Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 0da71265

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 mb_stride; ///< stride of some mb tables
153

    
154
    int chroma_qp; //QPc
155

    
156
    int prev_mb_skiped; //FIXME remove (IMHO not used)
157

    
158
    //prediction stuff
159
    int chroma_pred_mode;
160
    int intra16x16_pred_mode;
161
    
162
    int8_t intra4x4_pred_mode_cache[5*8];
163
    int8_t (*intra4x4_pred_mode)[8];
164
    void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
165
    void (*pred8x8  [4+3])(uint8_t *src, int stride);
166
    void (*pred16x16[4+3])(uint8_t *src, int stride);
167
    unsigned int topleft_samples_available;
168
    unsigned int top_samples_available;
169
    unsigned int topright_samples_available;
170
    unsigned int left_samples_available;
171

    
172
    /**
173
     * non zero coeff count cache.
174
     * is 64 if not available.
175
     */
176
    uint8_t non_zero_count_cache[6*8];
177
    uint8_t (*non_zero_count)[16];
178

    
179
    /**
180
     * Motion vector cache.
181
     */
182
    int16_t mv_cache[2][5*8][2];
183
    int8_t ref_cache[2][5*8];
184
#define LIST_NOT_USED -1 //FIXME rename?
185
#define PART_NOT_AVAILABLE -2
186
    
187
    /**
188
     * is 1 if the specific list MV&references are set to 0,0,-2.
189
     */
190
    int mv_cache_clean[2];
191

    
192
    int block_offset[16+8];
193
    int chroma_subblock_offset[16]; //FIXME remove
194
    
195
    uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
196
    uint16_t *mb2b8_xy;
197
    int b_stride;
198
    int b8_stride;
199

    
200
    SPS sps_buffer[MAX_SPS_COUNT];
201
    SPS sps; ///< current sps
202
    
203
    PPS pps_buffer[MAX_PPS_COUNT];
204
    /**
205
     * current pps
206
     */
207
    PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
208

    
209
    int slice_num;
210
    uint8_t *slice_table_base;
211
    uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
212
    int slice_type;
213
    int slice_type_fixed;
214
    
215
    //interlacing specific flags
216
    int mb_field_decoding_flag;
217
    
218
    int sub_mb_type[4];
219
    
220
    //POC stuff
221
    int poc_lsb;
222
    int poc_msb;
223
    int delta_poc_bottom;
224
    int delta_poc[2];
225
    int frame_num;
226
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
227
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
228
    int frame_num_offset;         ///< for POC type 2
229
    int prev_frame_num_offset;    ///< for POC type 2
230
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
231

    
232
    /**
233
     * frame_num for frames or 2*frame_num for field pics.
234
     */
235
    int curr_pic_num;
236
    
237
    /**
238
     * max_frame_num or 2*max_frame_num for field pics.
239
     */
240
    int max_pic_num;
241

    
242
    //Weighted pred stuff
243
    int luma_log2_weight_denom;
244
    int chroma_log2_weight_denom;
245
    int luma_weight[2][16];
246
    int luma_offset[2][16];
247
    int chroma_weight[2][16][2];
248
    int chroma_offset[2][16][2];
249
   
250
    //deblock
251
    int disable_deblocking_filter_idc;
252
    int slice_alpha_c0_offset_div2;
253
    int slice_beta_offset_div2;
254
     
255
    int redundant_pic_count;
256
    
257
    int direct_spatial_mv_pred;
258

    
259
    /**
260
     * num_ref_idx_l0/1_active_minus1 + 1
261
     */
262
    int ref_count[2];// FIXME split for AFF
263
    Picture *short_ref[16];
264
    Picture *long_ref[16];
265
    Picture default_ref_list[2][32];
266
    Picture ref_list[2][32]; //FIXME size?
267
    Picture field_ref_list[2][32]; //FIXME size?
268
    
269
    /**
270
     * memory management control operations buffer.
271
     */
272
    MMCO mmco[MAX_MMCO_COUNT];
273
    int mmco_index;
274
    
275
    int long_ref_count;  ///< number of actual long term references
276
    int short_ref_count; ///< number of actual short term references
277
    
278
    //data partitioning
279
    GetBitContext intra_gb;
280
    GetBitContext inter_gb;
281
    GetBitContext *intra_gb_ptr;
282
    GetBitContext *inter_gb_ptr;
283
    
284
    DCTELEM mb[16*24] __align8;
285
}H264Context;
286

    
287
static VLC coeff_token_vlc[4];
288
static VLC chroma_dc_coeff_token_vlc;
289

    
290
static VLC total_zeros_vlc[15];
291
static VLC chroma_dc_total_zeros_vlc[3];
292

    
293
static VLC run_vlc[6];
294
static VLC run7_vlc;
295

    
296
/**
297
 * fill a rectangle.
298
 * @param h height of the recatangle, should be a constant
299
 * @param w width of the recatangle, should be a constant
300
 * @param size the size of val (1 or 4), should be a constant
301
 */
302
static inline void fill_rectangle(void *p, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
303
    assert(size==1 || size==4);
304
    
305
    w      *= size;
306
    stride *= size;
307
    
308
//FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
309
    if(w==2 && h==2){
310
        *(uint16_t*)(p + 0)=
311
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
312
    }else if(w==2 && h==4){
313
        *(uint16_t*)(p + 0*stride)=
314
        *(uint16_t*)(p + 1*stride)=
315
        *(uint16_t*)(p + 2*stride)=
316
        *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
317
    }else if(w==4 && h==2){
318
        *(uint32_t*)(p + 0*stride)=
319
        *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
320
    }else if(w==4 && h==4){
321
        *(uint32_t*)(p + 0*stride)=
322
        *(uint32_t*)(p + 1*stride)=
323
        *(uint32_t*)(p + 2*stride)=
324
        *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
325
    }else if(w==8 && h==1){
326
        *(uint32_t*)(p + 0)=
327
        *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
328
    }else if(w==8 && h==2){
329
        *(uint32_t*)(p + 0 + 0*stride)=
330
        *(uint32_t*)(p + 4 + 0*stride)=
331
        *(uint32_t*)(p + 0 + 1*stride)=
332
        *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
333
    }else if(w==8 && h==4){
334
        *(uint64_t*)(p + 0*stride)=
335
        *(uint64_t*)(p + 1*stride)=
336
        *(uint64_t*)(p + 2*stride)=
337
        *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
338
    }else if(w==16 && h==2){
339
        *(uint64_t*)(p + 0+0*stride)=
340
        *(uint64_t*)(p + 8+0*stride)=
341
        *(uint64_t*)(p + 0+1*stride)=
342
        *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
343
    }else if(w==16 && h==4){
344
        *(uint64_t*)(p + 0+0*stride)=
345
        *(uint64_t*)(p + 8+0*stride)=
346
        *(uint64_t*)(p + 0+1*stride)=
347
        *(uint64_t*)(p + 8+1*stride)=
348
        *(uint64_t*)(p + 0+2*stride)=
349
        *(uint64_t*)(p + 8+2*stride)=
350
        *(uint64_t*)(p + 0+3*stride)=
351
        *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
352
    }else
353
        assert(0);
354
}
355

    
356
static inline void fill_caches(H264Context *h, int mb_type){
357
    MpegEncContext * const s = &h->s;
358
    const int mb_xy= s->mb_x + s->mb_y*h->mb_stride;
359
    int topleft_xy, top_xy, topright_xy, left_xy[2];
360
    int topleft_type, top_type, topright_type, left_type[2];
361
    int left_block[4];
362
    int i;
363

    
364
    //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it 
365
    
366
    if(h->sps.mb_aff){
367
    //FIXME
368
    }else{
369
        topleft_xy = mb_xy-1 - h->mb_stride;
370
        top_xy     = mb_xy   - h->mb_stride;
371
        topright_xy= mb_xy+1 - h->mb_stride;
372
        left_xy[0]   = mb_xy-1;
373
        left_xy[1]   = mb_xy-1;
374
        left_block[0]= 0;
375
        left_block[1]= 1;
376
        left_block[2]= 2;
377
        left_block[3]= 3;
378
    }
379

    
380
    topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
381
    top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
382
    topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
383
    left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
384
    left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
385

    
386
    if(IS_INTRA(mb_type)){
387
        h->topleft_samples_available= 
388
        h->top_samples_available= 
389
        h->left_samples_available= 0xFFFF;
390
        h->topright_samples_available= 0xEEEA;
391

    
392
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
393
            h->topleft_samples_available= 0xB3FF;
394
            h->top_samples_available= 0x33FF;
395
            h->topright_samples_available= 0x26EA;
396
        }
397
        for(i=0; i<2; i++){
398
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
399
                h->topleft_samples_available&= 0xDF5F;
400
                h->left_samples_available&= 0x5F5F;
401
            }
402
        }
403
        
404
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
405
            h->topleft_samples_available&= 0x7FFF;
406
        
407
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
408
            h->topright_samples_available&= 0xFBFF;
409
    
410
        if(IS_INTRA4x4(mb_type)){
411
            if(IS_INTRA4x4(top_type)){
412
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
413
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
414
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
415
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
416
            }else{
417
                int pred;
418
                if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
419
                    pred= 2;
420
                else{
421
                    pred= -1;
422
                }
423
                h->intra4x4_pred_mode_cache[4+8*0]=
424
                h->intra4x4_pred_mode_cache[5+8*0]=
425
                h->intra4x4_pred_mode_cache[6+8*0]=
426
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
427
            }
428
            for(i=0; i<2; i++){
429
                if(IS_INTRA4x4(left_type[i])){
430
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
431
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
432
                }else{
433
                    int pred;
434
                    if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
435
                        pred= 2;
436
                    else{
437
                        pred= -1;
438
                    }
439
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
440
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
441
                }
442
            }
443
        }
444
    }
445
    
446
    
447
/*
448
0 . T T. T T T T 
449
1 L . .L . . . . 
450
2 L . .L . . . . 
451
3 . T TL . . . . 
452
4 L . .L . . . . 
453
5 L . .. . . . . 
454
*/
455
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
456
    if(top_type){
457
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
458
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
459
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
460
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
461
    
462
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
463
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
464
    
465
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
466
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
467
    }else{
468
        h->non_zero_count_cache[4+8*0]=      
469
        h->non_zero_count_cache[5+8*0]=
470
        h->non_zero_count_cache[6+8*0]=
471
        h->non_zero_count_cache[7+8*0]=
472
    
473
        h->non_zero_count_cache[1+8*0]=
474
        h->non_zero_count_cache[2+8*0]=
475
    
476
        h->non_zero_count_cache[1+8*3]=
477
        h->non_zero_count_cache[2+8*3]= 64;
478
    }
479
    
480
    if(left_type[0]){
481
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
482
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
483
        h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
484
        h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
485
    }else{
486
        h->non_zero_count_cache[3+8*1]= 
487
        h->non_zero_count_cache[3+8*2]= 
488
        h->non_zero_count_cache[0+8*1]= 
489
        h->non_zero_count_cache[0+8*4]= 64;
490
    }
491
    
492
    if(left_type[1]){
493
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
494
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
495
        h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
496
        h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
497
    }else{
498
        h->non_zero_count_cache[3+8*3]= 
499
        h->non_zero_count_cache[3+8*4]= 
500
        h->non_zero_count_cache[0+8*2]= 
501
        h->non_zero_count_cache[0+8*5]= 64;
502
    }
503
    
504
#if 1
505
    if(IS_INTER(mb_type)){
506
        int list;
507
        for(list=0; list<2; list++){
508
            if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
509
                /*if(!h->mv_cache_clean[list]){
510
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
511
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
512
                    h->mv_cache_clean[list]= 1;
513
                }*/
514
                continue; //FIXME direct mode ...
515
            }
516
            h->mv_cache_clean[list]= 0;
517
            
518
            if(IS_INTER(topleft_type)){
519
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
520
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
521
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
522
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
523
            }else{
524
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
525
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
526
            }
527
            
528
            if(IS_INTER(top_type)){
529
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
530
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
531
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
532
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
533
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
534
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
535
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
536
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
537
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
538
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
539
            }else{
540
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
541
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
542
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
543
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
544
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
545
            }
546

    
547
            if(IS_INTER(topright_type)){
548
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
549
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
550
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
551
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
552
            }else{
553
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
554
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
555
            }
556
            
557
            //FIXME unify cleanup or sth
558
            if(IS_INTER(left_type[0])){
559
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
560
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
561
                *(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]];
562
                *(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]];
563
                h->ref_cache[list][scan8[0] - 1 + 0*8]= 
564
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
565
            }else{
566
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
567
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
568
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
569
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
570
            }
571
            
572
            if(IS_INTER(left_type[1])){
573
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
574
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
575
                *(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]];
576
                *(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]];
577
                h->ref_cache[list][scan8[0] - 1 + 2*8]= 
578
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
579
            }else{
580
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
581
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
582
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
583
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
584
            }
585

    
586
            h->ref_cache[list][scan8[5 ]+1] = 
587
            h->ref_cache[list][scan8[7 ]+1] = 
588
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewher else)
589
            h->ref_cache[list][scan8[4 ]] = 
590
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
591
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
592
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
593
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
594
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
595
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
596
        }
597
//FIXME
598

    
599
    }
600
#endif
601
}
602

    
603
static inline void write_back_intra_pred_mode(H264Context *h){
604
    MpegEncContext * const s = &h->s;
605
    const int mb_xy= s->mb_x + s->mb_y*h->mb_stride;
606

    
607
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
608
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
609
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
610
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
611
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
612
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
613
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
614
}
615

    
616
/**
617
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
618
 */
619
static inline int check_intra4x4_pred_mode(H264Context *h){
620
    MpegEncContext * const s = &h->s;
621
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
622
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
623
    int i;
624
    
625
    if(!(h->top_samples_available&0x8000)){
626
        for(i=0; i<4; i++){
627
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
628
            if(status<0){
629
                fprintf(stderr, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
630
                return -1;
631
            } else if(status){
632
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
633
            }
634
        }
635
    }
636
    
637
    if(!(h->left_samples_available&0x8000)){
638
        for(i=0; i<4; i++){
639
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
640
            if(status<0){
641
                fprintf(stderr, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
642
                return -1;
643
            } else if(status){
644
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
645
            }
646
        }
647
    }
648

    
649
    return 0;
650
} //FIXME cleanup like next
651

    
652
/**
653
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
654
 */
655
static inline int check_intra_pred_mode(H264Context *h, int mode){
656
    MpegEncContext * const s = &h->s;
657
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
658
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
659
    
660
    if(!(h->top_samples_available&0x8000)){
661
        mode= top[ mode ];
662
        if(mode<0){
663
            fprintf(stderr, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
664
            return -1;
665
        }
666
    }
667
    
668
    if(!(h->left_samples_available&0x8000)){
669
        mode= left[ mode ];
670
        if(mode<0){
671
            fprintf(stderr, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
672
            return -1;
673
        } 
674
    }
675

    
676
    return mode;
677
}
678

    
679
/**
680
 * gets the predicted intra4x4 prediction mode.
681
 */
682
static inline int pred_intra_mode(H264Context *h, int n){
683
    const int index8= scan8[n];
684
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
685
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
686
    const int min= FFMIN(left, top);
687

    
688
#ifdef TRACE
689
printf("mode:%d %d min:%d\n", left ,top, min);
690
#endif
691

    
692
    if(min<0) return DC_PRED;
693
    else      return min;
694
}
695

    
696
static inline void write_back_non_zero_count(H264Context *h){
697
    MpegEncContext * const s = &h->s;
698
    const int mb_xy= s->mb_x + s->mb_y*h->mb_stride;
699

    
700
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
701
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
702
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
703
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
704
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
705
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
706
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
707
    
708
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
709
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
710
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
711

    
712
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
713
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
714
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
715
}
716

    
717
/**
718
 * gets the predicted number of non zero coefficients.
719
 * @param n block index
720
 */
721
static inline int pred_non_zero_count(H264Context *h, int n){
722
    const int index8= scan8[n];
723
    const int left= h->non_zero_count_cache[index8 - 1];
724
    const int top = h->non_zero_count_cache[index8 - 8];
725
    int i= left + top;
726
    
727
    if(i<64) i= (i+1)>>1;
728

    
729
#ifdef TRACE
730
    printf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
731
#endif
732

    
733
    return i&31;
734
}
735

    
736
/**
737
 * gets the predicted MV.
738
 * @param n the block index
739
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
740
 * @param mx the x component of the predicted motion vector
741
 * @param my the y component of the predicted motion vector
742
 */
743
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
744
    MpegEncContext * const s = &h->s;
745
    const int index8= scan8[n];
746
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
747
    const int topright_ref= h->ref_cache[list][ index8 - 8 + part_width ];
748
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
749
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
750
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
751
    const int16_t * const C= h->mv_cache[list][ index8 - 8 + part_width ];
752
    
