_count[1]= h->pps.ref_count[1];
2931

    
2932
    if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
2933
        if(h->slice_type == B_TYPE){
2934
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
2935
        }
2936
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
2937
    
2938
        if(num_ref_idx_active_override_flag){
2939
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2940
            if(h->slice_type==B_TYPE)
2941
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2942

    
2943
            if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
2944
                fprintf(stderr, "reference overflow\n");
2945
                return -1;
2946
            }
2947
        }
2948
    }
2949

    
2950
    if(first_mb_in_slice == 0){
2951
        fill_default_ref_list(h);
2952
    }
2953

    
2954
    decode_ref_pic_list_reordering(h);
2955

    
2956
    if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) 
2957
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
2958
        pred_weight_table(h);
2959
    
2960
    if(s->current_picture.reference)
2961
        decode_ref_pic_marking(h);
2962
    //FIXME CABAC stuff
2963

    
2964
    s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta
2965
    //FIXME qscale / qp ... stuff
2966
    if(h->slice_type == SP_TYPE){
2967
        int sp_for_switch_flag= get_bits1(&s->gb);
2968
    }
2969
    if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
2970
        int slice_qs_delta= get_se_golomb(&s->gb);
2971
    }
2972

    
2973
    if( h->pps.deblocking_filter_parameters_present ) {
2974
        h->disable_deblocking_filter_idc= get_ue_golomb(&s->gb);
2975
        if( h->disable_deblocking_filter_idc  !=  1 ) {
2976
            h->slice_alpha_c0_offset_div2= get_se_golomb(&s->gb);
2977
            h->slice_beta_offset_div2= get_se_golomb(&s->gb);
2978
        }
2979
    }else
2980
        h->disable_deblocking_filter_idc= 0;
2981

    
2982
#if 0 //FMO
2983
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2984
        slice_group_change_cycle= get_bits(&s->gb, ?);
2985
#endif
2986

    
2987
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2988
        printf("mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n", 
2989
               first_mb_in_slice, 
2990
               ff_get_pict_type_char(h->slice_type),
2991
               pps_id, h->frame_num,
2992
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2993
               h->ref_count[0], h->ref_count[1],
2994
               s->qscale,
2995
               h->disable_deblocking_filter_idc
2996
               );
2997
    }
2998

    
2999
    return 0;
3000
}
3001

    
3002
/**
3003
 *
3004
 */
3005
static inline int get_level_prefix(GetBitContext *gb){
3006
    unsigned int buf;
3007
    int log;
3008
    
3009
    OPEN_READER(re, gb);
3010
    UPDATE_CACHE(re, gb);
3011
    buf=GET_CACHE(re, gb);
3012
    
3013
    log= 32 - av_log2(buf);
3014
#ifdef TRACE
3015
    print_bin(buf>>(32-log), log);
3016
    printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3017
#endif
3018

    
3019
    LAST_SKIP_BITS(re, gb, log);
3020
    CLOSE_READER(re, gb);
3021

    
3022
    return log-1;
3023
}
3024

    
3025
/**
3026
 * decodes a residual block.
3027
 * @param n block index
3028
 * @param scantable scantable
3029
 * @param max_coeff number of coefficients in the block
3030
 * @return <0 if an error occured
3031
 */
3032
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3033
    MpegEncContext * const s = &h->s;
3034
    const uint16_t *qmul= dequant_coeff[qp];
3035
    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};
3036
    int level[16], run[16];
3037
    int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3038

    
3039
    //FIXME put trailing_onex into the context
3040

    
3041
    if(n == CHROMA_DC_BLOCK_INDEX){
3042
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3043
        total_coeff= coeff_token>>2;
3044
    }else{    
3045
        if(n == LUMA_DC_BLOCK_INDEX){
3046
            total_coeff= pred_non_zero_count(h, 0);
3047
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3048
            total_coeff= coeff_token>>2;
3049
        }else{
3050
            total_coeff= pred_non_zero_count(h, n);
3051
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3052
            total_coeff= coeff_token>>2;
3053
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3054
        }
3055
    }
3056

    
3057
    //FIXME set last_non_zero?
3058

    
3059
    if(total_coeff==0)
3060
        return 0;
3061
        
3062
    trailing_ones= coeff_token&3;
3063
    tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3064
    assert(total_coeff<=16);
3065
    
3066
    for(i=0; i<trailing_ones; i++){
3067
        level[i]= 1 - 2*get_bits1(gb);
3068
    }
3069

    
3070
    suffix_length= total_coeff > 10 && trailing_ones < 3;
3071

    
3072
    for(; i<total_coeff; i++){
3073
        const int prefix= get_level_prefix(gb);
3074
        int level_code, mask;
3075

    
3076
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3077
            if(suffix_length)
3078
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3079
            else
3080
                level_code= (prefix<<suffix_length); //part
3081
        }else if(prefix==14){
3082
            if(suffix_length)
3083
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3084
            else
3085
                level_code= prefix + get_bits(gb, 4); //part
3086
        }else if(prefix==15){
3087
            level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3088
            if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3089
        }else{
3090
            fprintf(stderr, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3091
            return -1;
3092
        }
3093

    
3094
        if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3095

    
3096
        mask= -(level_code&1);
3097
        level[i]= (((2+level_code)>>1) ^ mask) - mask;
3098

    
3099
        if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3100

    
3101
#if 1
3102
        if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3103
#else        
3104
        if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3105
        ? == prefix > 2 or sth
3106
#endif
3107
        tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3108
    }
3109

    
3110
    if(total_coeff == max_coeff)
3111
        zeros_left=0;
3112
    else{
3113
        if(n == CHROMA_DC_BLOCK_INDEX)
3114
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3115
        else
3116
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3117
    }
3118
    
3119
    for(i=0; i<total_coeff-1; i++){
3120
        if(zeros_left <=0)
3121
            break;
3122
        else if(zeros_left < 7){
3123
            run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3124
        }else{
3125
            run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3126
        }
3127
        zeros_left -= run[i];
3128
    }
3129

    
3130
    if(zeros_left<0){
3131
        fprintf(stderr, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3132
        return -1;
3133
    }
3134
    
3135
    for(; i<total_coeff-1; i++){
3136
        run[i]= 0;
3137
    }
3138

    
3139
    run[i]= zeros_left;
3140

    
3141
    coeff_num=-1;
3142
    if(n > 24){
3143
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3144
            int j;
3145

