Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ c58222c5

History | View | Annotate | Download (150 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 log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
63
    int poc_type;                      ///< pic_order_cnt_type
64
    int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
65
    int delta_pic_order_always_zero_flag;
66
    int offset_for_non_ref_pic;
67
    int offset_for_top_to_bottom_field;
68
    int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
69
    int ref_frame_count;               ///< num_ref_frames
70
    int gaps_in_frame_num_allowed_flag;
71
    int mb_width;                      ///< frame_width_in_mbs_minus1 + 1
72
    int mb_height;                     ///< frame_height_in_mbs_minus1 + 1
73
    int frame_mbs_only_flag;
74
    int mb_aff;                        ///<mb_adaptive_frame_field_flag
75
    int direct_8x8_inference_flag;
76
    int crop;                   ///< frame_cropping_flag
77
    int crop_left;              ///< frame_cropping_rect_left_offset
78
    int crop_right;             ///< frame_cropping_rect_right_offset
79
    int crop_top;               ///< frame_cropping_rect_top_offset
80
    int crop_bottom;            ///< frame_cropping_rect_bottom_offset
81
    int vui_parameters_present_flag;
82
    int sar_width;
83
    int sar_height;
84
    short offset_for_ref_frame[256]; //FIXME dyn aloc?
85
}SPS;
86

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

    
107
/**
108
 * Memory management control operation opcode.
109
 */
110
typedef enum MMCOOpcode{
111
    MMCO_END=0,
112
    MMCO_SHORT2UNUSED,
113
    MMCO_LONG2UNUSED,
114
    MMCO_SHORT2LONG,
115
    MMCO_SET_MAX_LONG,
116
    MMCO_RESET, 
117
    MMCO_LONG,
118
} MMCOOpcode;
119

    
120
/**
121
 * Memory management control operation.
122
 */
123
typedef struct MMCO{
124
    MMCOOpcode opcode;
125
    int short_frame_num;
126
    int long_index;
127
} MMCO;
128

    
129
/**
130
 * H264Context
131
 */
132
typedef struct H264Context{
133
    MpegEncContext s;
134
    int nal_ref_idc;        
135
    int nal_unit_type;
136
#define NAL_SLICE                1
137
#define NAL_DPA                        2
138
#define NAL_DPB                        3
139
#define NAL_DPC                        4
140
#define NAL_IDR_SLICE                5
141
#define NAL_SEI                        6
142
#define NAL_SPS                        7
143
#define NAL_PPS                        8
144
#define NAL_PICTURE_DELIMITER        9
145
#define NAL_FILTER_DATA                10
146
    uint8_t *rbsp_buffer;
147
    int rbsp_buffer_size;
148

    
149
    int chroma_qp; //QPc
150

    
151
    int prev_mb_skiped; //FIXME remove (IMHO not used)
152

    
153
    //prediction stuff
154
    int chroma_pred_mode;
155
    int intra16x16_pred_mode;
156
    
157
    int8_t intra4x4_pred_mode_cache[5*8];
158
    int8_t (*intra4x4_pred_mode)[8];
159
    void (*pred4x4  [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
160
    void (*pred8x8  [4+3])(uint8_t *src, int stride);
161
    void (*pred16x16[4+3])(uint8_t *src, int stride);
162
    unsigned int topleft_samples_available;
163
    unsigned int top_samples_available;
164
    unsigned int topright_samples_available;
165
    unsigned int left_samples_available;
166

    
167
    /**
168
     * non zero coeff count cache.
169
     * is 64 if not available.
170
     */
171
    uint8_t non_zero_count_cache[6*8];
172
    uint8_t (*non_zero_count)[16];
173

    
174
    /**
175
     * Motion vector cache.
176
     */
177
    int16_t mv_cache[2][5*8][2];
178
    int8_t ref_cache[2][5*8];
179
#define LIST_NOT_USED -1 //FIXME rename?
180
#define PART_NOT_AVAILABLE -2
181
    
182
    /**
183
     * is 1 if the specific list MV&references are set to 0,0,-2.
184
     */
185
    int mv_cache_clean[2];
186

    
187
    int block_offset[16+8];
188
    int chroma_subblock_offset[16]; //FIXME remove
189
    
190
    uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
191
    uint16_t *mb2b8_xy;
192
    int b_stride;
193
    int b8_stride;
194

    
195
    int halfpel_flag;
196
    int thirdpel_flag;
197

    
198
    int unknown_svq3_flag;
199
    int next_slice_index;
200

    
201
    SPS sps_buffer[MAX_SPS_COUNT];
202
    SPS sps; ///< current sps
203
    
204
    PPS pps_buffer[MAX_PPS_COUNT];
205
    /**
206
     * current pps
207
     */
208
    PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
209

    
210
    int slice_num;
211
    uint8_t *slice_table_base;
212
    uint8_t *slice_table;      ///< slice_table_base + mb_stride + 1
213
    int slice_type;
214
    int slice_type_fixed;
215
    