753
    assert(part_width==1 || part_width==2 || part_width==4);
754
    
755
/* mv_cache
756
  B . . A T T T T 
757
  U . . L . . , .
758
  U . . L . . . .
759
  U . . L . . , .
760
  . . . L . . . .
761
*/
762
    if(topright_ref != PART_NOT_AVAILABLE){
763
        if((topright_ref==ref) + (top_ref==ref) + (left_ref==ref) == 1){
764
            *mx= A[0] + B[0] + C[0];
765
            *my= A[1] + B[1] + C[1];
766
        }else{
767
            *mx= mid_pred(A[0], B[0], C[0]);
768
            *my= mid_pred(A[1], B[1], C[1]);
769
        }
770
    }else{
771
        const int topleft_ref= h->ref_cache[list][ index8 - 9 ];
772
        const int16_t * const D= h->mv_cache[list][ index8 - 9 ];
773
        if(top_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){ //FIXME check rare FMO case where std isnt clear
774
            *mx= A[0];
775
            *my= A[1];
776
        }else{
777
            if((topleft_ref==ref) + (top_ref==ref) + (left_ref==ref) == 1){
778
                *mx= A[0] + B[0] + D[0];
779
                *my= A[1] + B[1] + D[1];
780
            }else{
781
                *mx= mid_pred(A[0], B[0], D[0]);
782
                *my= mid_pred(A[1], B[1], D[1]);
783
            }
784
        }
785
        
786
#ifdef TRACE
787
printf("topleft: %2d %2d %2d ", topleft_ref, D[0], D[1]);
788
#endif
789
    }
790
#ifdef TRACE
791
printf("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], topright_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, s->mb_x, s->mb_y, n, list);
792
#endif
793
}
794

    
795
/**
796
 * gets the directionally predicted 16x8 MV.
797
 * @param n the block index
798
 * @param mx the x component of the predicted motion vector
799
 * @param my the y component of the predicted motion vector
800
 */
801
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
802
    MpegEncContext * const s = &h->s;
803
    if(n==0){
804
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
805
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
806

    
807
#ifdef TRACE
808
printf("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);
809
#endif
810
        
811
        if(top_ref == ref){
812
            *mx= B[0];
813
            *my= B[1];
814
            return;
815
        }
816
    }else{
817
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
818
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
819
        
820
#ifdef TRACE
821
printf("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);
822
#endif
823

    
824
        if(left_ref == ref){
825
            *mx= A[0];
826
            *my= A[1];
827
            return;
828
        }
829
    }
830

    
831
    //RARE
832
    pred_motion(h, n, 4, list, ref, mx, my);
833
}
834

    
835
/**
836
 * gets the directionally predicted 8x16 MV.
837
 * @param n the block index
838
 * @param mx the x component of the predicted motion vector
839
 * @param my the y component of the predicted motion vector
840
 */
841
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
842
    MpegEncContext * const s = &h->s;
843
    if(n==0){
844
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
845
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
846
        
847
#ifdef TRACE
848
printf("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);
849
#endif
850

    
851
        if(left_ref == ref){
852
            *mx= A[0];
853
            *my= A[1];
854
            return;
855
        }
856
    }else{
857
        const int topright_ref= h->ref_cache[list][ scan8[4] - 8 + 2 ];
858
        const int16_t * const C= h->mv_cache[list][ scan8[4] - 8 + 2 ];
859
        
860
#ifdef TRACE
861
printf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", topright_ref, C[0], C[1], s->mb_x, s->mb_y, n, list);
862
#endif
863

    
864
        if(topright_ref == ref){ 
865
            *mx= C[0];
866
            *my= C[1];
867
            return;
868
        }
869
        
870
        if(topright_ref == PART_NOT_AVAILABLE){ //insanity ...
871
            const int topleft_ref= h->ref_cache[list][ scan8[4] - 9 ];
872
            const int16_t * const D= h->mv_cache[list][ scan8[4] - 9 ];
873

    
874
#ifdef TRACE
875
printf("pred_8x16: insanity (%2d %2d %2d) at %2d %2d %d list %d", topleft_ref, D[0], D[1], s->mb_x, s->mb_y, n, list);
876
#endif
877
            if(topleft_ref == ref){ 
878
                *mx= D[0];
879
                *my= D[1];
880
                return;
881
            }
882
        }
883
    }
884

    
885
    //RARE
886
    pred_motion(h, n, 2, list, ref, mx, my);
887
}
888

    
889
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
890
    MpegEncContext * const s = &h->s;
891
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
892
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
893

    
894
#ifdef TRACE
895
printf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, s->mb_x, s->mb_y);
896
#endif
897

    
898
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
899
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
900
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
901
       
902
        *mx = *my = 0;
903
        return;
904
    }
905
        
906
    pred_motion(h, 0, 4, 0, 0, mx, my);
907

    
908
    return;
909
}
910

    
911
static inline void write_back_motion(H264Context *h, int mb_type){
912
    MpegEncContext * const s = &h->s;
913
    const int mb_xy= s->mb_x + s->mb_y*h->mb_stride;
914
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
915
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
916
    int list;
917

    
918
    for(list=0; list<2; list++){
919
        int y;
920
        if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
921
            if(1){ //FIXME skip or never read if mb_type doesnt use it
922
                for(y=0; y<4; y++){
923
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
924
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
925
                }
926
                for(y=0; y<2; y++){
927
                    *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
928
                }
929
            }
930
            continue; //FIXME direct mode ...
931
        }
932
        
933
        for(y=0; y<4; y++){
934
            *(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];
935
            *(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];
936
        }
937
        for(y=0; y<2; y++){
938
            s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
939
            s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
940
        }
941
    }
942
}
943

    
944
/**
945
 * Decodes a network abstraction layer unit.
946
 * @param consumed is the number of bytes used as input
947
 * @param length is the length of the array
948
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
949
 * @returns decoded bytes, might be src+1 if no escapes 
950
 */
951
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
952
    int i, si, di;
953
    uint8_t *dst;
954

    
955
//    src[0]&0x80;                //forbidden bit
956
    h->nal_ref_idc= src[0]>>5;
957
    h->nal_unit_type= src[0]&0x1F;
958

    
959
    src++; length--;
960
#if 0    
961
    for(i=0; i<length; i++)
962
        printf("%2X ", src[i]);
963
#endif
964
    for(i=0; i+1<length; i+=2){
965
        if(src[i]) continue;
966
        if(i>0 && src[i-1]==0) i--;
967
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
968
            if(src[i+2]!=3){
969
                /* startcode, so we must be past the end */
970
                length=i;
971
            }
972
            break;
973
        }
974
    }
975

    
976
    if(i>=length-1){ //no escaped 0
977
        *dst_length= length;
978
        *consumed= length+1; //+1 for the header
979
        return src; 
980
    }
981

    
982
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
983
    dst= h->rbsp_buffer;
984

    
985
//printf("deoding esc\n");
986
    si=di=0;
987
    while(si<length){ 
988
        //remove escapes (very rare 1:2^22)
989
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
990
            if(src[si+2]==3){ //escape
991
                dst[di++]= 0;
992
                dst[di++]= 0;
993
                si+=3;
994
            }else //next start code
995
                break;
996
        }
997

    
998
        dst[di++]= src[si++];
999
    }
1000

    
1001
    *dst_length= di;
1002
    *consumed= si + 1;//+1 for the header
1003
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1004
    return dst;
1005
}
1006

    
1007
/**
1008
 * @param src the data which should be escaped
1009
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1010
 * @param length the length of the src data
1011
 * @param dst_length the length of the dst array
1012
 * @returns length of escaped data in bytes or -1 if an error occured
1013
 */
1014
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1015
    int i, escape_count, si, di;
1016
    uint8_t *temp;
1017
    
1018
    assert(length>=0);
1019
    assert(dst_length>0);
1020
    
1021
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1022

    
1023
    if(length==0) return 1;
1024

    
1025
    escape_count= 0;
1026
    for(i=0; i<length; i+=2){
1027
        if(src[i]) continue;
1028
        if(i>0 && src[i-1]==0) 
1029
            i--;
1030
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1031
            escape_count++;
1032
            i+=2;
1033
        }
1034
    }
1035
    
1036
    if(escape_count==0){ 
1037
        if(dst+1 != src)
1038
            memcpy(dst+1, src, length);
1039
        return length + 1;
1040
    }
1041
    
1042
    if(length + escape_count + 1> dst_length)
1043
        return -1;
1044

    
1045
    //this should be damn rare (hopefully)
1046

    
1047
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1048
    temp= h->rbsp_buffer;
1049
//printf("encoding esc\n");
1050
    
1051
    si= 0;
1052
    di= 0;
1053
    while(si < length){
1054
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1055
            temp[di++]= 0; si++;
1056
            temp[di++]= 0; si++;
1057
            temp[di++]= 3; 
1058
            temp[di++]= src[si++];
1059
        }
1060
        else
1061
            temp[di++]= src[si++];
1062
    }
1063
    memcpy(dst+1, temp, length+escape_count);
1064
    
1065
    assert(di == length+escape_count);
1066
    
1067
    return di + 1;
1068
}
1069

    
1070
/**
1071
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1072
 */
1073
static void encode_rbsp_trailing(PutBitContext *pb){
1074
    int length;
1075
    put_bits(pb, 1, 1);
1076
    length= (-get_bit_count(pb))&7;
1077
    if(length) put_bits(pb, length, 0);
1078
}
1079

    
1080
/**
1081
 * identifies the exact end of the bitstream
1082
 * @return the length of the trailing, or 0 if damaged
1083
 */
1084
static int decode_rbsp_trailing(uint8_t *src){
1085
    int v= *src;
1086
    int r;
1087

    
1088
#ifdef TRACE
1089
printf("rbsp trailing %X\n", v);
1090
#endif
1091

    
1092
    for(r=1; r<9; r++){
1093
        if(v&1) return r;
1094
        v>>=1;
1095
    }
1096
    return 0;
1097
}
1098

    
1099
/**
1100
 * idct tranforms the 16 dc values and dequantize them.
1101
 * @param qp quantization parameter
1102
 */
1103
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1104
    const int qmul= dequant_coeff[qp][0];
1105
#define stride 16
1106
    int i;
1107
    int temp[16]; //FIXME check if this is a good idea
1108
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1109
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1110

    
1111
//memset(block, 64, 2*256);
1112
//return;
1113
    for(i=0; i<4; i++){
1114
        const int offset= y_offset[i];
1115
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1116
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1117
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1118
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1119

    
1120
        temp[4*i+0]= z0+z3;
1121
        temp[4*i+1]= z1+z2;
1122
        temp[4*i+2]= z1-z2;
1123
        temp[4*i+3]= z0-z3;
1124
    }
1125

    
1126
    for(i=0; i<4; i++){
1127
        const int offset= x_offset[i];
1128
        const int z0= temp[4*0+i] + temp[4*2+i];
1129
        const int z1= temp[4*0+i] - temp[4*2+i];
1130
        const int z2= temp[4*1+i] - temp[4*3+i];
1131
        const int z3= temp[4*1+i] + temp[4*3+i];
1132

    
1133
        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1134
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1135
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1136
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1137
    }
1138
}
1139

    
1140
/**
1141
 * dct tranforms the 16 dc values.
1142
 * @param qp quantization parameter ??? FIXME
1143
 */
1144
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1145
//    const int qmul= dequant_coeff[qp][0];
1146
    int i;
1147
    int temp[16]; //FIXME check if this is a good idea
1148
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1149
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1150

    
1151
    for(i=0; i<4; i++){
1152
        const int offset= y_offset[i];
1153
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1154
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1155
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1156
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1157

    
1158
        temp[4*i+0]= z0+z3;
1159
        temp[4*i+1]= z1+z2;
1160
        temp[4*i+2]= z1-z2;
1161
        temp[4*i+3]= z0-z3;
1162
    }
1163

    
1164
    for(i=0; i<4; i++){
1165
        const int offset= x_offset[i];
1166
        const int z0= temp[4*0+i] + temp[4*2+i];
1167
        const int z1= temp[4*0+i] - temp[4*2+i];
1168
        const int z2= temp[4*1+i] - temp[4*3+i];
1169
        const int z3= temp[4*1+i] + temp[4*3+i];
1170

    
1171
        block[stride*0 +offset]= (z0 + z3)>>1;
1172
        block[stride*2 +offset]= (z1 + z2)>>1;
1173
        block[stride*8 +offset]= (z1 - z2)>>1;
1174
        block[stride*10+offset]= (z0 - z3)>>1;
1175
    }
1176
}
1177
#undef xStride
1178
#undef stride
1179

    
1180
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1181
    const int qmul= dequant_coeff[qp][0];
1182
    const int stride= 16*2;
1183
    const int xStride= 16;
1184
    int a,b,c,d,e;
1185

    
1186
    a= block[stride*0 + xStride*0];
1187
    b= block[stride*0 + xStride*1];
1188
    c= block[stride*1 + xStride*0];
1189
    d= block[stride*1 + xStride*1];
1190

    
1191
    e= a-b;
1192
    a= a+b;
1193
    b= c-d;
1194
    c= c+d;
1195

    
1196
    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1197
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1198
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1199
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1200
}
1201

    
1202
static void chroma_dc_dct_c(DCTELEM *block){
1203
    const int stride= 16*2;
1204
    const int xStride= 16;
1205
    int a,b,c,d,e;
1206

    
1207
    a= block[stride*0 + xStride*0];
1208
    b= block[stride*0 + xStride*1];
1209
    c= block[stride*1 + xStride*0];
1210
    d= block[stride*1 + xStride*1];
1211

    
1212
    e= a-b;
1213
    a= a+b;
1214
    b= c-d;
1215
    c= c+d;
1216

    
1217
    block[stride*0 + xStride*0]= (a+c);
1218
    block[stride*0 + xStride*1]= (e+b);
1219
    block[stride*1 + xStride*0]= (a-c);
1220
    block[stride*1 + xStride*1]= (e-b);
1221
}
1222

    
1223
/**
1224
 * gets the chroma qp.
1225
 */
1226
static inline int get_chroma_qp(H264Context *h, int qscale){
1227
    
1228
    return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1229
}
1230

    
1231

    
1232
/**
1233
 *
1234
 */
1235
static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1236
    int i;
1237
    uint8_t *cm = cropTbl + MAX_NEG_CROP;
1238

    
1239
    block[0] += 32;
1240
#if 1
1241
    for(i=0; i<4; i++){
1242
        const int z0=  block[i + 4*0]     +  block[i + 4*2];
1243
        const int z1=  block[i + 4*0]     -  block[i + 4*2];
1244
        const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
1245
        const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
1246

    
1247
        block[i + 4*0]= z0 + z3;
1248
        block[i + 4*1]= z1 + z2;
1249
        block[i + 4*2]= z1 - z2;
1250
        block[i + 4*3]= z0 - z3;
1251
    }
1252

    
1253
    for(i=0; i<4; i++){
1254
        const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
1255
        const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
1256
        const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
1257
        const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
1258

    
1259
        dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ];
1260
        dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ];
1261
        dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ];
1262
        dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ];
1263
    }
1264
#else
1265
    for(i=0; i<4; i++){
1266
        const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
1267
        const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
1268
        const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
1269
        const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
1270

    
1271
        block[0 + 4*i]= z0 + z3;
1272
        block[1 + 4*i]= z1 + z2;
1273
        block[2 + 4*i]= z1 - z2;
1274
        block[3 + 4*i]= z0 - z3;
1275
    }
1276

    
1277
    for(i=0; i<4; i++){
1278
        const int z0=  block[i + 4*0]     +  block[i + 4*2];
1279
        const int z1=  block[i + 4*0]     -  block[i + 4*2];
1280
        const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
1281
        const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
1282

    
1283
        dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1284
        dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1285
        dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1286
        dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1287
    }
1288
#endif
1289
}
1290

    
1291
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1292
    int i;
1293
    //FIXME try int temp instead of block
1294
    
1295
    for(i=0; i<4; i++){
1296
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1297
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1298
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1299
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1300
        const int z0= d0 + d3;
1301
        const int z3= d0 - d3;
1302
        const int z1= d1 + d2;
1303
        const int z2= d1 - d2;
1304
        
1305
        block[0 + 4*i]=   z0 +   z1;
1306
        block[1 + 4*i]= 2*z3 +   z2;
1307
        block[2 + 4*i]=   z0 -   z1;
1308
        block[3 + 4*i]=   z3 - 2*z2;
1309
    }    
1310

    
1311
    for(i=0; i<4; i++){
1312
        const int z0= block[0*4 + i] + block[3*4 + i];
1313
        const int z3= block[0*4 + i] - block[3*4 + i];
1314
        const int z1= block[1*4 + i] + block[2*4 + i];
1315
        const int z2= block[1*4 + i] - block[2*4 + i];
1316
        