    
3146
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3147
            j= scantable[ coeff_num ];
3148

    
3149
            block[j]= level[i];
3150
        }
3151
    }else{
3152
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into  rundecode?
3153
            int j;
3154

    
3155
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3156
            j= scantable[ coeff_num ];
3157

    
3158
            block[j]= level[i] * qmul[j];
3159
//            printf("%d %d  ", block[j], qmul[j]);
3160
        }
3161
    }
3162
    return 0;
3163
}
3164

    
3165
/**
3166
 * decodes a macroblock
3167
 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3168
 */
3169
static int decode_mb(H264Context *h){
3170
    MpegEncContext * const s = &h->s;
3171
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3172
    int mb_type, partition_count, cbp;
3173

    
3174
    memset(h->mb, 0, sizeof(int16_t)*24*16); //FIXME avoid if allready clear (move after skip handlong?
3175

    
3176
    tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3177

    
3178
    if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3179
        if(s->mb_skip_run==-1)
3180
            s->mb_skip_run= get_ue_golomb(&s->gb);
3181
        
3182
        if (s->mb_skip_run--) {
3183
            int i, mx, my;
3184
            /* skip mb */
3185
#if 0 //FIXME
3186
            for(i=0;i<6;i++)
3187
                s->block_last_index[i] = -1;
3188
         s->mv_type = MV_TYPE_16X16;
3189
            /* if P type, zero motion vector is implied */
3190
            s->mv_dir = MV_DIR_FORWARD;
3191
            s->mb_skiped = 1;
3192
#endif
3193
//FIXME b frame
3194
            mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3195

    
3196
            memset(h->non_zero_count[mb_xy], 0, 16);
3197
            memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3198

    
3199
            if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3200
                h->mb_field_decoding_flag= get_bits1(&s->gb);
3201
            }
3202

    
3203
            if(h->mb_field_decoding_flag)
3204
                mb_type|= MB_TYPE_INTERLACED;
3205
            
3206
            fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3207
            pred_pskip_motion(h, &mx, &my);
3208
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3209
            fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, (mx&0xFFFF)+(my<<16), 4);
3210
            write_back_motion(h, mb_type);
3211

    
3212
            s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3213
            h->slice_table[ mb_xy ]= h->slice_num;
3214

    
3215
            h->prev_mb_skiped= 1;
3216
            return 0;
3217
        }
3218
    }
3219
    if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3220
        if((s->mb_y&1)==0)
3221
            h->mb_field_decoding_flag = get_bits1(&s->gb);
3222
    }else
3223
        h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3224
    
3225
    h->prev_mb_skiped= 0;
3226
    
3227
    mb_type= get_ue_golomb(&s->gb);
3228
    if(h->slice_type == B_TYPE){
3229
        if(mb_type < 23){
3230
            partition_count= b_mb_type_info[mb_type].partition_count;
3231
            mb_type=         b_mb_type_info[mb_type].type;
3232
        }else{
3233
            mb_type -= 23;
3234
            goto decode_intra_mb;
3235
        }
3236
    }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3237
        if(mb_type < 5){
3238
            partition_count= p_mb_type_info[mb_type].partition_count;
3239
            mb_type=         p_mb_type_info[mb_type].type;
3240
        }else{
3241
            mb_type -= 5;
3242
            goto decode_intra_mb;
3243
        }
3244
    }else{
3245
       assert(h->slice_type == I_TYPE);
3246
decode_intra_mb:
3247
        if(mb_type > 25){
3248
            fprintf(stderr, "mb_type %d in %c slice to large at %d %d\n", mb_type, ff_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);
3249
            return -1;
3250
        }
3251
        partition_count=0;
3252
        cbp= i_mb_type_info[mb_type].cbp;
3253
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3254
        mb_type= i_mb_type_info[mb_type].type;
3255
    }
3256

    
3257
    if(h->mb_field_decoding_flag)
3258
        mb_type |= MB_TYPE_INTERLACED;
3259

    
3260
    s->current_picture.mb_type[mb_xy]= mb_type;
3261
    h->slice_table[ mb_xy ]= h->slice_num;
3262
    
3263
    if(IS_INTRA_PCM(mb_type)){
3264
        const uint8_t *ptr;
3265
        int x, y, i;
3266
        
3267
        // we assume these blocks are very rare so we dont optimize it
3268
        align_get_bits(&s->gb);
3269
        
3270
        ptr= s->gb.buffer + get_bits_count(&s->gb);
3271
    
3272
        for(y=0; y<16; y++){
3273
            const int index= 4*(y&3) + 64*(y>>2);
3274
            for(x=0; x<16; x++){
3275
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3276
            }
3277
        }
3278
        for(y=0; y<8; y++){
3279
            const int index= 256 + 4*(y&3) + 32*(y>>2);
3280
            for(x=0; x<8; x++){
3281
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3282
            }
3283
        }
3284
        for(y=0; y<8; y++){
3285
            const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3286
            for(x=0; x<8; x++){
3287
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3288
            }
3289
        }
3290
    
3291
        skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3292
        
3293
        memset(h->non_zero_count[mb_xy], 16, 16);
3294
        
3295
        return 0;
3296
    }
3297
        
3298
    fill_caches(h, mb_type);
3299

    
3300
    //mb_pred
3301
    if(IS_INTRA(mb_type)){
3302
//            init_top_left_availability(h);
3303
            if(IS_INTRA4x4(mb_type)){
3304
                int i;
3305

    
3306
//                fill_intra4x4_pred_table(h);
3307
                for(i=0; i<16; i++){
3308
                    const int mode_coded= !get_bits1(&s->gb);
3309
                    const int predicted_mode=  pred_intra_mode(h, i);
3310
                    int mode;
3311

    
3312
                    if(mode_coded){
3313
                        const int rem_mode= get_bits(&s->gb, 3);
3314
                        if(rem_mode<predicted_mode)
3315
                            mode= rem_mode;
3316
                        else
3317
                            mode= rem_mode + 1;
3318
                    }else{
3319
                        mode= predicted_mode;
3320
                    }
3321
                    
3322
                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3323
                }
3324
                write_back_intra_pred_mode(h);
3325
                if( check_intra4x4_pred_mode(h) < 0)
3326
                    return -1;
3327
            }else{
3328
                h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3329
                if(h->intra16x16_pred_mode < 0)
3330
                    return -1;
3331
            }
3332
            h->chroma_pred_mode= get_ue_golomb(&s->gb);
3333