216
    //interlacing specific flags
217
    int mb_field_decoding_flag;
218
    
219
    int sub_mb_type[4];
220
    
221
    //POC stuff
222
    int poc_lsb;
223
    int poc_msb;
224
    int delta_poc_bottom;
225
    int delta_poc[2];
226
    int frame_num;
227
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
228
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
229
    int frame_num_offset;         ///< for POC type 2
230
    int prev_frame_num_offset;    ///< for POC type 2
231
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
232

    
233
    /**
234
     * frame_num for frames or 2*frame_num for field pics.
235
     */
236
    int curr_pic_num;
237
    
238
    /**
239
     * max_frame_num or 2*max_frame_num for field pics.
240
     */
241
    int max_pic_num;
242

    
243
    //Weighted pred stuff
244
    int luma_log2_weight_denom;
245
    int chroma_log2_weight_denom;
246
    int luma_weight[2][16];
247
    int luma_offset[2][16];
248
    int chroma_weight[2][16][2];
249
    int chroma_offset[2][16][2];
250
   
251
    //deblock
252
    int disable_deblocking_filter_idc;
253
    int slice_alpha_c0_offset_div2;
254
    int slice_beta_offset_div2;
255
     
256
    int redundant_pic_count;
257
    
258
    int direct_spatial_mv_pred;
259

    
260
    /**
261
     * num_ref_idx_l0/1_active_minus1 + 1
262
     */
263
    int ref_count[2];// FIXME split for AFF
264
    Picture *short_ref[16];
265
    Picture *long_ref[16];
266
    Picture default_ref_list[2][32];
267
    Picture ref_list[2][32]; //FIXME size?
268
    Picture field_ref_list[2][32]; //FIXME size?
269
    
270
    /**
271
     * memory management control operations buffer.
272
     */
273
    MMCO mmco[MAX_MMCO_COUNT];
274
    int mmco_index;
275
    
276
    int long_ref_count;  ///< number of actual long term references
277
    int short_ref_count; ///< number of actual short term references
278
    
279
    //data partitioning
280
    GetBitContext intra_gb;
281
    GetBitContext inter_gb;
282
    GetBitContext *intra_gb_ptr;
283
    GetBitContext *inter_gb_ptr;
284
    
285
    DCTELEM mb[16*24] __align8;
286
}H264Context;
287

    
288
static VLC coeff_token_vlc[4];
289
static VLC chroma_dc_coeff_token_vlc;
290

    
291
static VLC total_zeros_vlc[15];
292
static VLC chroma_dc_total_zeros_vlc[3];
293

    
294
static VLC run_vlc[6];
295
static VLC run7_vlc;
296

    
297
static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
298
static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
299

    
300
static inline uint32_t pack16to32(int a, int b){
301
#ifdef WORDS_BIGENDIAN
302
   return (b&0xFFFF) + (a<<16);
303
#else
304
   return (a&0xFFFF) + (b<<16);
305
#endif
306
}
307

    
308
/**
309
 * fill a rectangle.
310
 * @param h height of the recatangle, should be a constant
311
 * @param w width of the recatangle, should be a constant
312
 * @param size the size of val (1 or 4), should be a constant
313
 */
314
static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
315
    uint8_t *p= (uint8_t*)vp;
316
    assert(size==1 || size==4);
317
    
318
    w      *= size;
319
    stride *= size;
320
    
321
//FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
322
    if(w==2 && h==2){
323
        *(uint16_t*)(p + 0)=
324
        *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
325
    }else if(w==2 && h==4){
326
        *(uint16_t*)(p + 0*stride)=
327
        *(uint16_t*)(p + 1*stride)=
328
        *(uint16_t*)(p + 2*stride)=
329
        *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
330
    }else if(w==4 && h==1){
331
        *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
332
    }else if(w==4 && h==2){
333
        *(uint32_t*)(p + 0*stride)=
334
        *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
335
    }else if(w==4 && h==4){
336
        *(uint32_t*)(p + 0*stride)=
337
        *(uint32_t*)(p + 1*stride)=
338
        *(uint32_t*)(p + 2*stride)=
339
        *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
340
    }else if(w==8 && h==1){
341
        *(uint32_t*)(p + 0)=
342
        *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
343
    }else if(w==8 && h==2){
344
        *(uint32_t*)(p + 0 + 0*stride)=
345
        *(uint32_t*)(p + 4 + 0*stride)=
346
        *(uint32_t*)(p + 0 + 1*stride)=
347
        *(uint32_t*)(p + 4 + 1*stride)=  size==4 ? val : val*0x01010101;
348
    }else if(w==8 && h==4){
349
        *(uint64_t*)(p + 0*stride)=
350
        *(uint64_t*)(p + 1*stride)=
351
        *(uint64_t*)(p + 2*stride)=
352
        *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
353
    }else if(w==16 && h==2){
354
        *(uint64_t*)(p + 0+0*stride)=
355
        *(uint64_t*)(p + 8+0*stride)=
356
        *(uint64_t*)(p + 0+1*stride)=
357
        *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
358
    }else if(w==16 && h==4){
359
        *(uint64_t*)(p + 0+0*stride)=
360
        *(uint64_t*)(p + 8+0*stride)=
361
        *(uint64_t*)(p + 0+1*stride)=
362
        *(uint64_t*)(p + 8+1*stride)=
363
        *(uint64_t*)(p + 0+2*stride)=
364
        *(uint64_t*)(p + 8+2*stride)=
365
        *(uint64_t*)(p + 0+3*stride)=
366
        *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
367
    }else
368
        assert(0);
369
}
370