1317
        block[0*4 + i]=   z0 +   z1;
1318
        block[1*4 + i]= 2*z3 +   z2;
1319
        block[2*4 + i]=   z0 -   z1;
1320
        block[3*4 + i]=   z3 - 2*z2;
1321
    }
1322
}
1323

    
1324
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1325
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1326
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1327
    int i;
1328
    const int * const quant_table= quant_coeff[qscale];
1329
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1330
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1331
    const unsigned int threshold2= (threshold1<<1);
1332
    int last_non_zero;
1333

    
1334
    if(seperate_dc){
1335
        if(qscale<=18){
1336
            //avoid overflows
1337
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1338
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1339
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1340

    
1341
            int level= block[0]*quant_coeff[qscale+18][0];
1342
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1343
                if(level>0){
1344
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1345
                    block[0]= level;
1346
                }else{
1347
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1348
                    block[0]= -level;
1349
                }
1350
//                last_non_zero = i;
1351
            }else{
1352
                block[0]=0;
1353
            }
1354
        }else{
1355
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1356
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1357
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1358

    
1359
            int level= block[0]*quant_table[0];
1360
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1361
                if(level>0){
1362
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1363
                    block[0]= level;
1364
                }else{
1365
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1366
                    block[0]= -level;
1367
                }
1368
//                last_non_zero = i;
1369
            }else{
1370
                block[0]=0;
1371
            }
1372
        }
1373
        last_non_zero= 0;
1374
        i=1;
1375
    }else{
1376
        last_non_zero= -1;
1377
        i=0;
1378
    }
1379

    
1380
    for(; i<16; i++){
1381
        const int j= scantable[i];
1382
        int level= block[j]*quant_table[j];
1383

    
1384
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1385
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1386
        if(((unsigned)(level+threshold1))>threshold2){
1387
            if(level>0){
1388
                level= (bias + level)>>QUANT_SHIFT;
1389
                block[j]= level;
1390
            }else{
1391
                level= (bias - level)>>QUANT_SHIFT;
1392
                block[j]= -level;
1393
            }
1394
            last_non_zero = i;
1395
        }else{
1396
            block[j]=0;
1397
        }
1398
    }
1399

    
1400
    return last_non_zero;
1401
}
1402

    
1403
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1404
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1405
    ((uint32_t*)(src+0*stride))[0]= a;
1406
    ((uint32_t*)(src+1*stride))[0]= a;
1407
    ((uint32_t*)(src+2*stride))[0]= a;
1408
    ((uint32_t*)(src+3*stride))[0]= a;
1409
}
1410

    
1411
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1412
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1413
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1414
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1415
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1416
}
1417

    
1418
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1419
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1420
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
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_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1429
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1430
    
1431
    ((uint32_t*)(src+0*stride))[0]= 
1432
    ((uint32_t*)(src+1*stride))[0]= 
1433
    ((uint32_t*)(src+2*stride))[0]= 
1434
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1435
}
1436

    
1437
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1438
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1439
    
1440
    ((uint32_t*)(src+0*stride))[0]= 
1441
    ((uint32_t*)(src+1*stride))[0]= 
1442
    ((uint32_t*)(src+2*stride))[0]= 
1443
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1444
}
1445

    
1446
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1447
    ((uint32_t*)(src+0*stride))[0]= 
1448
    ((uint32_t*)(src+1*stride))[0]= 
1449
    ((uint32_t*)(src+2*stride))[0]= 
1450
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1451
}
1452

    
1453

    
1454
#define LOAD_TOP_RIGHT_EDGE\
1455
    const int t4= topright[0];\
1456
    const int t5= topright[1];\
1457
    const int t6= topright[2];\
1458
    const int t7= topright[3];\
1459

    
1460
#define LOAD_LEFT_EDGE\
1461
    const int l0= src[-1+0*stride];\
1462
    const int l1= src[-1+1*stride];\
1463
    const int l2= src[-1+2*stride];\
1464
    const int l3= src[-1+3*stride];\
1465

    
1466
#define LOAD_TOP_EDGE\
1467
    const int t0= src[ 0-1*stride];\
1468
    const int t1= src[ 1-1*stride];\
1469
    const int t2= src[ 2-1*stride];\
1470
    const int t3= src[ 3-1*stride];\
1471

    
1472
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1473
    const int lt= src[-1-1*stride];
1474
    LOAD_TOP_EDGE
1475
    LOAD_LEFT_EDGE
1476

    
1477
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1478
    src[0+2*stride]=
1479
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1480
    src[0+1*stride]=
1481
    src[1+2*stride]=
1482
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1483
    src[0+0*stride]=
1484
    src[1+1*stride]=
1485
    src[2+2*stride]=
1486
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1487
    src[1+0*stride]=
1488
    src[2+1*stride]=
1489
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1490
    src[2+0*stride]=
1491
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1492
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1493
};
1494

    
1495
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1496
    LOAD_TOP_EDGE    
1497
    LOAD_TOP_RIGHT_EDGE    
1498
//    LOAD_LEFT_EDGE    
1499

    
1500
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1501
    src[1+0*stride]=
1502
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1503
    src[2+0*stride]=
1504
    src[1+1*stride]=
1505
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1506
    src[3+0*stride]=
1507
    src[2+1*stride]=
1508
    src[1+2*stride]=
1509
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1510
    src[3+1*stride]=
1511
    src[2+2*stride]=
1512
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1513
    src[3+2*stride]=
1514
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1515
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1516
};
1517

    
1518
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1519
    const int lt= src[-1-1*stride];
1520
    LOAD_TOP_EDGE    
1521
    LOAD_LEFT_EDGE    
1522
    const __attribute__((unused)) int unu= l3;
1523

    
1524
    src[0+0*stride]=
1525
    src[1+2*stride]=(lt + t0 + 1)>>1;
1526
    src[1+0*stride]=
1527
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1528
    src[2+0*stride]=
1529
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1530
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1531
    src[0+1*stride]=
1532
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1533
    src[1+1*stride]=
1534
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1535
    src[2+1*stride]=
1536
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1537
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1538
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1539
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1540
};
1541

    
1542
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1543
    LOAD_TOP_EDGE    
1544
    LOAD_TOP_RIGHT_EDGE    
1545
    const __attribute__((unused)) int unu= t7;
1546

    
1547
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1548
    src[1+0*stride]=
1549
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1550
    src[2+0*stride]=
1551
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1552
    src[3+0*stride]=
1553
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1554
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1555
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1556
    src[1+1*stride]=
1557
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1558
    src[2+1*stride]=
1559
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1560
    src[3+1*stride]=
1561
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1562
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1563
};
1564

    
1565
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1566
    LOAD_LEFT_EDGE    
1567

    
1568
    src[0+0*stride]=(l0 + l1 + 1)>>1;
1569
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1570
    src[2+0*stride]=
1571
    src[0+1*stride]=(l1 + l2 + 1)>>1;
1572
    src[3+0*stride]=
1573
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1574
    src[2+1*stride]=
1575
    src[0+2*stride]=(l2 + l3 + 1)>>1;
1576
    src[3+1*stride]=
1577
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1578
    src[3+2*stride]=
1579
    src[1+3*stride]=
1580
    src[0+3*stride]=
1581
    src[2+2*stride]=
1582
    src[2+3*stride]=
1583
    src[3+3*stride]=l3;
1584
};
1585
    
1586
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1587
    const int lt= src[-1-1*stride];
1588
    LOAD_TOP_EDGE    
1589
    LOAD_LEFT_EDGE    
1590
    const __attribute__((unused)) int unu= t3;
1591

    
1592
    src[0+0*stride]=
1593
    src[2+1*stride]=(lt + l0 + 1)>>1;
1594
    src[1+0*stride]=
1595
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1596
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1597
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1598
    src[0+1*stride]=
1599
    src[2+2*stride]=(l0 + l1 + 1)>>1;
1600
    src[1+1*stride]=
1601
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1602
    src[0+2*stride]=
1603
    src[2+3*stride]=(l1 + l2+ 1)>>1;
1604
    src[1+2*stride]=
1605
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1606
    src[0+3*stride]=(l2 + l3 + 1)>>1;
1607
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1608
};
1609

    
1610
static void pred16x16_vertical_c(uint8_t *src, int stride){
1611
    int i;
1612
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1613
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1614
    const uint32_t c= ((uint32_t*)(src-stride))[2];
1615
    const uint32_t d= ((uint32_t*)(src-stride))[3];
1616
    
1617
    for(i=0; i<16; i++){
1618
        ((uint32_t*)(src+i*stride))[0]= a;
1619
        ((uint32_t*)(src+i*stride))[1]= b;
1620
        ((uint32_t*)(src+i*stride))[2]= c;
1621
        ((uint32_t*)(src+i*stride))[3]= d;
1622
    }
1623
}
1624

    
1625
static void pred16x16_horizontal_c(uint8_t *src, int stride){
1626
    int i;
1627

    
1628
    for(i=0; i<16; i++){
1629
        ((uint32_t*)(src+i*stride))[0]=
1630
        ((uint32_t*)(src+i*stride))[1]=
1631
        ((uint32_t*)(src+i*stride))[2]=
1632
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1633
    }
1634
}
1635

    
1636
static void pred16x16_dc_c(uint8_t *src, int stride){
1637
    int i, dc=0;
1638

    
1639
    for(i=0;i<16; i++){
1640
        dc+= src[-1+i*stride];
1641
    }
1642
    
1643
    for(i=0;i<16; i++){
1644
        dc+= src[i-stride];
1645
    }
1646

    
1647
    dc= 0x01010101*((dc + 16)>>5);
1648

    
1649
    for(i=0; i<16; i++){
1650
        ((uint32_t*)(src+i*stride))[0]=
1651
        ((uint32_t*)(src+i*stride))[1]=
1652
        ((uint32_t*)(src+i*stride))[2]=
1653
        ((uint32_t*)(src+i*stride))[3]= dc;
1654
    }
1655
}
1656

    
1657
static void pred16x16_left_dc_c(uint8_t *src, int stride){
1658
    int i, dc=0;
1659

    
1660
    for(i=0;i<16; i++){
1661
        dc+= src[-1+i*stride];
1662
    }
1663
    
1664
    dc= 0x01010101*((dc + 8)>>4);
1665

    
1666
    for(i=0; i<16; i++){
1667
        ((uint32_t*)(src+i*stride))[0]=
1668
        ((uint32_t*)(src+i*stride))[1]=
1669
        ((uint32_t*)(src+i*stride))[2]=
1670
        ((uint32_t*)(src+i*stride))[3]= dc;
1671
    }
1672
}
1673

    
1674
static void pred16x16_top_dc_c(uint8_t *src, int stride){
1675
    int i, dc=0;
1676

    
1677
    for(i=0;i<16; i++){
1678
        dc+= src[i-stride];
1679
    }
1680
    dc= 0x01010101*((dc + 8)>>4);
1681

    
1682
    for(i=0; i<16; i++){
1683
        ((uint32_t*)(src+i*stride))[0]=
1684
        ((uint32_t*)(src+i*stride))[1]=
1685
        ((uint32_t*)(src+i*stride))[2]=
1686
        ((uint32_t*)(src+i*stride))[3]= dc;
1687
    }
1688
}
1689

    
1690
static void pred16x16_128_dc_c(uint8_t *src, int stride){
1691
    int i;
1692

    
1693
    for(i=0; i<16; i++){
1694
        ((uint32_t*)(src+i*stride))[0]=
1695
        ((uint32_t*)(src+i*stride))[1]=
1696
        ((uint32_t*)(src+i*stride))[2]=
1697
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1698
    }
1699
}
1700

    
1701
static void pred16x16_plane_c(uint8_t *src, int stride){
1702
    uint8_t *cm = cropTbl + MAX_NEG_CROP;
1703
    int i, dx, dy, dc;
1704
    int temp[16];
1705
    
1706
    dc= 16*(src[15-stride] + src[-1+15*stride]);
1707
    
1708
    dx=dy=0;
1709
    for(i=1; i<9; i++){
1710
        dx += i*(src[7+i-stride] - src[7-i-stride]);
1711
        dy += i*(src[-1+(7+i)*stride] - src[-1+(7-i)*stride]);
1712
    }
1713
    dx= (5*dx+32)>>6;
1714
    dy= (5*dy+32)>>6;
1715
    
1716
    dc += 16;
1717

    
1718
    //FIXME modifiy dc,dx,dy to avoid -7
1719
    
1720
    for(i=0; i<16; i++)
1721
        temp[i]= dx*(i-7) + dc;
1722
    
1723
    if(   (dc - ABS(dx)*8 - ABS(dy)*8)>>5 < 0
1724
       || (dc + ABS(dx)*8 + ABS(dy)*8)>>5 > 255){
1725
    
1726
        for(i=0; i<16; i++){
1727
            int j;
1728
            for(j=0; j<16; j++)
1729
                src[j + i*stride]= cm[ (temp[j] + dy*(i-7))>>5 ];
1730
        }
1731
    }else{
1732
        for(i=0; i<16; i++){
1733
            int j;
1734
            for(j=0; j<16; j++)
1735
                src[j + i*stride]= (temp[j] + dy*(i-7))>>5;
1736
        }
1737
    }
1738
}
1739

    
1740
static void pred8x8_vertical_c(uint8_t *src, int stride){
1741
    int i;
1742
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1743
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1744
    
1745
    for(i=0; i<8; i++){
1746
        ((uint32_t*)(src+i*stride))[0]= a;
1747
        ((uint32_t*)(src+i*stride))[1]= b;
1748
    }
1749
}
1750

    
1751
static void pred8x8_horizontal_c(uint8_t *src, int stride){
1752
    int i;
1753

    
1754
    for(i=0; i<8; i++){
1755
        ((uint32_t*)(src+i*stride))[0]=
1756
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1757
    }
1758
}
1759

    
1760
static void pred8x8_128_dc_c(uint8_t *src, int stride){
1761
    int i;
1762

    
1763
    for(i=0; i<4; i++){
1764
        ((uint32_t*)(src+i*stride))[0]= 
1765
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1766
    }
1767
    for(i=4; i<8; i++){
1768
        ((uint32_t*)(src+i*stride))[0]= 
1769
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1770
    }
1771
}
1772

    
1773
static void pred8x8_left_dc_c(uint8_t *src, int stride){
1774
    int i;
1775
    int dc0, dc2;
1776

    
1777
    dc0=dc2=0;
1778
    for(i=0;i<4; i++){
1779
        dc0+= src[-1+i*stride];
1780
        dc2+= src[-1+(i+4)*stride];
1781
    }
1782
    dc0= 0x01010101*((dc0 + 2)>>2);
1783
    dc2= 0x01010101*((dc2 + 2)>>2);
1784

    
1785
    for(i=0; i<4; i++){
1786
        ((uint32_t*)(src+i*stride))[0]=
1787
        ((uint32_t*)(src+i*stride))[1]= dc0;
1788
    }
1789
    for(i=4; i<8; i++){
1790
        ((uint32_t*)(src+i*stride))[0]=
1791
        ((uint32_t*)(src+i*stride))[1]= dc2;
1792
    }
1793
}
1794

    
1795
static void pred8x8_top_dc_c(uint8_t *src, int stride){
1796
    int i;
1797
    int dc0, dc1;
1798

    
1799
    dc0=dc1=0;
1800
    for(i=0;i<4; i++){
1801
        dc0+= src[i-stride];
1802
        dc1+= src[4+i-stride];
1803
    }
1804
    dc0= 0x01010101*((dc0 + 2)>>2);
1805
    dc1= 0x01010101*((dc1 + 2)>>2);
1806

    
1807
    for(i=0; i<4; i++){
1808
        ((uint32_t*)(src+i*stride))[0]= dc0;
1809
        ((uint32_t*)(src+i*stride))[1]= dc1;
1810
    }
1811
    for(i=4; i<8; i++){
1812
        ((uint32_t*)(src+i*stride))[0]= dc0;
1813
        ((uint32_t*)(src+i*stride))[1]= dc1;
1814
    }
1815
}
1816

    
1817

    
1818
static void pred8x8_dc_c(uint8_t *src, int stride){
1819
    int i;
1820
    int dc0, dc1, dc2, dc3;
1821

    
1822
    dc0=dc1=dc2=0;
1823
    for(i=0;i<4; i++){
1824
        dc0+= src[-1+i*stride] + src[i-stride];
1825
        dc1+= src[4+i-stride];
1826
        dc2+= src[-1+(i+4)*stride];
1827
    }
1828
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1829
    dc0= 0x01010101*((dc0 + 4)>>3);
1830
    dc1= 0x01010101*((dc1 + 2)>>2);
1831
    dc2= 0x01010101*((dc2 + 2)>>2);
1832

    
1833
    for(i=0; i<4; i++){
1834
        ((uint32_t*)(src+i*stride))[0]= dc0;
1835
        ((uint32_t*)(src+i*stride))[1]= dc1;
1836
    }
1837
    for(i=4; i<8; i++){
1838
        ((uint32_t*)(src+i*stride))[0]= dc2;
1839
        ((uint32_t*)(src+i*stride))[1]= dc3;
1840
    }
1841
}
1842