    
3334
            h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3335
            if(h->chroma_pred_mode < 0)
3336
                return -1;
3337
    }else if(partition_count==4){
3338
        int i, j, sub_partition_count[4], list, ref[2][4];
3339
        
3340
        if(h->slice_type == B_TYPE){
3341
            for(i=0; i<4; i++){
3342
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3343
                if(h->sub_mb_type[i] >=13){
3344
                    fprintf(stderr, "B sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
3345
                    return -1;
3346
                }
3347
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3348
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3349
            }
3350
        }else{
3351
            assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3352
            for(i=0; i<4; i++){
3353
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3354
                if(h->sub_mb_type[i] >=4){
3355
                    fprintf(stderr, "P sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
3356
                    return -1;
3357
                }
3358
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3359
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3360
            }
3361
        }
3362
        
3363
        for(list=0; list<2; list++){
3364
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3365
            if(ref_count == 0) continue;
3366
            for(i=0; i<4; i++){
3367
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3368
                    ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3369
                }else{
3370
                 //FIXME
3371
                    ref[list][i] = -1;
3372
                }
3373
            }
3374
        }
3375
        
3376
        for(list=0; list<2; list++){
3377
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3378
            if(ref_count == 0) continue;
3379

    
3380
            for(i=0; i<4; i++){
3381
                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3382
                h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3383

    
3384
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3385
                    const int sub_mb_type= h->sub_mb_type[i];
3386
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3387
                    for(j=0; j<sub_partition_count[i]; j++){
3388
                        int mx, my;
3389
                        const int index= 4*i + block_width*j;
3390
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3391
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3392
                        mx += get_se_golomb(&s->gb);
3393
                        my += get_se_golomb(&s->gb);
3394
                        tprintf("final mv:%d %d\n", mx, my);
3395

    
3396
                        if(IS_SUB_8X8(sub_mb_type)){
3397
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= 
3398
                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3399
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= 
3400
                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3401
                        }else if(IS_SUB_8X4(sub_mb_type)){
3402
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3403
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3404
                        }else if(IS_SUB_4X8(sub_mb_type)){
3405
                            mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3406
                            mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3407
                        }else{
3408
                            assert(IS_SUB_4X4(sub_mb_type));
3409
                            mv_cache[ 0 ][0]= mx;
3410
                            mv_cache[ 0 ][1]= my;
3411
                        }
3412
                    }
3413
                }else{
3414
                    uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3415
                    p[0] = p[1]=
3416
                    p[8] = p[9]= 0;
3417
                }
3418
            }
3419
        }
3420
    }else if(!IS_DIRECT(mb_type)){
3421
        int list, mx, my, i;
3422
         //FIXME we should set ref_idx_l? to 0 if we use that later ...
3423
        if(IS_16X16(mb_type)){
3424
            for(list=0; list<2; list++){
3425
                if(h->ref_count[0]>0){
3426
                    if(IS_DIR(mb_type, 0, list)){
3427
                        const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3428
                        fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3429
                    }
3430
                }
3431
            }
3432
            for(list=0; list<2; list++){
3433
                if(IS_DIR(mb_type, 0, list)){
3434
                    pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3435
                    mx += get_se_golomb(&s->gb);
3436
                    my += get_se_golomb(&s->gb);
3437
                    tprintf("final mv:%d %d\n", mx, my);
3438

    
3439
                    fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, (mx&0xFFFF) + (my<<16), 4);
3440
                }
3441
            }
3442
        }
3443
        else if(IS_16X8(mb_type)){
3444
            for(list=0; list<2; list++){
3445
                if(h->ref_count[list]>0){
3446
                    for(i=0; i<2; i++){
3447
                        if(IS_DIR(mb_type, i, list)){
3448
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3449
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3450
                        }
3451
                    }
3452
                }
3453
            }
3454
            for(list=0; list<2; list++){
3455
                for(i=0; i<2; i++){
3456
                    if(IS_DIR(mb_type, i, list)){
3457
                        pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3458
                        mx += get_se_golomb(&s->gb);
3459
                        my += get_se_golomb(&s->gb);
3460
                        tprintf("final mv:%d %d\n", mx, my);
3461

    
3462
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (mx&0xFFFF) + (my<<16), 4);
3463
                    }
3464
                }
3465
            }
3466
        }else{
3467
            assert(IS_8X16(mb_type));
3468
            for(list=0; list<2; list++){
3469
                if(h->ref_count[list]>0){
3470
                    for(i=0; i<2; i++){
3471
                        if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3472
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3473
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3474
                        }
3475
                    }
3476
                }
3477
            }
3478
            for(list=0; list<2; list++){
3479
                for(i=0; i<2; i++){
3480
                    if(IS_DIR(mb_type, i, list)){
3481
                        pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3482
                        mx += get_se_golomb(&s->gb);
3483
                        my += get_se_golomb(&s->gb);
3484
                        tprintf("final mv:%d %d\n", mx, my);
3485

    
3486
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (mx&0xFFFF) + (my<<16), 4);
3487
                    }
3488
                }
3489
            }
3490
        }
3491
    }
3492
    
3493
    if(IS_INTER(mb_type))
3494
        write_back_motion(h, mb_type);
3495
    
3496
    if(!IS_INTRA16x16(mb_type)){
3497
        cbp= get_ue_golomb(&s->gb);
3498
        if(cbp > 47){
3499
            fprintf(stderr, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3500
            return -1;
3501
        }
3502
        
3503
        if(IS_INTRA4x4(mb_type))
3504
            cbp= golomb_to_intra4x4_cbp[cbp];
3505
        else
3506
            cbp= golomb_to_inter_cbp[cbp];
3507
    }
3508

    
3509
    if(cbp || IS_INTRA16x16(mb_type)){
3510
        int i8x8, i4x4, chroma_idx;
3511
        int chroma_qp, dquant;
3512
        GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3513
        const uint8_t *scan, *dc_scan;
3514
        
3515
//        fill_non_zero_count_cache(h);
3516

    
3517
        if(IS_INTERLACED(mb_type)){
3518
            scan= field_scan;
3519
            dc_scan= luma_dc_field_scan;
3520
        }else{
3521
            scan= zigzag_scan;
3522
            dc_scan= luma_dc_zigzag_scan;
3523
        }
3524

    
3525
        dquant= get_se_golomb(&s->gb);
3526

    
3527
        if( dquant > 25 || dquant < -26 ){
3528
            fprintf(stderr, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3529
            return -1;
3530
        }
3531
        
3532
        s->qscale += dquant;
3533
        if(((unsigned)s->qscale) > 51){
3534
            if(s->qscale<0) s->qscale+= 52;
3535
            else            s->qscale-= 52;
3536
        }
3537
        
3538
        h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3539
        if(IS_INTRA16x16(mb_type)){
3540
            if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3541
                return -1; //FIXME continue if partotioned and other retirn -1 too
3542
            }
3543

    
3544
            assert((cbp&15) == 0 || (cbp&15) == 15);
3545

    
3546
            if(cbp&15){
3547
                for(i8x8=0; i8x8<4; i8x8++){
3548
                    for(i4x4=0; i4x4<4; i4x4++){
3549
                        const int index= i4x4 + 4*i8x8;
3550
                        if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3551
                            return -1;
3552
                        }
3553
                    }
3554
                }
3555
            }else{
3556
                memset(&h->non_zero_count_cache[8], 0, 8*4); //FIXME stupid & slow
3557
            }
3558
        }else{
3559
            for(i8x8=0; i8x8<4; i8x8++){
3560
                if(cbp & (1<<i8x8)){
3561
                    for(i4x4=0; i4x4<4; i4x4++){
3562
                        const int index= i4x4 + 4*i8x8;
3563
                        
3564
                        if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3565
                            return -1;
3566
                        }
3567
                    }
3568
                }else{
3569
                    uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3570
                    nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3571
                }
3572
            }
3573
        }
3574
        
3575
        if(cbp&0x30){
3576
            for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3577
                if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3578
                    return -1;
3579
                }
3580
        }
3581

    
3582
        if(cbp&0x20){
3583
            for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3584
                for(i4x4=0; i4x4<4; i4x4++){
3585
                    const int index= 16 + 4*chroma_idx + i4x4;
3586
                    if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3587
                        return -1;
3588
                    }
3589
                }
3590
            }
3591
        }else{
3592
            uint8_t * const nnz= &h->non_zero_count_cache[0];
3593
            nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3594
            nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3595
        }
3596
    }else{
3597
        memset(&h->non_zero_count_cache[8], 0, 8*5);
3598
    }
3599
    write_back_non_zero_count(h);
3600

    
3601
    return 0;
3602
}
3603

    
3604
static int decode_slice(H264Context *h){
3605
    MpegEncContext * const s = &h->s;
3606
    const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3607

    
3608
    s->mb_skip_run= -1;
3609
    
3610
#if 1
3611
    for(;;){
3612
        int ret= decode_mb(h);
3613
            
3614
        hl_decode_mb(h);
3615
        
3616
        if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
3617
            s->mb_y++;
3618
            ret= decode_mb(h);
3619
            
3620
            hl_decode_mb(h);
3621
            s->mb_y--;
3622
        }
3623

    
3624
        if(ret<0){
3625
            fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3626
            ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3627

    
3628
            return -1;
3629
        }
3630
        
3631
        if(++s->mb_x >= s->mb_width){
3632
            s->mb_x=0;
3633
            ff_draw_horiz_band(s, 16*s->mb_y, 16);
3634
            if(++s->mb_y >= s->mb_height){
3635
                tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3636

    
3637
                if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3638
                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3639

    
3640
                    return 0;
3641
                }else{
3642
                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3643

    
3644
                    return -1;
3645
                }
3646
            }
3647
        }
3648
        
3649
        if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3650
            if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3651
                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3652

    
3653
                return 0;
3654
            }else{
3655
                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3656

    
3657
                return -1;
3658
            }
3659
        }
3660
    }
3661
#endif
3662
#if 0
3663
    for(;s->mb_y < s->mb_height; s->mb_y++){
3664
        for(;s->mb_x < s->mb_width; s->mb_x++){
3665
            int ret= decode_mb(h);
3666
            
3667
            hl_decode_mb(h);
3668

3669
            if(ret<0){
3670
                fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3671
                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3672

3673
                return -1;
3674
            }
3675
        
3676
            if(++s->mb_x >= s->mb_width){
3677
                s->mb_x=0;
3678
                if(++s->mb_y >= s->mb_height){
3679
                    if(get_bits_count(s->gb) == s->gb.size_in_bits){
3680
                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3681

3682
                        return 0;
3683
                    }else{
3684
                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3685

3686
                        return -1;
3687
                    }
3688
                }
3689
            }
3690
        
3691
            if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
3692
                if(get_bits_count(s->gb) == s->gb.size_in_bits){
3693
                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3694

3695
                    return 0;
3696
                }else{
3697
                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3698

3699
                    return -1;
3700
                }
3701
            }
3702
        }
3703
        s->mb_x=0;
3704
        ff_draw_horiz_band(s, 16*s->mb_y, 16);
3705
    }
3706
#endif
3707
    return -1; //not reached
3708
}
3709

    
3710
static inline int decode_vui_parameters(H264Context *h, SPS *sps){
3711
    MpegEncContext * const s = &h->s;
3712
    int aspect_ratio_info_present_flag, aspect_ratio_idc;
3713