    
1843
static void pred8x8_plane_c(uint8_t *src, int stride){
1844
    uint8_t *cm = cropTbl + MAX_NEG_CROP;
1845
    int i, dx, dy, dc;
1846
    int temp[8];
1847

    
1848
    dc= 16*(src[7-stride] + src[-1+7*stride]);
1849
    
1850
    dx=dy=0;
1851
    for(i=1; i<5; i++){
1852
        dx += i*(src[3+i-stride] - src[3-i-stride]);
1853
        dy += i*(src[-1+(3+i)*stride] - src[-1+(3-i)*stride]);
1854
    }
1855
    dx= (17*dx+16)>>5;
1856
    dy= (17*dy+16)>>5;
1857
    
1858
    dc += 16;
1859
    
1860
    //FIXME modifiy dc,dx,dy to avoid -3
1861
    
1862
    for(i=0; i<8; i++)
1863
        temp[i]= dx*(i-3) + dc;
1864
    
1865
    if(   (dc - ABS(dx)*4 - ABS(dy)*4)>>5 < 0
1866
       || (dc + ABS(dx)*4 + ABS(dy)*4)>>5 > 255){
1867
    
1868
        for(i=0; i<8; i++){
1869
            int j;
1870
            for(j=0; j<8; j++)
1871
                src[j + i*stride]= cm[ (temp[j] + dy*(i-3))>>5 ];
1872
        }
1873
    }else{
1874
        for(i=0; i<8; i++){
1875
            int j;
1876
            for(j=0; j<8; j++)
1877
                src[j + i*stride]= (temp[j] + dy*(i-3))>>5;
1878
        }
1879
    }
1880
}
1881

    
1882
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1883
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1884
                           int src_x_offset, int src_y_offset,
1885
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1886
    MpegEncContext * const s = &h->s;
1887
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1888
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1889
    const int luma_xy= (mx&3) + ((my&3)<<2);
1890
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1891
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1892
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1893
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1894
    int extra_height= extra_width;
1895
    int emu=0;
1896
    const int full_mx= mx>>2;
1897
    const int full_my= my>>2;
1898
    
1899
    assert(pic->data[0]);
1900
    
1901
    if(mx&7) extra_width -= 3;
1902
    if(my&7) extra_height -= 3;
1903
    
1904
    if(   full_mx < 0-extra_width 
1905
       || full_my < 0-extra_height 
1906
       || full_mx + 16/*FIXME*/ > s->width + extra_width 
1907
       || full_my + 16/*FIXME*/ > s->height + extra_height){
1908
        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);
1909
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1910
        emu=1;
1911
    }
1912
    
1913
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1914
    if(!square){
1915
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1916
    }
1917
    
1918
    if(s->flags&CODEC_FLAG_GRAY) return;
1919
    
1920
    if(emu){
1921
        ff_emulated_edge_mc(s, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
1922
            src_cb= s->edge_emu_buffer;
1923
    }
1924
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1925

    
1926
    if(emu){
1927
        ff_emulated_edge_mc(s, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
1928
            src_cr= s->edge_emu_buffer;
1929
    }
1930
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1931
}
1932

    
1933
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1934
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1935
                           int x_offset, int y_offset,
1936
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1937
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1938
                           int list0, int list1){
1939
    MpegEncContext * const s = &h->s;
1940
    qpel_mc_func *qpix_op=  qpix_put;
1941
    h264_chroma_mc_func chroma_op= chroma_put;
1942
    
1943
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
1944
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
1945
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
1946
    x_offset += 8*s->mb_x;
1947
    y_offset += 8*s->mb_y;
1948
    
1949
    if(list0){
1950
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[0] ] ];
1951
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1952
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1953
                           qpix_op, chroma_op);
1954

    
1955
        qpix_op=  qpix_avg;
1956
        chroma_op= chroma_avg;
1957
    }
1958

    
1959
    if(list1){
1960
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[0] ] ];
1961
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1962
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1963
                           qpix_op, chroma_op);
1964
    }
1965
}
1966

    
1967
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1968
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1969
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
1970
    MpegEncContext * const s = &h->s;
1971
    const int mb_xy= s->mb_x + s->mb_y*h->mb_stride;
1972
    const int mb_type= s->current_picture.mb_type[mb_xy];
1973
    
1974
    assert(IS_INTER(mb_type));
1975
    
1976
    if(IS_16X16(mb_type)){
1977
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1978
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1979
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1980
    }else if(IS_16X8(mb_type)){
1981
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1982
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1983
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1984
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1985
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1986
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1987
    }else if(IS_8X16(mb_type)){
1988
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
1989
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1990
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1991
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
1992
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1993
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1994
    }else{
1995
        int i;
1996
        
1997
        assert(IS_8X8(mb_type));
1998

    
1999
        for(i=0; i<4; i++){
2000
            const int sub_mb_type= h->sub_mb_type[i];
2001
            const int n= 4*i;
2002
            int x_offset= (i&1)<<2;
2003
            int y_offset= (i&2)<<1;
2004

    
2005
            if(IS_SUB_8X8(sub_mb_type)){
2006
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2007
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2008
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2009
            }else if(IS_SUB_8X4(sub_mb_type)){
2010
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2011
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2012
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2013
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2014
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2015
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2016
            }else if(IS_SUB_4X8(sub_mb_type)){
2017
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2018
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2019
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2020
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2021
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2022
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2023
            }else{
2024
                int j;
2025
                assert(IS_SUB_4X4(sub_mb_type));
2026
                for(j=0; j<4; j++){
2027
                    int sub_x_offset= x_offset + 2*(j&1);
2028
                    int sub_y_offset= y_offset +   (j&2);
2029
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2030
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2031
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2032
                }
2033
            }
2034
        }
2035
    }
2036
}
2037

    
2038
static void decode_init_vlc(H264Context *h){
2039
    static int done = 0;
2040

    
2041
    if (!done) {
2042
        int i;
2043
        done = 1;
2044

    
2045
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2046
                 &chroma_dc_coeff_token_len [0], 1, 1,
2047
                 &chroma_dc_coeff_token_bits[0], 1, 1);
2048

    
2049
        for(i=0; i<4; i++){
2050
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2051
                     &coeff_token_len [i][0], 1, 1,
2052
                     &coeff_token_bits[i][0], 1, 1);
2053
        }
2054

    
2055
        for(i=0; i<3; i++){
2056
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2057
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2058
                     &chroma_dc_total_zeros_bits[i][0], 1, 1);
2059
        }
2060
        for(i=0; i<15; i++){
2061
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2062
                     &total_zeros_len [i][0], 1, 1,
2063
                     &total_zeros_bits[i][0], 1, 1);
2064
        }
2065

    
2066
        for(i=0; i<6; i++){
2067
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2068
                     &run_len [i][0], 1, 1,
2069
                     &run_bits[i][0], 1, 1);
2070
        }
2071
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2072
                 &run_len [6][0], 1, 1,
2073
                 &run_bits[6][0], 1, 1);
2074
    }
2075
}
2076

    
2077
/**
2078
 * Sets the intra prediction function pointers.
2079
 */
2080
static void init_pred_ptrs(H264Context *h){
2081
//    MpegEncContext * const s = &h->s;
2082

    
2083
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2084
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2085
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2086
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2087
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2088
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2089
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2090
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2091
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2092
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2093
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2094
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2095

    
2096
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2097
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2098
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2099
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2100
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2101
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2102
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2103

    
2104
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2105
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2106
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2107
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2108
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2109
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2110
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2111
}
2112

    
2113
//FIXME factorize
2114
#define CHECKED_ALLOCZ(p, size)\
2115
{\
2116
    p= av_mallocz(size);\
2117
    if(p==NULL){\
2118
        perror("malloc");\
2119
        goto fail;\
2120
    }\
2121
}
2122

    
2123
static void free_tables(H264Context *h){
2124
    MpegEncContext * const s = &h->s;
2125

    
2126
    av_freep(&h->intra4x4_pred_mode);
2127
    av_freep(&h->non_zero_count);
2128
    av_freep(&h->slice_table_base);
2129
    h->slice_table= NULL;
2130
    
2131
    av_freep(&h->mb2b_xy);
2132
    av_freep(&h->mb2b8_xy);
2133
}
2134

    
2135
/**
2136
 * allocates tables.
2137
 * needs widzh/height
2138
 */
2139
static int alloc_tables(H264Context *h){
2140
    MpegEncContext * const s = &h->s;
2141
    const int big_mb_num= h->mb_stride * (s->mb_height+1);
2142
    int x,y;
2143

    
2144
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2145
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2146
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2147

    
2148
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2149
    h->slice_table= h->slice_table_base + h->mb_stride + 1;
2150

    
2151
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint16_t));
2152
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2153
    for(y=0; y<s->mb_height; y++){
2154
        for(x=0; x<s->mb_width; x++){
2155
            const int mb_xy= x + y*h->mb_stride;
2156
            const int b_xy = 4*x + 4*y*h->b_stride;
2157
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2158
        
2159
            h->mb2b_xy [mb_xy]= b_xy;
2160
            h->mb2b8_xy[mb_xy]= b8_xy;
2161
        }
2162
    }
2163
    
2164
    return 0;
2165
fail:
2166
    free_tables(h);
2167
    return -1;
2168
}
2169

    
2170
static void common_init(H264Context *h){
2171
    MpegEncContext * const s = &h->s;
2172
    int i;
2173

    
2174
    s->width = s->avctx->width;
2175
    s->height = s->avctx->height;
2176
    s->codec_id= s->avctx->codec->id;
2177
    
2178
    init_pred_ptrs(h);
2179

    
2180
    s->decode=1; //FIXME
2181
}
2182

    
2183
static int decode_init(AVCodecContext *avctx){
2184
    H264Context *h= avctx->priv_data;
2185
    MpegEncContext * const s = &h->s;
2186

    
2187
    s->avctx = avctx;
2188
    common_init(h);
2189

    
2190
    s->out_format = FMT_H264;
2191
    s->workaround_bugs= avctx->workaround_bugs;
2192

    
2193
    // set defaults
2194
    s->progressive_sequence=1;
2195
//    s->decode_mb= ff_h263_decode_mb;
2196
    s->low_delay= 1;
2197
    avctx->pix_fmt= PIX_FMT_YUV420P;
2198

    
2199
    decode_init_vlc(h);
2200
    
2201
    return 0;
2202
}
2203

    
2204
static void frame_start(H264Context *h){
2205
    MpegEncContext * const s = &h->s;
2206
    int i;
2207

    
2208
    MPV_frame_start(s, s->avctx);
2209
    ff_er_frame_start(s);
2210
    h->mmco_index=0;
2211

    
2212
    assert(s->linesize && s->uvlinesize);
2213

    
2214
    for(i=0; i<16; i++){
2215
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2216
        h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2217
    }
2218
    for(i=0; i<4; i++){
2219
        h->block_offset[16+i]=
2220
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2221
    }
2222

    
2223
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2224
}
2225

    
2226
static void hl_decode_mb(H264Context *h){
2227
    MpegEncContext * const s = &h->s;
2228
    const int mb_x= s->mb_x;
2229
    const int mb_y= s->mb_y;
2230
    const int mb_xy= mb_x + mb_y*h->mb_stride;
2231
    const int mb_type= s->current_picture.mb_type[mb_xy];
2232
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2233
    int linesize, uvlinesize /*dct_offset*/;
2234
    int i;
2235

    
2236
    if(!s->decode)
2237
        return;
2238

    
2239
    if(s->mb_skiped){
2240
    }
2241

    
2242
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2243
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2244
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2245

    
2246
    if (h->mb_field_decoding_flag) {
2247
        linesize = s->linesize * 2;
2248
        uvlinesize = s->uvlinesize * 2;
2249
        if(mb_y&1){ //FIXME move out of this func?
2250
            dest_y -= s->linesize*15;
2251
            dest_cb-= s->linesize*7;
2252
            dest_cr-= s->linesize*7;
2253
        }
2254
    } else {
2255
        linesize = s->linesize;
2256
        uvlinesize = s->uvlinesize;
2257
//        dct_offset = s->linesize * 16;
2258
    }
2259

    
2260
    if(IS_INTRA(mb_type)){
2261
        if(!(s->flags&CODEC_FLAG_GRAY)){
2262
            h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2263
            h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2264
        }
2265

    
2266
        if(IS_INTRA4x4(mb_type)){
2267
            if(!s->encoding){
2268
                for(i=0; i<16; i++){
2269
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2270
                    uint8_t *topright= ptr + 4 - linesize;
2271
                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2272
                    const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2273
                    int tr;
2274

    
2275
                    if(!topright_avail){
2276
                        tr= ptr[3 - linesize]*0x01010101;
2277
                        topright= (uint8_t*) &tr;
2278
                    }
2279

    
2280
                    h->pred4x4[ dir ](ptr, topright, linesize);
2281
                    if(h->non_zero_count_cache[ scan8[i] ])
2282
                        h264_add_idct_c(ptr, h->mb + i*16, linesize);
2283
                }
2284
            }
2285
        }else{
2286
            h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2287
            h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2288
        }
2289
    }else{
2290
        hl_motion(h, dest_y, dest_cb, dest_cr,
2291
                  s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
2292
                  s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2293
    }
2294

    
2295

    
2296
    if(!IS_INTRA4x4(mb_type)){
2297
        for(i=0; i<16; i++){
2298
            if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2299
                uint8_t * const ptr= dest_y + h->block_offset[i];
2300
                h264_add_idct_c(ptr, h->mb + i*16, linesize);
2301
            }
2302
        }
2303
    }
2304

    
2305
    if(!(s->flags&CODEC_FLAG_GRAY)){
2306
        chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2307
        chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2308
        for(i=16; i<16+4; i++){
2309
            if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2310
                uint8_t * const ptr= dest_cb + h->block_offset[i];
2311
                h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2312
            }
2313
        }
2314
        for(i=20; i<20+4; i++){
2315
            if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2316
                uint8_t * const ptr= dest_cr + h->block_offset[i];
2317
                h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2318
            }
2319
        }
2320
    }
2321
}
2322

    
2323
static void decode_mb_cabac(H264Context *h){
2324
//    MpegEncContext * const s = &h->s;
2325
}
2326

    
2327
/**
2328
 * fills the default_ref_list.
2329
 */
2330
static int fill_default_ref_list(H264Context *h){
2331
    MpegEncContext * const s = &h->s;
2332
    int i;
2333
    Picture sorted_short_ref[16];
2334
    
2335
    if(h->slice_type==B_TYPE){
2336
        int out_i;
2337
        int limit= -1;
2338

    
2339
        for(out_i=0; out_i<h->short_ref_count; out_i++){
2340
            int best_i=-1;
2341
            int best_poc=-1;
2342

    
2343
            for(i=0; i<h->short_ref_count; i++){
2344
                const int poc= h->short_ref[i]->poc;
2345
                if(poc > limit && poc < best_poc){
2346
                    best_poc= poc;
2347
                    best_i= i;
2348
                }
2349
            }
2350
            
2351
            assert(best_i != -1);
2352
            
2353
            limit= best_poc;
2354
            sorted_short_ref[out_i]= *h->short_ref[best_i];
2355
        }
2356
    }
2357

    
2358
    if(s->picture_structure == PICT_FRAME){
2359
        if(h->slice_type==B_TYPE){
2360
            const int current_poc= s->current_picture_ptr->poc;
2361
            int list;
2362

    
2363
            for(list=0; list<2; list++){
2364
                int index=0;
2365

    
2366
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2367
                    const int i2= list ? h->short_ref_count - i - 1 : i;
2368
                    const int poc= sorted_short_ref[i2].poc;
2369
                    
2370
                    if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2371

    
2372
                    if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2373
                        h->default_ref_list[list][index  ]= sorted_short_ref[i2];
2374
                        h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2375
                    }
2376
                }
2377

    
2378
                for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2379
                    if(h->long_ref[i]->reference != 3) continue;
2380

    
2381
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
2382
                    h->default_ref_list[ list ][index++].pic_id= i;;
2383
                }
2384
                
2385
                if(h->long_ref_count > 1 && h->short_ref_count==0){
2386
                    Picture temp= h->default_ref_list[1][0];
2387
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
2388
                    h->default_ref_list[1][0] = temp;
2389
                }
2390

    
2391
                if(index < h->ref_count[ list ])
2392
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2393
            }
2394
        }else{
2395
            int index=0;
2396
            for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2397
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2398
                h->default_ref_list[0][index  ]= *h->short_ref[i];
2399
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2400
            }
2401
            for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2402
                if(h->long_ref[i]->reference != 3) continue;
2403
                h->default_ref_list[0][index  ]= *h->long_ref[i];
2404
                h->default_ref_list[0][index++].pic_id= i;;
2405
            }
2406
            if(index < h->ref_count[0])
2407
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2408
        }
2409
    }else{ //FIELD
2410
        if(h->slice_type==B_TYPE){
2411
        }else{
2412
            //FIXME second field balh
2413
        }
2414
    }
2415
    return 0;
2416
}
2417