    
3714
    aspect_ratio_info_present_flag= get_bits1(&s->gb);
3715
    
3716
    if( aspect_ratio_info_present_flag ) {
3717
        aspect_ratio_idc= get_bits(&s->gb, 8);
3718
        if( aspect_ratio_idc == EXTENDED_SAR ) {
3719
            sps->sar_width= get_bits(&s->gb, 16);
3720
            sps->sar_height= get_bits(&s->gb, 16);
3721
        }else if(aspect_ratio_idc < 16){
3722
            sps->sar_width=  pixel_aspect[aspect_ratio_idc][0];
3723
            sps->sar_height= pixel_aspect[aspect_ratio_idc][1];
3724
        }else{
3725
            fprintf(stderr, "illegal aspect ratio\n");
3726
            return -1;
3727
        }
3728
    }else{
3729
        sps->sar_width= 
3730
        sps->sar_height= 0;
3731
    }
3732
//            s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
3733
#if 0
3734
| overscan_info_present_flag                        |0  |u(1)    |
3735
| if( overscan_info_present_flag )                  |   |        |
3736
|  overscan_appropriate_flag                        |0  |u(1)    |
3737
| video_signal_type_present_flag                    |0  |u(1)    |
3738
| if( video_signal_type_present_flag ) {            |   |        |
3739
|  video_format                                     |0  |u(3)    |
3740
|  video_full_range_flag                            |0  |u(1)    |
3741
|  colour_description_present_flag                  |0  |u(1)    |
3742
|  if( colour_description_present_flag ) {          |   |        |
3743
|   colour_primaries                                |0  |u(8)    |
3744
|   transfer_characteristics                        |0  |u(8)    |
3745
|   matrix_coefficients                             |0  |u(8)    |
3746
|  }                                                |   |        |
3747
| }                                                 |   |        |
3748
| chroma_location_info_present_flag                 |0  |u(1)    |
3749
| if ( chroma_location_info_present_flag ) {        |   |        |
3750
|  chroma_sample_location_type_top_field            |0  |ue(v)   |
3751
|  chroma_sample_location_type_bottom_field         |0  |ue(v)   |
3752
| }                                                 |   |        |
3753
| timing_info_present_flag                          |0  |u(1)    |
3754
| if( timing_info_present_flag ) {                  |   |        |
3755
|  num_units_in_tick                                |0  |u(32)   |
3756
|  time_scale                                       |0  |u(32)   |
3757
|  fixed_frame_rate_flag                            |0  |u(1)    |
3758
| }                                                 |   |        |
3759
| nal_hrd_parameters_present_flag                   |0  |u(1)    |
3760
| if( nal_hrd_parameters_present_flag  = =  1)      |   |        |
3761
|  hrd_parameters( )                                |   |        |
3762
| vcl_hrd_parameters_present_flag                   |0  |u(1)    |
3763
| if( vcl_hrd_parameters_present_flag  = =  1)      |   |        |
3764
|  hrd_parameters( )                                |   |        |
3765
| if( ( nal_hrd_parameters_present_flag  = =  1  | ||   |        |
3766
|                                                   |   |        |
3767
|( vcl_hrd_parameters_present_flag  = =  1 ) )      |   |        |
3768
|  low_delay_hrd_flag                               |0  |u(1)    |
3769
| bitstream_restriction_flag                        |0  |u(1)    |
3770
| if( bitstream_restriction_flag ) {                |0  |u(1)    |
3771
|  motion_vectors_over_pic_boundaries_flag          |0  |u(1)    |
3772
|  max_bytes_per_pic_denom                          |0  |ue(v)   |
3773
|  max_bits_per_mb_denom                            |0  |ue(v)   |
3774
|  log2_max_mv_length_horizontal                    |0  |ue(v)   |
3775
|  log2_max_mv_length_vertical                      |0  |ue(v)   |
3776
|  num_reorder_frames                               |0  |ue(v)   |
3777
|  max_dec_frame_buffering                          |0  |ue(v)   |
3778
| }                                                 |   |        |
3779
|}                                                  |   |        |
3780
#endif
3781
    return 0;
3782
}
3783

    
3784
static inline int decode_seq_parameter_set(H264Context *h){
3785
    MpegEncContext * const s = &h->s;
3786
    int profile_idc, level_idc, multiple_slice_groups, arbitrary_slice_order, redundant_slices;
3787
    int sps_id, i;
3788
    SPS *sps;
3789
    
3790
    profile_idc= get_bits(&s->gb, 8);
3791
    level_idc= get_bits(&s->gb, 8);
3792
    multiple_slice_groups= get_bits1(&s->gb);
3793
    arbitrary_slice_order= get_bits1(&s->gb);
3794
    redundant_slices= get_bits1(&s->gb);
3795
    
3796
    sps_id= get_ue_golomb(&s->gb);
3797
    
3798
    sps= &h->sps_buffer[ sps_id ];
3799
    
3800
    sps->profile_idc= profile_idc;
3801
    sps->level_idc= level_idc;
3802
    sps->multiple_slice_groups= multiple_slice_groups;
3803
    sps->arbitrary_slice_order= arbitrary_slice_order;
3804
    sps->redundant_slices= redundant_slices;
3805
    
3806
    sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
3807

    
3808
    sps->poc_type= get_ue_golomb(&s->gb);
3809
    
3810
    if(sps->poc_type == 0){ //FIXME #define
3811
        sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
3812
    } else if(sps->poc_type == 1){//FIXME #define
3813
        sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
3814
        sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
3815
        sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
3816
        sps->poc_cycle_length= get_ue_golomb(&s->gb);
3817
        
3818
        for(i=0; i<sps->poc_cycle_length; i++)
3819
            sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
3820
    }
3821
    if(sps->poc_type > 2){
3822
        fprintf(stderr, "illegal POC type %d\n", sps->poc_type);
3823
        return -1;
3824
    }
3825

    
3826
    sps->ref_frame_count= get_ue_golomb(&s->gb);
3827
    sps->required_frame_num_update_behaviour_flag= get_bits1(&s->gb);
3828
    sps->mb_width= get_ue_golomb(&s->gb) + 1;
3829
    sps->mb_height= get_ue_golomb(&s->gb) + 1;
3830
    sps->frame_mbs_only_flag= get_bits1(&s->gb);
3831
    if(!sps->frame_mbs_only_flag)
3832
        sps->mb_aff= get_bits1(&s->gb);
3833
    else
3834
        sps->mb_aff= 0;
3835

    
3836
    sps->direct_8x8_inference_flag= get_bits1(&s->gb);
3837

    
3838
    sps->vui_parameters_present_flag= get_bits1(&s->gb);
3839
    if( sps->vui_parameters_present_flag )
3840
        decode_vui_parameters(h, sps);
3841
    
3842
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3843
        printf("sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s %s\n", 
3844
               sps_id, sps->profile_idc, sps->level_idc,
3845
               sps->poc_type,
3846
               sps->ref_frame_count,
3847
               sps->mb_width, sps->mb_height,
3848
               sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
3849
               sps->direct_8x8_inference_flag ? "8B8" : "",
3850
               sps->vui_parameters_present_flag ? "VUI" : ""
3851
               );
3852
    }
3853
    return 0;
3854
}
3855

    
3856
static inline int decode_picture_parameter_set(H264Context *h){
3857
    MpegEncContext * const s = &h->s;
3858
    int pps_id= get_ue_golomb(&s->gb);
3859
    PPS *pps= &h->pps_buffer[pps_id];
3860
    
3861
    pps->sps_id= get_ue_golomb(&s->gb);
3862
    pps->cabac= get_bits1(&s->gb);
3863
    pps->pic_order_present= get_bits1(&s->gb);
3864
    pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
3865
    if(pps->slice_group_count > 1 ){
3866
        pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
3867
fprintf(stderr, "FMO not supported\n");
3868
        switch(pps->mb_slice_group_map_type){
3869
        case 0:
3870
#if 0
3871
|   for( i = 0; i <= num_slice_groups_minus1; i++ ) |   |        |
3872
|    run_length[ i ]                                |1  |ue(v)   |
3873
#endif
3874
            break;
3875
        case 2:
3876
#if 0
3877
|   for( i = 0; i < num_slice_groups_minus1; i++ )  |   |        |
3878
|{                                                  |   |        |
3879
|    top_left_mb[ i ]                               |1  |ue(v)   |
3880
|    bottom_right_mb[ i ]                           |1  |ue(v)   |
3881
|   }                                               |   |        |
3882
#endif
3883
            break;
3884
        case 3:
3885
        case 4:
3886
        case 5:
3887
#if 0
3888
|   slice_group_change_direction_flag               |1  |u(1)    |
3889
|   slice_group_change_rate_minus1                  |1  |ue(v)   |
3890
#endif
3891
            break;
3892
        case 6:
3893
#if 0
3894
|   slice_group_id_cnt_minus1                       |1  |ue(v)   |
3895
|   for( i = 0; i <= slice_group_id_cnt_minus1; i++ |   |        |
3896
|)                                                  |   |        |
3897
|    slice_group_id[ i ]                            |1  |u(v)    |
3898
#endif
3899
        }
3900
    }
3901
    pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3902
    pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3903
    if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
3904
        fprintf(stderr, "reference overflow (pps)\n");
3905
        return -1;
3906
    }
3907
    
3908
    pps->weighted_pred= get_bits1(&s->gb);
3909
    pps->weighted_bipred_idc= get_bits(&s->gb, 2);
3910
    pps->init_qp= get_se_golomb(&s->gb) + 26;
3911
    pps->init_qs= get_se_golomb(&s->gb) + 26;
3912
    pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
3913
    pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
3914
    pps->constrained_intra_pred= get_bits1(&s->gb);
3915
    pps->redundant_pic_cnt_present = get_bits1(&s->gb);
3916
    pps->crop= get_bits1(&s->gb);
3917
    if(pps->crop){
3918
        pps->crop_left  = get_ue_golomb(&s->gb);
3919
        pps->crop_right = get_ue_golomb(&s->gb);
3920
        pps->crop_top   = get_ue_golomb(&s->gb);
3921
        pps->crop_bottom= get_ue_golomb(&s->gb);
3922
    }else{
3923
        pps->crop_left  = 
3924
        pps->crop_right = 
3925
        pps->crop_top   = 
3926
        pps->crop_bottom= 0;
3927
    }
3928
    
3929
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3930
        printf("pps:%d sps:%d %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s crop:%d/%d/%d/%d\n", 
3931
               pps_id, pps->sps_id,
3932
               pps->cabac ? "CABAC" : "CAVLC",
3933
               pps->slice_group_count,
3934
               pps->ref_count[0], pps->ref_count[1],
3935
               pps->weighted_pred ? "weighted" : "",
3936
               pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
3937
               pps->deblocking_filter_parameters_present ? "LPAR" : "",
3938
               pps->constrained_intra_pred ? "CONSTR" : "",
3939
               pps->redundant_pic_cnt_present ? "REDU" : "",
3940
               pps->crop_left, pps->crop_right, 
3941
               pps->crop_top, pps->crop_bottom
3942
               );
3943
    }
3944
    
3945
    return 0;
3946
}
3947

    
3948
/**
3949
 * finds the end of the current frame in the bitstream.
3950
 * @return the position of the first byte of the next frame, or -1
3951
 */
3952
static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){
3953
    ParseContext *pc= &s->parse_context;
3954
    int last_addr, i;
3955
    uint32_t state;
3956
//printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
3957
//    mb_addr= pc->mb_addr - 1;
3958
    state= pc->state;
3959
    //FIXME this will fail with slices
3960
    for(i=0; i<buf_size; i++){
3961
        state= (state<<8) | buf[i];
3962
        if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
3963
            if(pc->frame_start_found){
3964
                pc->state=-1; 
3965
                pc->frame_start_found= 0;
3966
                return i-3;
3967
            }
3968
            pc->frame_start_found= 1;
3969
        }
3970
    }
3971
    
3972
    pc->state= state;
3973
    return -1;
3974
}
3975

    
3976
static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
3977
    MpegEncContext * const s = &h->s;
3978
    AVCodecContext * const avctx= s->avctx;
3979
    int buf_index=0;
3980
    int i;
3981
#if 0    
3982
    for(i=0; i<32; i++){
3983
        printf("%X ", buf[i]);
3984
    }
3985
#endif
3986
    for(;;){
3987
        int consumed;
3988
        int dst_length;
3989
        int bit_length;
3990
        uint8_t *ptr;
3991
        
3992
        // start code prefix search
3993
        for(; buf_index + 3 < buf_size; buf_index++){
3994
            // this should allways succeed in the first iteration
3995
            if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
3996
                break;
3997
        }
3998
        
3999
        if(buf_index+3 >= buf_size) break;
4000
        
4001
        buf_index+=3;
4002
        
4003
        ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index);
4004
        if(ptr[dst_length - 1] == 0) dst_length--;
4005
        bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
4006

    
4007
        if(s->avctx->debug&FF_DEBUG_STARTCODE){
4008
            printf("NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
4009
        }
4010
        
4011
        buf_index += consumed;
4012

    
4013
        if(h->nal_ref_idc < s->hurry_up)
4014
            continue;
4015
        
4016
        switch(h->nal_unit_type){
4017
        case NAL_IDR_SLICE:
4018
            idr(h); //FIXME ensure we dont loose some frames if there is reordering
4019
        case NAL_SLICE:
4020
            init_get_bits(&s->gb, ptr, bit_length);
4021
            h->intra_gb_ptr=
4022
            h->inter_gb_ptr= &s->gb;
4023
            s->data_partitioning = 0;
4024
            