    
2418
static int decode_ref_pic_list_reordering(H264Context *h){
2419
    MpegEncContext * const s = &h->s;
2420
    int list;
2421
    
2422
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2423
    
2424
    for(list=0; list<2; list++){
2425
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2426

    
2427
        if(get_bits1(&s->gb)){
2428
            int pred= h->curr_pic_num;
2429
            int index;
2430

    
2431
            for(index=0; ; index++){
2432
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2433
                int pic_id;
2434
                int i;
2435
                
2436
                
2437
                if(index >= h->ref_count[list]){
2438
                    fprintf(stderr, "reference count overflow\n");
2439
                    return -1;
2440
                }
2441
                
2442
                if(reordering_of_pic_nums_idc<3){
2443
                    if(reordering_of_pic_nums_idc<2){
2444
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2445

    
2446
                        if(abs_diff_pic_num >= h->max_pic_num){
2447
                            fprintf(stderr, "abs_diff_pic_num overflow\n");
2448
                            return -1;
2449
                        }
2450

    
2451
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2452
                        else                                pred+= abs_diff_pic_num;
2453
                        pred &= h->max_pic_num - 1;
2454
                    
2455
                        for(i= h->ref_count[list]-1; i>=index; i--){
2456
                            if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2457
                                break;
2458
                        }
2459
                    }else{
2460
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2461

    
2462
                        for(i= h->ref_count[list]-1; i>=index; i--){
2463
                            if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2464
                                break;
2465
                        }
2466
                    }
2467

    
2468
                    if(i < index){
2469
                        fprintf(stderr, "reference picture missing during reorder\n");
2470
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2471
                    }else if(i > index){
2472
                        Picture tmp= h->ref_list[list][i];
2473
                        for(; i>index; i--){
2474
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2475
                        }
2476
                        h->ref_list[list][index]= tmp;
2477
                    }
2478
                }else if(reordering_of_pic_nums_idc==3) 
2479
                    break;
2480
                else{
2481
                    fprintf(stderr, "illegal reordering_of_pic_nums_idc\n");
2482
                    return -1;
2483
                }
2484
            }
2485
        }
2486

    
2487
        if(h->slice_type!=B_TYPE) break;
2488
    }
2489
    return 0;    
2490
}
2491

    
2492
static int pred_weight_table(H264Context *h){
2493
    MpegEncContext * const s = &h->s;
2494
    int list, i;
2495
    
2496
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2497
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2498

    
2499
    for(list=0; list<2; list++){
2500
        for(i=0; i<h->ref_count[list]; i++){
2501
            int luma_weight_flag, chroma_weight_flag;
2502
            
2503
            luma_weight_flag= get_bits1(&s->gb);
2504
            if(luma_weight_flag){
2505
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
2506
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
2507
            }
2508

    
2509
            chroma_weight_flag= get_bits1(&s->gb);
2510
            if(chroma_weight_flag){
2511
                int j;
2512
                for(j=0; j<2; j++){
2513
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2514
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2515
                }
2516
            }
2517
        }
2518
        if(h->slice_type != B_TYPE) break;
2519
    }
2520
    return 0;
2521
}
2522

    
2523
/**
2524
 * instantaneos decoder refresh.
2525
 */
2526
static void idr(H264Context *h){
2527
    int i;
2528

    
2529
    for(i=0; i<h->long_ref_count; i++){
2530
        h->long_ref[i]->reference=0;
2531
        h->long_ref[i]= NULL;
2532
    }
2533
    h->long_ref_count=0;
2534

    
2535
    for(i=0; i<h->short_ref_count; i++){
2536
        h->short_ref[i]->reference=0;
2537
        h->short_ref[i]= NULL;
2538
    }
2539
    h->short_ref_count=0;
2540
}
2541

    
2542
//static void 
2543
/**
2544
 *
2545
 * @return the removed picture or NULL if an error occures
2546
 */
2547
static Picture * remove_short(H264Context *h, int frame_num){
2548
    int i;
2549
    
2550
    for(i=0; i<h->short_ref_count; i++){
2551
        Picture *pic= h->short_ref[i];
2552
        if(pic->frame_num == frame_num){
2553
            h->short_ref[i]= NULL;
2554
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2555
            h->short_ref_count--;
2556
            return pic;
2557
        }
2558
    }
2559
    return NULL;
2560
}
2561

    
2562
/**
2563
 *
2564
 * @return the removed picture or NULL if an error occures
2565
 */
2566
static Picture * remove_long(H264Context *h, int i){
2567
    Picture *pic;
2568

    
2569
    if(i >= h->long_ref_count) return NULL;
2570
    pic= h->long_ref[i];
2571
    if(pic==NULL) return NULL;
2572
    
2573
    h->long_ref[i]= NULL;
2574
    memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2575
    h->long_ref_count--;
2576

    
2577
    return pic;
2578
}
2579

    
2580
/**
2581
 * Executes the reference picture marking (memory management control operations).
2582
 */
2583
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2584
    MpegEncContext * const s = &h->s;
2585
    int i;
2586
    int current_is_long=0;
2587
    Picture *pic;
2588
    
2589
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2590
        printf("no mmco here\n");
2591
        
2592
    for(i=0; i<mmco_count; i++){
2593
        if(s->avctx->debug&FF_DEBUG_MMCO)
2594
            printf("mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
2595

    
2596
        switch(mmco[i].opcode){
2597
        case MMCO_SHORT2UNUSED:
2598
            pic= remove_short(h, mmco[i].short_frame_num);
2599
            if(pic==NULL) return -1;
2600
            pic->reference= 0;
2601
            break;
2602
        case MMCO_SHORT2LONG:
2603
            pic= remove_long(h, mmco[i].long_index);
2604
            if(pic) pic->reference=0;
2605
            
2606
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2607
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
2608
            break;
2609
        case MMCO_LONG2UNUSED:
2610
            pic= remove_long(h, mmco[i].long_index);
2611
            if(pic==NULL) return -1;
2612
            pic->reference= 0;
2613
            break;
2614
        case MMCO_LONG:
2615
            pic= remove_long(h, mmco[i].long_index);
2616
            if(pic) pic->reference=0;
2617
            
2618
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2619
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
2620
            h->long_ref_count++;
2621
            
2622
            current_is_long=1;
2623
            break;
2624
        case MMCO_SET_MAX_LONG:
2625
            assert(mmco[i].long_index <= 16);
2626
            while(mmco[i].long_index < h->long_ref_count){
2627
                pic= remove_long(h, mmco[i].long_index);
2628
                pic->reference=0;
2629
            }
2630
            while(mmco[i].long_index > h->long_ref_count){
2631
                h->long_ref[ h->long_ref_count++ ]= NULL;
2632
            }
2633
            break;
2634
        case MMCO_RESET:
2635
            while(h->short_ref_count){
2636
                pic= remove_short(h, h->short_ref[0]->frame_num);
2637
                pic->reference=0;
2638
            }
2639
            while(h->long_ref_count){
2640
                pic= remove_long(h, h->long_ref_count-1);
2641
                pic->reference=0;
2642
            }
2643
            break;
2644
        default: assert(0);
2645
        }
2646
    }
2647
    
2648
    if(!current_is_long){
2649
        pic= remove_short(h, s->current_picture_ptr->frame_num);
2650
        if(pic){
2651
            pic->reference=0;
2652
            fprintf(stderr, "illegal short term buffer state detected\n");
2653
        }
2654
        
2655
        if(h->short_ref_count)
2656
            memmove(&h->short_ref[1], &h->short_ref[0], (h->short_ref_count - 1)*sizeof(Picture*));
2657
        h->short_ref[0]= s->current_picture_ptr;    
2658
        h->short_ref[0]->long_ref=0;
2659
        h->short_ref_count++;
2660
    }
2661
    
2662
    return 0; 
2663
}
2664

    
2665
static int decode_ref_pic_marking(H264Context *h){
2666
    MpegEncContext * const s = &h->s;
2667
    int i;
2668
    
2669
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2670
        s->broken_link= get_bits1(&s->gb) -1;
2671
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2672
        if(h->mmco[0].long_index == -1)
2673
            h->mmco_index= 0;
2674
        else{
2675
            h->mmco[0].opcode= MMCO_LONG;
2676
            h->mmco_index= 1;
2677
        } 
2678
    }else{
2679
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2680
            for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) { 
2681
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2682

    
2683
                h->mmco[i].opcode= opcode;
2684
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2685
                    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
2686
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2687
                        fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2688
                        return -1;
2689
                    }*/
2690
                }
2691
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2692
                    h->mmco[i].long_index= get_ue_golomb(&s->gb);
2693
                    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){
2694
                        fprintf(stderr, "illegal long ref in memory management control operation %d\n", opcode);
2695
                        return -1;
2696
                    }
2697
                }
2698
                    
2699
                if(opcode > MMCO_LONG){
2700
                    fprintf(stderr, "illegal memory management control operation %d\n", opcode);
2701
                    return -1;
2702
                }
2703
            }
2704
            h->mmco_index= i;
2705
        }else{
2706
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2707

    
2708
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2709
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2710
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2711
                h->mmco_index= 1;
2712
            }else
2713
                h->mmco_index= 0;
2714
        }
2715
    }
2716
    
2717
    return 0; 
2718
}
2719

    
2720
static int init_poc(H264Context *h){
2721
    MpegEncContext * const s = &h->s;
2722
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2723
    int field_poc[2];
2724

    
2725
    if(h->nal_unit_type == NAL_IDR_SLICE){
2726
        h->frame_num_offset= 0;
2727
    }else{
2728
        if(h->frame_num < h->prev_frame_num)
2729
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2730
        else
2731
            h->frame_num_offset= h->prev_frame_num_offset;
2732
    }
2733

    
2734
    if(h->sps.poc_type==0){
2735
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2736

    
2737
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2738
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2739
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2740
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2741
        else
2742
            h->poc_msb = h->prev_poc_msb;
2743
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2744
        field_poc[0] = 
2745
        field_poc[1] = h->poc_msb + h->poc_lsb;
2746
        if(s->picture_structure == PICT_FRAME) 
2747
            field_poc[1] += h->delta_poc_bottom;
2748
    }else if(h->sps.poc_type==1){
2749
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2750
        int i;
2751

    
2752
        if(h->sps.poc_cycle_length != 0)
2753
            abs_frame_num = h->frame_num_offset + h->frame_num;
2754
        else
2755
            abs_frame_num = 0;
2756

    
2757
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
2758
            abs_frame_num--;
2759
            
2760
        expected_delta_per_poc_cycle = 0;
2761
        for(i=0; i < h->sps.poc_cycle_length; i++)
2762
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2763

    
2764
        if(abs_frame_num > 0){
2765
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2766
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2767

    
2768
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2769
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
2770
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2771
        } else
2772
            expectedpoc = 0;
2773

    
2774
        if(h->nal_ref_idc == 0) 
2775
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2776
        
2777
        field_poc[0] = expectedpoc + h->delta_poc[0];
2778
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2779

    
2780
        if(s->picture_structure == PICT_FRAME)
2781
            field_poc[1] += h->delta_poc[1];
2782
    }else{
2783
        int poc;
2784
        if(h->nal_unit_type == NAL_IDR_SLICE){
2785
            poc= 0;
2786
        }else{
2787
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2788
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2789
        }
2790
        field_poc[0]= poc;
2791
        field_poc[1]= poc;
2792
    }
2793
    
2794
    if(s->picture_structure != PICT_BOTTOM_FIELD)
2795
        s->current_picture_ptr->field_poc[0]= field_poc[0];
2796
    if(s->picture_structure != PICT_TOP_FIELD)
2797
        s->current_picture_ptr->field_poc[1]= field_poc[1];
2798
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2799
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2800

    
2801
    return 0;
2802
}
2803

    
2804
/**
2805
 * decodes a slice header.
2806
 * this will allso call MPV_common_init() and frame_start() as needed
2807
 */
2808
static int decode_slice_header(H264Context *h){
2809
    MpegEncContext * const s = &h->s;
2810
    int first_mb_in_slice, pps_id;
2811
    int num_ref_idx_active_override_flag;
2812
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2813
    float new_aspect;
2814

    
2815
    s->current_picture.reference= h->nal_ref_idc != 0;
2816

    
2817
    first_mb_in_slice= get_ue_golomb(&s->gb);
2818

    
2819
    h->slice_type= get_ue_golomb(&s->gb);
2820
    if(h->slice_type > 9){
2821
        fprintf(stderr, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
2822
    }
2823
    if(h->slice_type > 4){
2824
        h->slice_type -= 5;
2825
        h->slice_type_fixed=1;
2826
    }else
2827
        h->slice_type_fixed=0;
2828
    
2829
    h->slice_type= slice_type_map[ h->slice_type ];
2830
    
2831
    s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
2832
        
2833
    pps_id= get_ue_golomb(&s->gb);
2834
    if(pps_id>255){
2835
        fprintf(stderr, "pps_id out of range\n");
2836
        return -1;
2837
    }
2838
    h->pps= h->pps_buffer[pps_id];
2839
    h->sps= h->sps_buffer[ h->pps.sps_id ];
2840
    
2841
    s->mb_width= h->sps.mb_width;
2842
    s->mb_height= h->sps.mb_height;
2843
    h->mb_stride= s->mb_width + 1;
2844
    
2845
    h->b_stride=  s->mb_width*4;
2846
    h->b8_stride= s->mb_width*2;
2847

    
2848
    s->mb_x = first_mb_in_slice % s->mb_width;
2849
    s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
2850
    
2851
    s->width = 16*s->mb_width - 2*(h->pps.crop_left + h->pps.crop_right );
2852
    if(h->sps.frame_mbs_only_flag)
2853
        s->height= 16*s->mb_height - 2*(h->pps.crop_top  + h->pps.crop_bottom);
2854
    else
2855
        s->height= 16*s->mb_height - 4*(h->pps.crop_top  + h->pps.crop_bottom); //FIXME recheck
2856
    
2857
    if(h->pps.crop_left || h->pps.crop_top){
2858
        fprintf(stderr, "insane croping not completly supported, this could look slightly wrong ...\n");
2859
    }
2860

    
2861
    if(s->aspected_height) //FIXME emms at end of slice ?
2862
        new_aspect= h->sps.sar_width*s->width / (float)(s->height*h->sps.sar_height);
2863
    else
2864
        new_aspect=0;
2865

    
2866
    if (s->context_initialized 
2867
        && (   s->width != s->avctx->width || s->height != s->avctx->height 
2868
            || ABS(new_aspect - s->avctx->aspect_ratio) > 0.001)) {
2869
        free_tables(h);
2870
        MPV_common_end(s);
2871
    }
2872
    if (!s->context_initialized) {
2873
        if (MPV_common_init(s) < 0)
2874
            return -1;
2875

    
2876
        alloc_tables(h);
2877

    
2878
        s->avctx->width = s->width;
2879
        s->avctx->height = s->height;
2880
        s->avctx->aspect_ratio= new_aspect;
2881
    }
2882

    
2883
    if(first_mb_in_slice == 0){
2884
        frame_start(h);
2885
    }
2886

    
2887
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2888

    
2889
    if(h->sps.frame_mbs_only_flag){
2890
        s->picture_structure= PICT_FRAME;
2891
    }else{
2892
        if(get_bits1(&s->gb)) //field_pic_flag
2893
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2894
        else
2895
            s->picture_structure= PICT_FRAME;
2896
    }
2897

    
2898
    if(s->picture_structure==PICT_FRAME){
2899
        h->curr_pic_num=   h->frame_num;
2900
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2901
    }else{
2902
        h->curr_pic_num= 2*h->frame_num;
2903
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2904
    }
2905
        
2906
    if(h->nal_unit_type == NAL_IDR_SLICE){
2907
        int idr_pic_id= get_ue_golomb(&s->gb);
2908
    }
2909
   
2910
    if(h->sps.poc_type==0){
2911
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2912
        
2913
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2914
            h->delta_poc_bottom= get_se_golomb(&s->gb);
2915
        }
2916
    }
2917
    
2918
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2919
        h->delta_poc[0]= get_se_golomb(&s->gb);
2920
        
2921
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2922
            h->delta_poc[1]= get_se_golomb(&s->gb);
2923
    }
2924
    
2925
    init_poc(h);
2926
    
2927
    if(h->pps.redundant_pic_cnt_present){
2928
        h->redundant_pic_count= get_ue_golomb(&s->gb);
2929
    }
2930

    
2931
    //set defaults, might be overriden a few line later
2932
    h->ref_count[0]= h->pps.ref_count[0];
2933
    h->ref_count[1]= h->pps.ref_count[1];
2934