4025
            if(decode_slice_header(h) < 0) return -1;
4026
            if(h->redundant_pic_count==0)
4027
                decode_slice(h);
4028
            break;
4029
        case NAL_DPA:
4030
            init_get_bits(&s->gb, ptr, bit_length);
4031
            h->intra_gb_ptr=
4032
            h->inter_gb_ptr= NULL;
4033
            s->data_partitioning = 1;
4034
            
4035
            if(decode_slice_header(h) < 0) return -1;
4036
            break;
4037
        case NAL_DPB:
4038
            init_get_bits(&h->intra_gb, ptr, bit_length);
4039
            h->intra_gb_ptr= &h->intra_gb;
4040
            break;
4041
        case NAL_DPC:
4042
            init_get_bits(&h->inter_gb, ptr, bit_length);
4043
            h->inter_gb_ptr= &h->inter_gb;
4044

    
4045
            if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning)
4046
                decode_slice(h);
4047
            break;
4048
        case NAL_SEI:
4049
            break;
4050
        case NAL_SPS:
4051
            init_get_bits(&s->gb, ptr, bit_length);
4052
            decode_seq_parameter_set(h);
4053
            
4054
            if(s->flags& CODEC_FLAG_LOW_DELAY)
4055
                s->low_delay=1;
4056
      
4057
            avctx->has_b_frames= !s->low_delay;
4058
            break;
4059
        case NAL_PPS:
4060
            init_get_bits(&s->gb, ptr, bit_length);
4061
            
4062
            decode_picture_parameter_set(h);
4063

    
4064
            break;
4065
        case NAL_PICTURE_DELIMITER:
4066
            break;
4067
        case NAL_FILTER_DATA:
4068
            break;
4069
        }        
4070

    
4071
        //FIXME move after where irt is set
4072
        s->current_picture.pict_type= s->pict_type;
4073
        s->current_picture.key_frame= s->pict_type == I_TYPE;
4074
    }
4075
    
4076
    if(!s->current_picture_ptr) return buf_index; //no frame
4077
    
4078
    h->prev_frame_num_offset= h->frame_num_offset;
4079
    h->prev_frame_num= h->frame_num;
4080
    if(s->current_picture_ptr->reference){
4081
        h->prev_poc_msb= h->poc_msb;
4082
        h->prev_poc_lsb= h->poc_lsb;
4083
    }
4084
    if(s->current_picture_ptr->reference)
4085
        execute_ref_pic_marking(h, h->mmco, h->mmco_index);
4086
    else
4087
        assert(h->mmco_index==0);
4088

    
4089
    ff_er_frame_end(s);
4090
    MPV_frame_end(s);
4091

    
4092
    return buf_index;
4093
}
4094

    
4095
/**
4096
 * retunrs the number of bytes consumed for building the current frame
4097
 */
4098
static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
4099
    if(s->flags&CODEC_FLAG_TRUNCATED){
4100
        pos -= s->parse_context.last_index;
4101
        if(pos<0) pos=0; // FIXME remove (uneeded?)
4102
        
4103
        return pos;
4104
    }else{
4105
        if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
4106
        if(pos+10>buf_size) pos=buf_size; // oops ;)
4107

    
4108
        return pos;
4109
    }
4110
}
4111

    
4112
static int decode_frame(AVCodecContext *avctx, 
4113
                             void *data, int *data_size,
4114
                             uint8_t *buf, int buf_size)
4115
{
4116
    H264Context *h = avctx->priv_data;
4117
    MpegEncContext *s = &h->s;
4118
    AVFrame *pict = data; 
4119
    float new_aspect;
4120
    int buf_index;
4121
    
4122
    s->flags= avctx->flags;
4123

    
4124
    *data_size = 0;
4125
   
4126
   /* no supplementary picture */
4127
    if (buf_size == 0) {
4128
        return 0;
4129
    }
4130
    
4131
    if(s->flags&CODEC_FLAG_TRUNCATED){
4132
        int next= find_frame_end(s, buf, buf_size);
4133
        
4134
        if( ff_combine_frame(s, next, &buf, &buf_size) < 0 )
4135
            return buf_size;
4136
//printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
4137
    }
4138

    
4139
    if(s->avctx->extradata_size && s->picture_number==0){
4140
        if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) ) 
4141
            return -1;
4142
    }
4143

    
4144
    buf_index=decode_nal_units(h, buf, buf_size);
4145
    if(buf_index < 0) 
4146
        return -1;
4147

    
4148
    //FIXME do something with unavailable reference frames    
4149
 
4150
//    if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
4151
#if 0
4152
    if(s->pict_type==B_TYPE || s->low_delay){
4153
        *pict= *(AVFrame*)&s->current_picture;
4154
    } else {
4155
        *pict= *(AVFrame*)&s->last_picture;
4156
    }
4157
#endif
4158
    if(!s->current_picture_ptr){
4159
        fprintf(stderr, "error, NO frame\n");
4160
        return -1;
4161
    }
4162

    
4163
    *pict= *(AVFrame*)&s->current_picture; //FIXME 
4164
    ff_print_debug_info(s, s->current_picture_ptr);
4165
    assert(pict->data[0]);
4166
//printf("out %d\n", (int)pict->data[0]);
4167
#if 0 //?
4168

4169
    /* Return the Picture timestamp as the frame number */
4170
    /* we substract 1 because it is added on utils.c    */
4171
    avctx->frame_number = s->picture_number - 1;
4172
#endif
4173
#if 0
4174
    /* dont output the last pic after seeking */
4175
    if(s->last_picture_ptr || s->low_delay)
4176
    //Note this isnt a issue as a IDR pic should flush teh buffers
4177
#endif
4178
        *data_size = sizeof(AVFrame);
4179
    return get_consumed_bytes(s, buf_index, buf_size);
4180
}
4181
#if 0
4182
static inline void fill_mb_avail(H264Context *h){
4183
    MpegEncContext * const s = &h->s;
4184
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4185

4186
    if(s->mb_y){
4187
        h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
4188
        h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
4189
        h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
4190
    }else{
4191
        h->mb_avail[0]=
4192
        h->mb_avail[1]=
4193
        h->mb_avail[2]= 0;
4194
    }
4195
    h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
4196
    h->mb_avail[4]= 1; //FIXME move out
4197
    h->mb_avail[5]= 0; //FIXME move out
4198
}
4199
#endif
4200

    
4201
#if 0 //selftest
4202
#define COUNT 8000
4203
#define SIZE (COUNT*40)
4204
int main(){
4205
    int i;
4206
    uint8_t temp[SIZE];
4207
    PutBitContext pb;
4208
    GetBitContext gb;
4209
//    int int_temp[10000];
4210
    DSPContext dsp;
4211
    AVCodecContext avctx;
4212
    
4213
    dsputil_init(&dsp, &avctx);
4214

4215
    init_put_bits(&pb, temp, SIZE, NULL, NULL);
4216
    printf("testing unsigned exp golomb\n");
4217
    for(i=0; i<COUNT; i++){
4218
        START_TIMER
4219
        set_ue_golomb(&pb, i);
4220
        STOP_TIMER("set_ue_golomb");
4221
    }
4222
    flush_put_bits(&pb);
4223
    
4224
    init_get_bits(&gb, temp, 8*SIZE);
4225
    for(i=0; i<COUNT; i++){
4226
        int j, s;
4227
        
4228
        s= show_bits(&gb, 24);
4229
        
4230
        START_TIMER
4231
        j= get_ue_golomb(&gb);
4232
        if(j != i){
4233
            printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4234
//            return -1;
4235
        }
4236
        STOP_TIMER("get_ue_golomb");
4237
    }
4238
    
4239
    
4240
    init_put_bits(&pb, temp, SIZE, NULL, NULL);
4241
    printf("testing signed exp golomb\n");
4242
    for(i=0; i<COUNT; i++){
4243
        START_TIMER
4244
        set_se_golomb(&pb, i - COUNT/2);
4245
        STOP_TIMER("set_se_golomb");
4246
    }
4247
    flush_put_bits(&pb);
4248
    
4249
    init_get_bits(&gb, temp, 8*SIZE);
4250
    for(i=0; i<COUNT; i++){
4251
        int j, s;
4252
        
4253
        s= show_bits(&gb, 24);
4254
        
4255
        START_TIMER
4256
        j= get_se_golomb(&gb);
4257
        if(j != i - COUNT/2){
4258
            printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4259
//            return -1;
4260
        }
4261
        STOP_TIMER("get_se_golomb");
4262
    }
4263

4264
    printf("testing 4x4 (I)DCT\n");
4265
    
4266
    DCTELEM block[16];
4267
    uint8_t src[16], ref[16];
4268
    uint64_t error= 0, max_error=0;
4269

4270
    for(i=0; i<COUNT; i++){
4271
        int j;
4272
//        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
4273
        for(j=0; j<16; j++){
4274
            ref[j]= random()%255;
4275
            src[j]= random()%255;
4276
        }
4277

4278
        h264_diff_dct_c(block, src, ref, 4);
4279
        
4280
        //normalize
4281
        for(j=0; j<16; j++){
4282
//            printf("%d ", block[j]);
4283
            block[j]= block[j]*4;
4284
            if(j&1) block[j]= (block[j]*4 + 2)/5;
4285
            if(j&4) block[j]= (block[j]*4 + 2)/5;
4286
        }
4287
//        printf("\n");
4288
        
4289
        h264_add_idct_c(ref, block, 4);
4290
/*        for(j=0; j<16; j++){
4291
            printf("%d ", ref[j]);
4292
        }
4293
        printf("\n");*/
4294
            
4295
        for(j=0; j<16; j++){
4296
            int diff= ABS(src[j] - ref[j]);
4297
            
4298
            error+= diff*diff;
4299
            max_error= FFMAX(max_error, diff);
4300
        }
4301
    }
4302
    printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
4303
#if 0
4304
    printf("testing quantizer\n");
4305
    for(qp=0; qp<52; qp++){
4306
        for(i=0; i<16; i++)
4307
            src1_block[i]= src2_block[i]= random()%255;
4308
        
4309
    }
4310
#endif
4311
    printf("Testing NAL layer\n");
4312
    
4313
    uint8_t bitstream[COUNT];
4314
    uint8_t nal[COUNT*2];
4315
    H264Context h;
4316
    memset(&h, 0, sizeof(H264Context));
4317
    
4318
    for(i=0; i<COUNT; i++){
4319
        int zeros= i;
4320
        int nal_length;
4321
        int consumed;
4322
        int out_length;
4323
        uint8_t *out;
4324
        int j;
4325
        
4326
        for(j=0; j<COUNT; j++){
4327
            bitstream[j]= (random() % 255) + 1;
4328
        }
4329
        
4330
        for(j=0; j<zeros; j++){
4331
            int pos= random() % COUNT;
4332
            while(bitstream[pos] == 0){
4333
                pos++;
4334
                pos %= COUNT;
4335
            }
4336
            bitstream[pos]=0;
4337
        }
4338
        
4339
        START_TIMER
4340
        
4341
        nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
4342
        if(nal_length<0){
4343
            printf("encoding failed\n");
4344
            return -1;
4345
        }
4346
        
4347
        out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
4348

    
4349
        STOP_TIMER("NAL")
4350
        
4351
        if(out_length != COUNT){
4352
            printf("incorrect length %d %d\n", out_length, COUNT);
4353
            return -1;
4354
        }
4355
        
4356
        if(consumed != nal_length){
4357
            printf("incorrect consumed length %d %d\n", nal_length, consumed);
4358
            return -1;
4359
        }
4360
        
4361
        if(memcmp(bitstream, out, COUNT)){
4362
            printf("missmatch\n");
4363
            return -1;
4364
        }
4365
    }
4366
    
4367
    printf("Testing RBSP\n");
4368
    
4369
    
4370
    return 0;
4371
}
4372
#endif
4373

    
4374

    
4375
static int decode_end(AVCodecContext *avctx)
4376
{
4377
    H264Context *h = avctx->priv_data;
4378
    MpegEncContext *s = &h->s;
4379
    
4380
    free_tables(h); //FIXME cleanup init stuff perhaps
4381
    MPV_common_end(s);
4382

    
4383
//    memset(h, 0, sizeof(H264Context));
4384
        
4385
    return 0;
4386
}
4387

    
4388

    
4389
AVCodec h264_decoder = {
4390
    "h264",
4391
    CODEC_TYPE_VIDEO,
4392
    CODEC_ID_H264,
4393
    sizeof(H264Context),
4394
    decode_init,
4395
    NULL,
4396
    decode_end,
4397
    decode_frame,
4398
    /*CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_DR1 | */CODEC_CAP_TRUNCATED,
4399
};
4400