    
2935
    if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
2936
        if(h->slice_type == B_TYPE){
2937
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
2938
        }
2939
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
2940
    
2941
        if(num_ref_idx_active_override_flag){
2942
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2943
            if(h->slice_type==B_TYPE)
2944
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2945

    
2946
            if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
2947
                fprintf(stderr, "reference overflow\n");
2948
                return -1;
2949
            }
2950
        }
2951
    }
2952

    
2953
    if(first_mb_in_slice == 0){
2954
        fill_default_ref_list(h);
2955
    }
2956

    
2957
    decode_ref_pic_list_reordering(h);
2958

    
2959
    if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) 
2960
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
2961
        pred_weight_table(h);
2962
    
2963
    if(s->current_picture.reference)
2964
        decode_ref_pic_marking(h);
2965
    //FIXME CABAC stuff
2966

    
2967
    s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta
2968
    //FIXME qscale / qp ... stuff
2969
    if(h->slice_type == SP_TYPE){
2970
        int sp_for_switch_flag= get_bits1(&s->gb);
2971
    }
2972
    if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
2973
        int slice_qs_delta= get_se_golomb(&s->gb);
2974
    }
2975

    
2976
    if( h->pps.deblocking_filter_parameters_present ) {
2977
        h->disable_deblocking_filter_idc= get_ue_golomb(&s->gb);
2978
        if( h->disable_deblocking_filter_idc  !=  1 ) {
2979
            h->slice_alpha_c0_offset_div2= get_se_golomb(&s->gb);
2980
            h->slice_beta_offset_div2= get_se_golomb(&s->gb);
2981
        }
2982
    }else
2983
        h->disable_deblocking_filter_idc= 0;
2984

    
2985
#if 0 //FMO
2986
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2987
        slice_group_change_cycle= get_bits(&s->gb, ?);
2988
#endif
2989

    
2990
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2991
        printf("mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n", 
2992
               first_mb_in_slice, 
2993
               ff_get_pict_type_char(h->slice_type),
2994
               pps_id, h->frame_num,
2995
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
2996
               h->ref_count[0], h->ref_count[1],
2997
               s->qscale,
2998
               h->disable_deblocking_filter_idc
2999
               );
3000
    }
3001

    
3002
    return 0;
3003
}
3004

    
3005
/**
3006
 *
3007
 */
3008
static inline int get_level_prefix(GetBitContext *gb){
3009
    unsigned int buf;
3010
    int log;
3011
    
3012
    OPEN_READER(re, gb);
3013
    UPDATE_CACHE(re, gb);
3014
    buf=GET_CACHE(re, gb);
3015
    
3016
    log= 32 - av_log2(buf);
3017
#ifdef TRACE
3018
    print_bin(buf>>(32-log), log);
3019
    printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3020
#endif
3021

    
3022
    LAST_SKIP_BITS(re, gb, log);
3023
    CLOSE_READER(re, gb);
3024

    
3025
    return log-1;
3026
}
3027

    
3028
/**
3029
 * decodes a residual block.
3030
 * @param n block index
3031
 * @param scantable scantable
3032
 * @param max_coeff number of coefficients in the block
3033
 * @return <0 if an error occured
3034
 */
3035
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3036
    MpegEncContext * const s = &h->s;
3037
    const uint16_t *qmul= dequant_coeff[qp];
3038
    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};
3039
    int level[16], run[16];
3040
    int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3041

    
3042
    //FIXME put trailing_onex into the context
3043

    
3044
    if(n == CHROMA_DC_BLOCK_INDEX){
3045
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3046
        total_coeff= coeff_token>>2;
3047
    }else{    
3048
        if(n == LUMA_DC_BLOCK_INDEX){
3049
            total_coeff= pred_non_zero_count(h, 0);
3050
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3051
            total_coeff= coeff_token>>2;
3052
        }else{
3053
            total_coeff= pred_non_zero_count(h, n);
3054
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3055
            total_coeff= coeff_token>>2;
3056
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3057
        }
3058
    }
3059

    
3060
    //FIXME set last_non_zero?
3061

    
3062
    if(total_coeff==0)
3063
        return 0;
3064
        
3065
    trailing_ones= coeff_token&3;
3066
#ifdef TRACE
3067
    printf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3068
#endif
3069
    assert(total_coeff<=16);
3070
    
3071
    for(i=0; i<trailing_ones; i++){
3072
        level[i]= 1 - 2*get_bits1(gb);
3073
    }
3074

    
3075
    suffix_length= total_coeff > 10 && trailing_ones < 3;
3076

    
3077
    for(; i<total_coeff; i++){
3078
        const int prefix= get_level_prefix(gb);
3079
        int level_code, mask;
3080

    
3081
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3082
            if(suffix_length)
3083
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3084
            else
3085
                level_code= (prefix<<suffix_length); //part
3086
        }else if(prefix==14){
3087
            if(suffix_length)
3088
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3089
            else
3090
                level_code= prefix + get_bits(gb, 4); //part
3091
        }else if(prefix==15){
3092
            level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3093
            if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3094
        }else{
3095
            fprintf(stderr, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3096
            return -1;
3097
        }
3098

    
3099
        if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3100

    
3101
        mask= -(level_code&1);
3102
        level[i]= (((2+level_code)>>1) ^ mask) - mask;
3103

    
3104
        if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3105

    
3106
#if 1
3107
        if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3108
#else        
3109
        if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3110
        ? == prefix > 2 or sth
3111
#endif
3112
#ifdef TRACE
3113
printf("level: %d suffix_length:%d\n", level[i], suffix_length);
3114
#endif
3115
    }
3116

    
3117
    if(total_coeff == max_coeff)
3118
        zeros_left=0;
3119
    else{
3120
        if(n == CHROMA_DC_BLOCK_INDEX)
3121
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3122
        else
3123
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3124
    }
3125
    
3126
    for(i=0; i<total_coeff-1; i++){
3127
        if(zeros_left <=0)
3128
            break;
3129
        else if(zeros_left < 7){
3130
            run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3131
        }else{
3132
            run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3133
        }
3134
        zeros_left -= run[i];
3135
    }
3136

    
3137
    if(zeros_left<0){
3138
        fprintf(stderr, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3139
        return -1;
3140
    }
3141
    
3142
    for(; i<total_coeff-1; i++){
3143
        run[i]= 0;
3144
    }
3145

    
3146
    run[i]= zeros_left;
3147

    
3148
    coeff_num=-1;
3149
    if(n > 24){
3150
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3151
            int j;
3152

    
3153
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3154
            j= scantable[ coeff_num ];
3155

    
3156
            block[j]= level[i];
3157
        }
3158
    }else{
3159
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into  rundecode?
3160
            int j;
3161

    
3162
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3163
            j= scantable[ coeff_num ];
3164

    
3165
            block[j]= level[i] * qmul[j];
3166
//            printf("%d %d  ", block[j], qmul[j]);
3167
        }
3168
    }
3169
    return 0;
3170
}
3171

    
3172
/**
3173
 * decodes a macroblock
3174
 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3175
 */
3176
static int decode_mb(H264Context *h){
3177
    MpegEncContext * const s = &h->s;
3178
    const int mb_xy= s->mb_x + s->mb_y*h->mb_stride;
3179
    int mb_type, /*ref0,*/ partition_count, cbp;
3180

    
3181
    memset(h->mb, 0, sizeof(int16_t)*24*16); //FIXME avoid if allready clear (move after skip handlong?
3182

    
3183
#ifdef TRACE
3184
    printf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3185
#endif
3186

    
3187
    if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3188
        if(s->mb_skip_run==-1)
3189
            s->mb_skip_run= get_ue_golomb(&s->gb);
3190
        
3191
        if (s->mb_skip_run--) {
3192
            int i, mx, my;
3193
            /* skip mb */
3194
#if 0 //FIXME
3195
            for(i=0;i<6;i++)
3196
                s->block_last_index[i] = -1;
3197
         s->mv_type = MV_TYPE_16X16;
3198
            /* if P type, zero motion vector is implied */
3199
            s->mv_dir = MV_DIR_FORWARD;
3200
            s->mb_skiped = 1;
3201
#endif
3202
//FIXME b frame
3203
            mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3204

    
3205
            memset(h->non_zero_count[mb_xy], 0, 16);
3206
            memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3207

    
3208
            if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3209
                h->mb_field_decoding_flag= get_bits1(&s->gb);
3210
            }
3211

    
3212
            if(h->mb_field_decoding_flag)
3213
                mb_type|= MB_TYPE_INTERLACED;
3214
            
3215
            fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3216
            pred_pskip_motion(h, &mx, &my);
3217
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3218
            fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, (mx&0xFFFF)+(my<<16), 4);
3219
            write_back_motion(h, mb_type);
3220

    
3221
            s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3222
            h->slice_table[ mb_xy ]= h->slice_num;
3223

    
3224
            h->prev_mb_skiped= 1;
3225
            return 0;
3226
        }
3227
    }
3228
    if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3229
        if((s->mb_y&1)==0)
3230
            h->mb_field_decoding_flag = get_bits1(&s->gb);
3231
    }else
3232
        h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3233
    
3234
    h->prev_mb_skiped= 0;
3235
    
3236
    mb_type= get_ue_golomb(&s->gb);
3237
    if(h->slice_type == B_TYPE){
3238
        if(mb_type < 23){
3239
            partition_count= b_mb_type_info[mb_type].partition_count;
3240
            mb_type=         b_mb_type_info[mb_type].type;
3241
        }else{
3242
            mb_type -= 23;
3243
            goto decode_intra_mb;
3244
        }
3245
    }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3246
        if(mb_type < 5){
3247
            partition_count= p_mb_type_info[mb_type].partition_count;
3248
            mb_type=         p_mb_type_info[mb_type].type;
3249
        }else{
3250
            mb_type -= 5;
3251
            goto decode_intra_mb;
3252
        }
3253
    }else{
3254
       assert(h->slice_type == I_TYPE);
3255
decode_intra_mb:
3256
        if(mb_type > 25){
3257
            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);
3258
            return -1;
3259
        }
3260
        partition_count=0;
3261
        cbp= i_mb_type_info[mb_type].cbp;
3262
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3263
        mb_type= i_mb_type_info[mb_type].type;
3264
    }
3265

    
3266
    if(h->mb_field_decoding_flag)
3267
        mb_type |= MB_TYPE_INTERLACED;
3268

    
3269
    s->current_picture.mb_type[mb_xy]= mb_type;
3270
    h->slice_table[ mb_xy ]= h->slice_num;
3271
    
3272
    if(IS_INTRA_PCM(mb_type)){
3273
        const uint8_t *ptr;
3274
        int x, y, i;
3275
        
3276
        // we assume these blocks are very rare so we dont optimize it
3277
        align_get_bits(&s->gb);
3278
        
3279
        ptr= s->gb.buffer + get_bits_count(&s->gb);
3280
    
3281
        for(y=0; y<16; y++){
3282
            const int index= 4*(y&3) + 64*(y>>2);
3283
            for(x=0; x<16; x++){
3284
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3285
            }
3286
        }
3287
        for(y=0; y<8; y++){
3288
            const int index= 256 + 4*(y&3) + 32*(y>>2);
3289
            for(x=0; x<8; x++){
3290
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3291
            }
3292
        }
3293
        for(y=0; y<8; y++){
3294
            const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3295
            for(x=0; x<8; x++){
3296
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3297
            }
3298
        }
3299
    
3300
        skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3301
        
3302
        memset(h->non_zero_count[mb_xy], 16, 16);
3303
        
3304
        return 0;
3305
    }
3306
        
3307
    fill_caches(h, mb_type);
3308

    
3309
    //mb_pred
3310
    if(IS_INTRA(mb_type)){
3311
//            init_top_left_availability(h);
3312
            if(IS_INTRA4x4(mb_type)){
3313
                int i;
3314

    
3315
//                fill_intra4x4_pred_table(h);
3316
                for(i=0; i<16; i++){
3317
                    const int mode_coded= !get_bits1(&s->gb);
3318
                    const int predicted_mode=  pred_intra_mode(h, i);
3319
                    int mode;
3320

    
3321
                    if(mode_coded){
3322
                        const int rem_mode= get_bits(&s->gb, 3);
3323
                        if(rem_mode<predicted_mode)
3324
                            mode= rem_mode;
3325
                        else
3326
                            mode= rem_mode + 1;
3327
                    }else{
3328
                        mode= predicted_mode;
3329
                    }
3330
                    
3331
                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3332
                }
3333
                write_back_intra_pred_mode(h);
3334
                if( check_intra4x4_pred_mode(h) < 0)
3335
                    return -1;
3336
            }else{
3337
                h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3338
                if(h->intra16x16_pred_mode < 0)
3339
                    return -1;
3340
            }
3341
            h->chroma_pred_mode= get_ue_golomb(&s->gb);
3342

    
3343
            h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3344
            if(h->chroma_pred_mode < 0)
3345
                return -1;
3346
    }else if(partition_count==4){
3347
        int i, j, sub_partition_count[4], list, ref[2][4];
3348
        
3349
        if(h->slice_type == B_TYPE){
3350
            for(i=0; i<4; i++){
3351
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3352
                if(h->sub_mb_type[i] >=13){
3353
                    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);
3354
                    return -1;
3355
                }
3356
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3357
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3358
            }
3359
        }else{
3360
            assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3361
            for(i=0; i<4; i++){
3362
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3363
                if(h->sub_mb_type[i] >=4){
3364
                    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);
3365
                    return -1;
3366
                }
3367
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3368
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3369
            }
3370
        }
3371
        
3372
        for(list=0; list<2; list++){
3373
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3374
            if(ref_count == 0) continue;
3375
            for(i=0; i<4; i++){
3376
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3377
                    ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3378
                }else{
3379
                 //FIXME
3380
                    ref[list][i] = -1;
3381
                }
3382
            }
3383
        }
3384
        
3385
        for(list=0; list<2; list++){
3386
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3387
            if(ref_count == 0) continue;
3388

    
3389
            for(i=0; i<4; i++){
3390
                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3391
                h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3392

    
3393
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3394
                    const int sub_mb_type= h->sub_mb_type[i];
3395
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3396
                    for(j=0; j<sub_partition_count[i]; j++){
3397
                        int mx, my;
3398
                        const int index= 4*i + block_width*j;
3399
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3400
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3401
                        mx += get_se_golomb(&s->gb);
3402
                        my += get_se_golomb(&s->gb);
3403
#ifdef TRACE
3404
printf("final mv:%d %d\n", mx, my);
3405
#endif
3406
                        if(IS_SUB_8X8(sub_mb_type)){
3407
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= 
3408
                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3409
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= 
3410
                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3411
                        }else if(IS_SUB_8X4(sub_mb_type)){
3412
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3413
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3414
                        }else if(IS_SUB_4X8(sub_mb_type)){
3415
                            mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3416
                            mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3417
                        }else{
3418
                            assert(IS_SUB_4X4(sub_mb_type));
3419
                            mv_cache[ 0 ][0]= mx;
3420
                            mv_cache[ 0 ][1]= my;
3421
                        }
3422
                    }
3423
                }else{
3424
                    uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3425
                    p[0] = p[1]=
3426
                    p[8] = p[9]= 0;
3427
                }
3428
            }
3429
        }
3430
    }else if(!IS_DIRECT(mb_type)){
3431
        int list, mx, my, i;
3432
         //FIXME we should set ref_idx_l? to 0 if we use that later ...
3433
        if(IS_16X16(mb_type)){
3434
            for(list=0; list<2; list++){
3435
                if(h->ref_count[0]>0){
3436
                    if(IS_DIR(mb_type, 0, list)){
3437
                        const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3438
                        fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3439
                    }
3440
                }
3441
            }
3442
            for(list=0; list<2; list++){
3443
                if(IS_DIR(mb_type, 0, list)){
3444
                    pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3445
                    mx += get_se_golomb(&s->gb);
3446
                    my += get_se_golomb(&s->gb);
3447
#ifdef TRACE
3448
printf("final mv:%d %d\n", mx, my);
3449
#endif
3450
                    fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, (mx&0xFFFF) + (my<<16), 4);
3451
                }
3452
            }
3453
        }
3454
        else if(IS_16X8(mb_type)){
3455
            for(list=0; list<2; list++){
3456
                if(h->ref_count[list]>0){
3457
                    for(i=0; i<2; i++){
3458
                        if(IS_DIR(mb_type, i, list)){
3459
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3460
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3461
                        }
3462
                    }
3463
                }
3464
            }
3465
            for(list=0; list<2; list++){
3466
                for(i=0; i<2; i++){
3467
                    if(IS_DIR(mb_type, i, list)){
3468
                        pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3469
                        mx += get_se_golomb(&s->gb);
3470
                        my += get_se_golomb(&s->gb);
3471
#ifdef TRACE
3472
printf("final mv:%d %d\n", mx, my);
3473
#endif
3474
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (mx&0xFFFF) + (my<<16), 4);
3475
                    }
3476
                }
3477
            }
3478
        }else{
3479
            assert(IS_8X16(mb_type));
3480
            for(list=0; list<2; list++){
3481
                if(h->ref_count[list]>0){
3482
                    for(i=0; i<2; i++){
3483
                        if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3484
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3485
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3486
                        }
3487
                    }
3488
                }
3489
            }
3490
            for(list=0; list<2; list++){
3491
                for(i=0; i<2; i++){
3492
                    if(IS_DIR(mb_type, i, list)){
3493
                        pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3494
                        mx += get_se_golomb(&s->gb);
3495
                        my += get_se_golomb(&s->gb);
3496
#ifdef TRACE
3497
printf("final mv:%d %d\n", mx, my);
3498
#endif
3499
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (mx&0xFFFF) + (my<<16), 4);
3500
                    }
3501
                }
3502
            }
3503
        }
3504
    }
3505
    
3506
    if(IS_INTER(mb_type))
3507
        write_back_motion(h, mb_type);
3508
    
3509
    if(!IS_INTRA16x16(mb_type)){
3510
        cbp= get_ue_golomb(&s->gb);
3511
        if(cbp > 47){
3512
            fprintf(stderr, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3513
            return -1;
3514
        }
3515
        
3516
        if(IS_INTRA4x4(mb_type))
3517
            cbp= golomb_to_intra4x4_cbp[cbp];
3518
        else
3519
            cbp= golomb_to_inter_cbp[cbp];
3520
    }
3521

    
3522
    if(cbp || IS_INTRA16x16(mb_type)){
3523
        int i8x8, i4x4, chroma_idx;
3524
        int chroma_qp, dquant;
3525
        GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3526
        const uint8_t *scan, *dc_scan;
3527
        
3528
//        fill_non_zero_count_cache(h);
3529

    
3530
        if(IS_INTERLACED(mb_type)){
3531
            scan= field_scan;
3532
            dc_scan= luma_dc_field_scan;
3533
        }else{
3534
            scan= zigzag_scan;
3535
            dc_scan= luma_dc_zigzag_scan;
3536
        }
3537

    
3538
        dquant= get_se_golomb(&s->gb);
3539

    
3540
        if( dquant > 25 || dquant < -26 ){
3541
            fprintf(stderr, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3542
            return -1;
3543
        }
3544
        
3545
        s->qscale += dquant;
3546
        if(((unsigned)s->qscale) > 51){
3547
            if(s->qscale<0) s->qscale+= 52;
3548
            else            s->qscale-= 52;
3549
        }
3550
        
3551
        h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3552
        if(IS_INTRA16x16(mb_type)){
3553
            if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3554
                return -1; //FIXME continue if partotioned and other retirn -1 too
3555
            }
3556

    
3557
            assert((cbp&15) == 0 || (cbp&15) == 15);
3558

    
3559
            if(cbp&15){
3560
                for(i8x8=0; i8x8<4; i8x8++){
3561
                    for(i4x4=0; i4x4<4; i4x4++){
3562
                        const int index= i4x4 + 4*i8x8;
3563
                        if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3564
                            return -1;
3565
                        }
3566
                    }
3567
                }
3568
            }else{
3569
                memset(&h->non_zero_count_cache[8], 0, 8*4); //FIXME stupid & slow
3570
            }
3571
        }else{
3572
            for(i8x8=0; i8x8<4; i8x8++){
3573
                if(cbp & (1<<i8x8)){
3574
                    for(i4x4=0; i4x4<4; i4x4++){
3575
                        const int index= i4x4 + 4*i8x8;
3576
                        
3577
                        if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3578
                            return -1;
3579
                        }
3580
                    }
3581
                }else{
3582
                    uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3583
                    nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3584
                }
3585
            }
3586
        }
3587
        
3588
        if(cbp&0x30){
3589
            for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3590
                if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3591
                    return -1;
3592
                }
3593
        }
3594

    
3595
        if(cbp&0x20){
3596
            for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3597
                for(i4x4=0; i4x4<4; i4x4++){
3598
                    const int index= 16 + 4*chroma_idx + i4x4;
3599
                    if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3600
                        return -1;
3601
                    }
3602
                }
3603
            }
3604
        }else{
3605
            uint8_t * const nnz= &h->non_zero_count_cache[0];
3606
            nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3607
            nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3608
        }
3609
    }else{
3610
        memset(&h->non_zero_count_cache[8], 0, 8*5);
3611
    }
3612
    write_back_non_zero_count(h);
3613

    
3614
    return 0;
3615
}
3616

    
3617
static int decode_slice(H264Context *h){
3618
    MpegEncContext * const s = &h->s;
3619
    const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3620

    
3621
    s->mb_skip_run= -1;
3622
    
3623
#if 1
3624
    for(;;){
3625
        int ret= decode_mb(h);
3626
            
3627
        hl_decode_mb(h);
3628
        
3629
        if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
3630
            s->mb_y++;
3631
            ret= decode_mb(h);
3632
            
3633
            hl_decode_mb(h);
3634
            s->mb_y--;
3635
        }
3636

    
3637
        if(ret<0){
3638
            fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3639
            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);
3640

    
3641
            return -1;
3642
        }
3643
        
3644
        if(++s->mb_x >= s->mb_width){
3645
            s->mb_x=0;
3646
            ff_draw_horiz_band(s, 16*s->mb_y, 16);
3647
            if(++s->mb_y >= s->mb_height){
3648
#ifdef TRACE
3649
printf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3650
#endif
3651
                if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3652
                    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);
3653

    
3654
                    return 0;
3655
                }else{
3656
                    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);
3657

    
3658
                    return -1;
3659
                }
3660
            }
3661
        }
3662
        
3663
        if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3664
            if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3665
                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);
3666

    
3667
                return 0;
3668
            }else{
3669
                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);
3670

    
3671
                return -1;
3672
            }
3673
        }
3674
    }
3675
#endif
3676
#if 0
3677
    for(;s->mb_y < s->mb_height; s->mb_y++){
3678
        for(;s->mb_x < s->mb_width; s->mb_x++){
3679
            int ret= decode_mb(h);
3680
            
3681
            hl_decode_mb(h);
3682

3683
            if(ret<0){
3684
                fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3685
                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);
3686

3687
                return -1;
3688
            }
3689
        
3690
            if(++s->mb_x >= s->mb_width){
3691
                s->mb_x=0;
3692
                if(++s->mb_y >= s->mb_height){
3693
                    if(get_bits_count(s->gb) == s->gb.size_in_bits){
3694
                        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);
3695

3696
                        return 0;
3697
                    }else{
3698
                        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);
3699

3700
                        return -1;
3701
                    }
3702
                }
3703
            }
3704
        
3705
            if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
3706
                if(get_bits_count(s->gb) == s->gb.size_in_bits){
3707
                    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);
3708

3709
                    return 0;
3710
                }else{
3711
                    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);
3712

3713
                    return -1;
3714
                }
3715
            }
3716
        }
3717
        s->mb_x=0;
3718
        ff_draw_horiz_band(s, 16*s->mb_y, 16);
3719
    }
3720
#endif
3721
    return -1; //not reached
3722
}
3723

    
3724
static inline int decode_vui_parameters(H264Context *h, SPS *sps){
3725
    MpegEncContext * const s = &h->s;
3726
    int aspect_ratio_info_present_flag, aspect_ratio_idc;
3727

    
3728
    aspect_ratio_info_present_flag= get_bits1(&s->gb);
3729
    
3730
    if( aspect_ratio_info_present_flag ) {
3731
        aspect_ratio_idc= get_bits(&s->gb, 8);
3732
        if( aspect_ratio_idc == EXTENDED_SAR ) {
3733
            sps->sar_width= get_bits(&s->gb, 16);
3734
            sps->sar_height= get_bits(&s->gb, 16);
3735
        }else if(aspect_ratio_idc < 16){
3736
            sps->sar_width=  pixel_aspect[aspect_ratio_idc][0];
3737
            sps->sar_height= pixel_aspect[aspect_ratio_idc][1];
3738
        }else{
3739
            fprintf(stderr, "illegal aspect ratio\n");
3740
            return -1;
3741
        }
3742
    }else{
3743
        sps->sar_width= 
3744
        sps->sar_height= 0;
3745
    }
3746
//            s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
3747
#if 0
3748
| overscan_info_present_flag                        |0  |u(1)    |
3749
| if( overscan_info_present_flag )                  |   |        |
3750
|  overscan_appropriate_flag                        |0  |u(1)    |
3751
| video_signal_type_present_flag                    |0  |u(1)    |
3752
| if( video_signal_type_present_flag ) {            |   |        |
3753
|  video_format                                     |0  |u(3)    |
3754
|  video_full_range_flag                            |0  |u(1)    |
3755
|  colour_description_present_flag                  |0  |u(1)    |
3756
|  if( colour_description_present_flag ) {          |   |        |
3757
|   colour_primaries                                |0  |u(8)    |
3758
|   transfer_characteristics                        |0  |u(8)    |
3759
|   matrix_coefficients                             |0  |u(8)    |
3760
|  }                                                |   |        |
3761
| }                                                 |   |        |
3762
| chroma_location_info_present_flag                 |0  |u(1)    |
3763
| if ( chroma_location_info_present_flag ) {        |   |        |
3764
|  chroma_sample_location_type_top_field            |0  |ue(v)   |
3765
|  chroma_sample_location_type_bottom_field         |0  |ue(v)   |
3766
| }                                                 |   |        |
3767
| timing_info_present_flag                          |0  |u(1)    |
3768
| if( timing_info_present_flag ) {                  |   |        |
3769
|  num_units_in_tick                                |0  |u(32)   |
3770
|  time_scale                                       |0  |u(32)   |
3771
|  fixed_frame_rate_flag                            |0  |u(1)    |
3772
| }                                                 |   |        |
3773
| nal_hrd_parameters_present_flag                   |0  |u(1)    |
3774
| if( nal_hrd_parameters_present_flag  = =  1)      |   |        |
3775
|  hrd_parameters( )                                |   |        |
3776
| vcl_hrd_parameters_present_flag                   |0  |u(1)    |
3777
| if( vcl_hrd_parameters_present_flag  = =  1)      |   |        |
3778
|  hrd_parameters( )                                |   |        |
3779
| if( ( nal_hrd_parameters_present_flag  = =  1  | ||   |        |
3780
|                                                   |   |        |
3781
|( vcl_hrd_parameters_present_flag  = =  1 ) )      |   |        |
3782
|  low_delay_hrd_flag                               |0  |u(1)    |
3783
| bitstream_restriction_flag                        |0  |u(1)    |
3784
| if( bitstream_restriction_flag ) {                |0  |u(1)    |
3785
|  motion_vectors_over_pic_boundaries_flag          |0  |u(1)    |
3786
|  max_bytes_per_pic_denom                          |0  |ue(v)   |
3787
|  max_bits_per_mb_denom                            |0  |ue(v)   |
3788
|  log2_max_mv_length_horizontal                    |0  |ue(v)   |
3789
|  log2_max_mv_length_vertical                      |0  |ue(v)   |
3790
|  num_reorder_frames                               |0  |ue(v)   |
3791
|  max_dec_frame_buffering                          |0  |ue(v)   |
3792
| }                                                 |   |        |
3793
|}                                                  |   |        |
3794
#endif
3795
    return 0;
3796
}
3797

    
3798
static inline int decode_seq_parameter_set(H264Context *h){
3799
    MpegEncContext * const s = &h->s;
3800
    int profile_idc, level_idc, multiple_slice_groups, arbitrary_slice_order, redundant_slices;
3801
    int sps_id, i;
3802
    SPS *sps;
3803
    
3804
    profile_idc= get_bits(&s->gb, 8);
3805
    level_idc= get_bits(&s->gb, 8);
3806
    multiple_slice_groups= get_bits1(&s->gb);
3807
    arbitrary_slice_order= get_bits1(&s->gb);
3808
    redundant_slices= get_bits1(&s->gb);
3809
    
3810
    sps_id= get_ue_golomb(&s->gb);
3811
    
3812
    sps= &h->sps_buffer[ sps_id ];
3813
    
3814
    sps->profile_idc= profile_idc;
3815
    sps->level_idc= level_idc;
3816
    sps->multiple_slice_groups= multiple_slice_groups;
3817
    sps->arbitrary_slice_order= arbitrary_slice_order;
3818
    sps->redundant_slices= redundant_slices;
3819
    
3820
    sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
3821

    
3822
    sps->poc_type= get_ue_golomb(&s->gb);
3823
    
3824
    if(sps->poc_type == 0){ //FIXME #define
3825
        sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
3826
    } else if(sps->poc_type == 1){//FIXME #define
3827
        sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
3828
        sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
3829
        sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
3830
        sps->poc_cycle_length= get_ue_golomb(&s->gb);
3831
        
3832
        for(i=0; i<sps->poc_cycle_length; i++)
3833
            sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
3834
    }
3835
    if(sps->poc_type > 2){
3836
        fprintf(stderr, "illegal POC type %d\n", sps->poc_type);
3837
        return -1;
3838
    }
3839

    
3840
    sps->ref_frame_count= get_ue_golomb(&s->gb);
3841
    sps->required_frame_num_update_behaviour_flag= get_bits1(&s->gb);
3842
    sps->mb_width= get_ue_golomb(&s->gb) + 1;
3843
    sps->mb_height= get_ue_golomb(&s->gb) + 1;
3844
    sps->frame_mbs_only_flag= get_bits1(&s->gb);
3845
    if(!sps->frame_mbs_only_flag)
3846
        sps->mb_aff= get_bits1(&s->gb);
3847
    else
3848
        sps->mb_aff= 0;
3849

    
3850
    sps->direct_8x8_inference_flag= get_bits1(&s->gb);
3851

    
3852
    sps->vui_parameters_present_flag= get_bits1(&s->gb);
3853
    if( sps->vui_parameters_present_flag )
3854
        decode_vui_parameters(h, sps);
3855
    
3856
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3857
        printf("sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s %s\n", 
3858
               sps_id, sps->profile_idc, sps->level_idc,
3859
               sps->poc_type,
3860
               sps->ref_frame_count,
3861
               sps->mb_width, sps->mb_height,
3862
               sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
3863
               sps->direct_8x8_inference_flag ? "8B8" : "",
3864
               sps->vui_parameters_present_flag ? "VUI" : ""
3865
               );
3866
    }
3867
    return 0;
3868
}
3869

    
3870
static inline int decode_picture_parameter_set(H264Context *h){
3871
    MpegEncContext * const s = &h->s;
3872
    int pps_id= get_ue_golomb(&s->gb);
3873
    PPS *pps= &h->pps_buffer[pps_id];
3874
    
3875
    pps->sps_id= get_ue_golomb(&s->gb);
3876
    pps->cabac= get_bits1(&s->gb);
3877
    pps->pic_order_present= get_bits1(&s->gb);
3878
    pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
3879
    if(pps->slice_group_count > 1 ){
3880
        pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
3881
fprintf(stderr, "FMO not supported\n");
3882
        switch(pps->mb_slice_group_map_type){
3883
        case 0:
3884
#if 0
3885
|   for( i = 0; i <= num_slice_groups_minus1; i++ ) |   |        |
3886
|    run_length[ i ]                                |1  |ue(v)   |
3887
#endif
3888
            break;
3889
        case 2:
3890
#if 0
3891
|   for( i = 0; i < num_slice_groups_minus1; i++ )  |   |        |
3892
|{                                                  |   |        |
3893
|    top_left_mb[ i ]                               |1  |ue(v)   |
3894
|    bottom_right_mb[ i ]                           |1  |ue(v)   |
3895
|   }                                               |   |        |
3896
#endif
3897
            break;
3898
        case 3:
3899
        case 4:
3900
        case 5:
3901
#if 0
3902
|   slice_group_change_direction_flag               |1  |u(1)    |
3903
|   slice_group_change_rate_minus1                  |1  |ue(v)   |
3904
#endif
3905
            break;
3906
        case 6:
3907
#if 0
3908
|   slice_group_id_cnt_minus1                       |1  |ue(v)   |
3909
|   for( i = 0; i <= slice_group_id_cnt_minus1; i++ |   |        |
3910
|)                                                  |   |        |
3911
|    slice_group_id[ i ]                            |1  |u(v)    |
3912
#endif
3913
        }
3914
    }
3915
    pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3916
    pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3917
    if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
3918
        fprintf(stderr, "reference overflow (pps)\n");
3919
        return -1;
3920
    }
3921
    
3922
    pps->weighted_pred= get_bits1(&s->gb);
3923
    pps->weighted_bipred_idc= get_bits(&s->gb, 2);
3924
    pps->init_qp= get_se_golomb(&s->gb) + 26;
3925
    pps->init_qs= get_se_golomb(&s->gb) + 26;
3926
    pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
3927
    pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
3928
    pps->constrained_intra_pred= get_bits1(&s->gb);
3929
    pps->redundant_pic_cnt_present = get_bits1(&s->gb);
3930
    pps->crop= get_bits1(&s->gb);
3931
    if(pps->crop){
3932
        pps->crop_left  = get_ue_golomb(&s->gb);
3933
        pps->crop_right = get_ue_golomb(&s->gb);
3934
        pps->crop_top   = get_ue_golomb(&s->gb);
3935
        pps->crop_bottom= get_ue_golomb(&s->gb);
3936
    }else{
3937
        pps->crop_left  = 
3938
        pps->crop_right = 
3939
        pps->crop_top   = 
3940
        pps->crop_bottom= 0;
3941
    }
3942
    
3943
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3944
        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", 
3945
               pps_id, pps->sps_id,
3946
               pps->cabac ? "CABAC" : "CAVLC",
3947
               pps->slice_group_count,
3948
               pps->ref_count[0], pps->ref_count[1],
3949
               pps->weighted_pred ? "weighted" : "",
3950
               pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
3951
               pps->deblocking_filter_parameters_present ? "LPAR" : "",
3952
               pps->constrained_intra_pred ? "CONSTR" : "",
3953
               pps->redundant_pic_cnt_present ? "REDU" : "",
3954
               pps->crop_left, pps->crop_right, 
3955
               pps->crop_top, pps->crop_bottom
3956
               );
3957
    }
3958
    
3959
    return 0;
3960
}
3961

    
3962
/**
3963
 * finds the end of the current frame in the bitstream.
3964
 * @return the position of the first byte of the next frame, or -1
3965
 */
3966
static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){
3967
    ParseContext *pc= &s->parse_context;
3968
    int last_addr, i;
3969
    uint32_t state;
3970
//printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
3971
//    mb_addr= pc->mb_addr - 1;
3972
    state= pc->state;
3973
    //FIXME this will fail with slices
3974
    for(i=0; i<buf_size; i++){
3975
        state= (state<<8) | buf[i];
3976
        if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
3977
            if(pc->frame_start_found){
3978
                pc->state=-1; 
3979
                pc->frame_start_found= 0;
3980
                return i-3;
3981
            }
3982
            pc->frame_start_found= 1;
3983
        }
3984
    }
3985
    
3986
    pc->state= state;
3987
    return -1;
3988
}
3989

    
3990
static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
3991
    MpegEncContext * const s = &h->s;
3992
    AVCodecContext * const avctx= s->avctx;
3993
    int buf_index=0;
3994
    int i;
3995
#if 0    
3996
    for(i=0; i<32; i++){
3997
        printf("%X ", buf[i]);
3998
    }
3999
#endif
4000
    for(;;){
4001
        int consumed;
4002
        int dst_length;
4003
        int bit_length;
4004
        uint8_t *ptr;
4005
        
4006
        // start code prefix search
4007
        for(; buf_index + 3 < buf_size; buf_index++){
4008
            // this should allways succeed in the first iteration
4009
            if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
4010
                break;
4011
        }
4012
        
4013
        if(buf_index+3 >= buf_size) break;
4014
        
4015
        buf_index+=3;
4016
        
4017
        ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index);
4018
        if(ptr[dst_length - 1] == 0) dst_length--;
4019
        bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
4020

    
4021
        if(s->avctx->debug&FF_DEBUG_STARTCODE){
4022
            printf("NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
4023
        }
4024
        
4025
        buf_index += consumed;
4026

    
4027
        if(h->nal_ref_idc < s->hurry_up)
4028
            continue;
4029
        
4030
        switch(h->nal_unit_type){
4031
        case NAL_IDR_SLICE:
4032
            idr(h); //FIXME ensure we dont loose some frames if there is reordering
4033
        case NAL_SLICE:
4034
            init_get_bits(&s->gb, ptr, bit_length);
4035
            h->intra_gb_ptr=
4036
            h->inter_gb_ptr= &s->gb;
4037
            s->data_partitioning = 0;
4038
            
4039
            if(decode_slice_header(h) < 0) return -1;
4040
            if(h->redundant_pic_count==0)
4041
                decode_slice(h);
4042
            break;
4043
        case NAL_DPA:
4044
            init_get_bits(&s->gb, ptr, bit_length);
4045
            h->intra_gb_ptr=
4046
            h->inter_gb_ptr= NULL;
4047
            s->data_partitioning = 1;
4048
            
4049
            if(decode_slice_header(h) < 0) return -1;
4050
            break;
4051
        case NAL_DPB:
4052
            init_get_bits(&h->intra_gb, ptr, bit_length);
4053
            h->intra_gb_ptr= &h->intra_gb;
4054
            break;
4055
        case NAL_DPC:
4056
            init_get_bits(&h->inter_gb, ptr, bit_length);
4057
            h->inter_gb_ptr= &h->inter_gb;
4058
            
4059
            if(h->redundant_pic_count==0)
4060
                decode_slice(h);
4061
            break;
4062
        case NAL_SEI:
4063
            break;
4064
        case NAL_SPS:
4065
            init_get_bits(&s->gb, ptr, bit_length);
4066
            decode_seq_parameter_set(h);
4067
            
4068
            if(s->flags& CODEC_FLAG_LOW_DELAY)
4069
                s->low_delay=1;
4070
      
4071
            avctx->has_b_frames= !s->low_delay;
4072
            break;
4073
        case NAL_PPS:
4074
            init_get_bits(&s->gb, ptr, bit_length);
4075
            
4076
            decode_picture_parameter_set(h);
4077

    
4078
            break;
4079
        case NAL_PICTURE_DELIMITER:
4080
            break;
4081
        case NAL_FILTER_DATA:
4082
            break;
4083
        }        
4084

    
4085
        //FIXME move after where irt is set
4086
        s->current_picture.pict_type= s->pict_type;
4087
        s->current_picture.key_frame= s->pict_type == I_TYPE;
4088
    }
4089
    
4090
    h->prev_frame_num_offset= h->frame_num_offset;
4091
    h->prev_frame_num= h->frame_num;
4092
    if(s->current_picture_ptr->reference){
4093
        h->prev_poc_msb= h->poc_msb;
4094
        h->prev_poc_lsb= h->poc_lsb;
4095
    }
4096
    if(s->current_picture_ptr->reference)
4097
        execute_ref_pic_marking(h, h->mmco, h->mmco_index);
4098
    else
4099
        assert(h->mmco_index=0);
4100

    
4101
    ff_er_frame_end(s);
4102
    MPV_frame_end(s);
4103

    
4104
    return buf_index;
4105
}
4106

    
4107
/**
4108
 * retunrs the number of bytes consumed for building the current frame
4109
 */
4110
static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
4111
    if(s->flags&CODEC_FLAG_TRUNCATED){
4112
        pos -= s->parse_context.last_index;
4113
        if(pos<0) pos=0; // FIXME remove (uneeded?)
4114
        
4115
        return pos;
4116
    }else{
4117
        if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
4118
        if(pos+10>buf_size) pos=buf_size; // oops ;)
4119

    
4120
        return pos;
4121
    }
4122
}
4123

    
4124
static int decode_frame(AVCodecContext *avctx, 
4125
                             void *data, int *data_size,
4126
                             uint8_t *buf, int buf_size)
4127
{
4128
    H264Context *h = avctx->priv_data;
4129
    MpegEncContext *s = &h->s;
4130
    AVFrame *pict = data; 
4131
    float new_aspect;
4132
    int buf_index;
4133
    
4134
    s->flags= avctx->flags;
4135

    
4136
    *data_size = 0;
4137
   
4138
   /* no supplementary picture */
4139
    if (buf_size == 0) {
4140
        return 0;
4141
    }
4142
    
4143
    if(s->flags&CODEC_FLAG_TRUNCATED){
4144
        int next= find_frame_end(s, buf, buf_size);
4145
        
4146
        if( ff_combine_frame(s, next, &buf, &buf_size) < 0 )
4147
            return buf_size;
4148
//printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
4149
    }
4150

    
4151
    if(s->avctx->extradata_size && s->picture_number==0){
4152
        if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) ) 
4153
            return -1;
4154
    }
4155

    
4156
    buf_index=decode_nal_units(h, buf, buf_size);
4157
    if(buf_index < 0) 
4158
        return -1;
4159

    
4160
    //FIXME do something with unavailable reference frames    
4161
 
4162
//    if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
4163
#if 0
4164
    if(s->pict_type==B_TYPE || s->low_delay){
4165
        *pict= *(AVFrame*)&s->current_picture;
4166
    } else {
4167
        *pict= *(AVFrame*)&s->last_picture;
4168
    }
4169
#endif
4170
    *pict= *(AVFrame*)&s->current_picture; //FIXME 
4171
    assert(pict->data[0]);
4172
//printf("out %d\n", (int)pict->data[0]);
4173
    if(avctx->debug&FF_DEBUG_QP){
4174
        int8_t *qtab= pict->qscale_table;
4175
        int x,y;
4176
        
4177
        for(y=0; y<s->mb_height; y++){
4178
            for(x=0; x<s->mb_width; x++){
4179
                printf("%2d ", qtab[x + y*s->mb_width]);
4180
            }
4181
            printf("\n");
4182
        }
4183
        printf("\n");
4184
    }
4185
#if 0 //?
4186

4187
    /* Return the Picture timestamp as the frame number */
4188
    /* we substract 1 because it is added on utils.c    */
4189
    avctx->frame_number = s->picture_number - 1;
4190
#endif
4191
#if 0
4192
    /* dont output the last pic after seeking */
4193
    if(s->last_picture_ptr || s->low_delay)
4194
    //Note this isnt a issue as a IDR pic should flush teh buffers
4195
#endif
4196
        *data_size = sizeof(AVFrame);
4197
    return get_consumed_bytes(s, buf_index, buf_size);
4198
}
4199
#if 0
4200
static inline void fill_mb_avail(H264Context *h){
4201
    MpegEncContext * const s = &h->s;
4202
    const int mb_xy= s->mb_x + s->mb_y*h->mb_stride;
4203

4204
    if(s->mb_y){
4205
        h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - h->mb_stride - 1] == h->slice_num;
4206
        h->mb_avail[1]=                            h->slice_table[mb_xy - h->mb_stride    ] == h->slice_num;
4207
        h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - h->mb_stride + 1] == h->slice_num;
4208
    }else{
4209
        h->mb_avail[0]=
4210
        h->mb_avail[1]=
4211
        h->mb_avail[2]= 0;
4212
    }
4213
    h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
4214
    h->mb_avail[4]= 1; //FIXME move out
4215
    h->mb_avail[5]= 0; //FIXME move out
4216
}
4217
#endif
4218

    
4219
#if 0 //selftest
4220
#define COUNT 8000
4221
#define SIZE (COUNT*40)
4222
int main(){
4223
    int i;
4224
    uint8_t temp[SIZE];
4225
    PutBitContext pb;
4226
    GetBitContext gb;
4227
//    int int_temp[10000];
4228
    DSPContext dsp;
4229
    AVCodecContext avctx;
4230
    
4231
    dsputil_init(&dsp, &avctx);
4232

4233
    init_put_bits(&pb, temp, SIZE, NULL, NULL);
4234
    printf("testing unsigned exp golomb\n");
4235
    for(i=0; i<COUNT; i++){
4236
        START_TIMER
4237
        set_ue_golomb(&pb, i);
4238
        STOP_TIMER("set_ue_golomb");
4239
    }
4240
    flush_put_bits(&pb);
4241
    
4242
    init_get_bits(&gb, temp, 8*SIZE);
4243
    for(i=0; i<COUNT; i++){
4244
        int j, s;
4245
        
4246
        s= show_bits(&gb, 24);
4247
        
4248
        START_TIMER
4249
        j= get_ue_golomb(&gb);
4250
        if(j != i){
4251
            printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4252
//            return -1;
4253
        }
4254
        STOP_TIMER("get_ue_golomb");
4255
    }
4256
    
4257
    
4258
    init_put_bits(&pb, temp, SIZE, NULL, NULL);
4259
    printf("testing signed exp golomb\n");
4260
    for(i=0; i<COUNT; i++){
4261
        START_TIMER
4262
        set_se_golomb(&pb, i - COUNT/2);
4263
        STOP_TIMER("set_se_golomb");
4264
    }
4265
    flush_put_bits(&pb);
4266
    
4267
    init_get_bits(&gb, temp, 8*SIZE);
4268
    for(i=0; i<COUNT; i++){
4269
        int j, s;
4270
        
4271
        s= show_bits(&gb, 24);
4272
        
4273
        START_TIMER
4274
        j= get_se_golomb(&gb);
4275
        if(j != i - COUNT/2){
4276
            printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4277
//            return -1;
4278
        }
4279
        STOP_TIMER("get_se_golomb");
4280
    }
4281

4282
    printf("testing 4x4 (I)DCT\n");
4283
    
4284
    DCTELEM block[16];
4285
    uint8_t src[16], ref[16];
4286
    uint64_t error= 0, max_error=0;
4287

4288
    for(i=0; i<COUNT; i++){
4289
        int j;
4290
//        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
4291
        for(j=0; j<16; j++){
4292
            ref[j]= random()%255;
4293
            src[j]= random()%255;
4294
        }
4295

4296
        h264_diff_dct_c(block, src, ref, 4);
4297
        
4298
        //normalize
4299
        for(j=0; j<16; j++){
4300
//            printf("%d ", block[j]);
4301
            block[j]= block[j]*4;
4302
            if(j&1) block[j]= (block[j]*4 + 2)/5;
4303
            if(j&4) block[j]= (block[j]*4 + 2)/5;
4304
        }
4305
//        printf("\n");
4306
        
4307
        h264_add_idct_c(ref, block, 4);
4308
/*        for(j=0; j<16; j++){
4309
            printf("%d ", ref[j]);
4310
        }
4311
        printf("\n");*/
4312
            
4313
        for(j=0; j<16; j++){
4314
            int diff= ABS(src[j] - ref[j]);
4315
            
4316
            error+= diff*diff;
4317
            max_error= FFMAX(max_error, diff);
4318
        }
4319
    }
4320
    printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
4321
#if 0
4322
    printf("testing quantizer\n");
4323
    for(qp=0; qp<52; qp++){
4324
        for(i=0; i<16; i++)
4325
            src1_block[i]= src2_block[i]= random()%255;
4326
        
4327
    }
4328
#endif
4329
    printf("Testing NAL layer\n");
4330
    
4331
    uint8_t bitstream[COUNT];
4332
    uint8_t nal[COUNT*2];
4333
    H264Context h;
4334
    memset(&h, 0, sizeof(H264Context));
4335
    
4336
    for(i=0; i<COUNT; i++){
4337
        int zeros= i;
4338
        int nal_length;
4339
        int consumed;
4340
        int out_length;
4341
        uint8_t *out;
4342
        int j;
4343
        
4344
        for(j=0; j<COUNT; j++){
4345
            bitstream[j]= (random() % 255) + 1;
4346
        }
4347
        
4348
        for(j=0; j<zeros; j++){
4349
            int pos= random() % COUNT;
4350
            while(bitstream[pos] == 0){
4351
                pos++;
4352
                pos %= COUNT;
4353
            }
4354
            bitstream[pos]=0;
4355
        }
4356
        
4357
        START_TIMER
4358
        
4359
        nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
4360
        if(nal_length<0){
4361
            printf("encoding failed\n");
4362
            return -1;
4363
        }
4364
        
4365
        out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
4366

    
4367
        STOP_TIMER("NAL")
4368
        
4369
        if(out_length != COUNT){
4370
            printf("incorrect length %d %d\n", out_length, COUNT);
4371
            return -1;
4372
        }
4373
        
4374
        if(consumed != nal_length){
4375
            printf("incorrect consumed length %d %d\n", nal_length, consumed);
4376
            return -1;
4377
        }
4378
        
4379
        if(memcmp(bitstream, out, COUNT)){
4380
            printf("missmatch\n");
4381
            return -1;
4382
        }
4383
    }
4384
    
4385
    printf("Testing RBSP\n");
4386
    
4387
    
4388
    return 0;
4389
}
4390
#endif
4391

    
4392

    
4393
static int decode_end(AVCodecContext *avctx)
4394
{
4395
    H264Context *h = avctx->priv_data;
4396
    MpegEncContext *s = &h->s;
4397
    
4398
    free_tables(h); //FIXME cleanup init stuff perhaps
4399
    MPV_common_end(s);
4400

    
4401
//    memset(h, 0, sizeof(H264Context));
4402
        
4403
    return 0;
4404
}
4405

    
4406

    
4407
AVCodec h264_decoder = {
4408
    "h264",
4409
    CODEC_TYPE_VIDEO,
4410
    CODEC_ID_H264,
4411
    sizeof(H264Context),
4412
    decode_init,
4413
    NULL,
4414
    decode_end,
4415
    decode_frame,
4416
    /*CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_DR1 | */CODEC_CAP_TRUNCATED,
4417
};
4418