    
371
static inline void fill_caches(H264Context *h, int mb_type){
372
    MpegEncContext * const s = &h->s;
373
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
374
    int topleft_xy, top_xy, topright_xy, left_xy[2];
375
    int topleft_type, top_type, topright_type, left_type[2];
376
    int left_block[4];
377
    int i;
378

    
379
    //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it 
380
    
381
    if(h->sps.mb_aff){
382
    //FIXME
383
        topleft_xy = 0; /* avoid warning */
384
        top_xy = 0; /* avoid warning */
385
        topright_xy = 0; /* avoid warning */
386
    }else{
387
        topleft_xy = mb_xy-1 - s->mb_stride;
388
        top_xy     = mb_xy   - s->mb_stride;
389
        topright_xy= mb_xy+1 - s->mb_stride;
390
        left_xy[0]   = mb_xy-1;
391
        left_xy[1]   = mb_xy-1;
392
        left_block[0]= 0;
393
        left_block[1]= 1;
394
        left_block[2]= 2;
395
        left_block[3]= 3;
396
    }
397

    
398
    topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
399
    top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
400
    topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
401
    left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
402
    left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
403

    
404
    if(IS_INTRA(mb_type)){
405
        h->topleft_samples_available= 
406
        h->top_samples_available= 
407
        h->left_samples_available= 0xFFFF;
408
        h->topright_samples_available= 0xEEEA;
409

    
410
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
411
            h->topleft_samples_available= 0xB3FF;
412
            h->top_samples_available= 0x33FF;
413
            h->topright_samples_available= 0x26EA;
414
        }
415
        for(i=0; i<2; i++){
416
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
417
                h->topleft_samples_available&= 0xDF5F;
418
                h->left_samples_available&= 0x5F5F;
419
            }
420
        }
421
        
422
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
423
            h->topleft_samples_available&= 0x7FFF;
424
        
425
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
426
            h->topright_samples_available&= 0xFBFF;
427
    
428
        if(IS_INTRA4x4(mb_type)){
429
            if(IS_INTRA4x4(top_type)){
430
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
431
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
432
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
433
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
434
            }else{
435
                int pred;
436
                if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
437
                    pred= 2;
438
                else{
439
                    pred= -1;
440
                }
441
                h->intra4x4_pred_mode_cache[4+8*0]=
442
                h->intra4x4_pred_mode_cache[5+8*0]=
443
                h->intra4x4_pred_mode_cache[6+8*0]=
444
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
445
            }
446
            for(i=0; i<2; i++){
447
                if(IS_INTRA4x4(left_type[i])){
448
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
449
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
450
                }else{
451
                    int pred;
452
                    if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
453
                        pred= 2;
454
                    else{
455
                        pred= -1;
456
                    }
457
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
458
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
459
                }
460
            }
461
        }
462
    }
463
    
464
    
465
/*
466
0 . T T. T T T T 
467
1 L . .L . . . . 
468
2 L . .L . . . . 
469
3 . T TL . . . . 
470
4 L . .L . . . . 
471
5 L . .. . . . . 
472
*/
473
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
474
    if(top_type){
475
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
476
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
477
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
478
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
479
    
480
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
481
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
482
    
483
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
484
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
485
    }else{
486
        h->non_zero_count_cache[4+8*0]=      
487
        h->non_zero_count_cache[5+8*0]=
488
        h->non_zero_count_cache[6+8*0]=
489
        h->non_zero_count_cache[7+8*0]=
490
    
491
        h->non_zero_count_cache[1+8*0]=
492
        h->non_zero_count_cache[2+8*0]=
493
    
494
        h->non_zero_count_cache[1+8*3]=
495
        h->non_zero_count_cache[2+8*3]= 64;
496
    }
497
    
498
    if(left_type[0]){
499
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
500
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
501
        h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
502
        h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
503
    }else{
504
        h->non_zero_count_cache[3+8*1]= 
505
        h->non_zero_count_cache[3+8*2]= 
506
        h->non_zero_count_cache[0+8*1]= 
507
        h->non_zero_count_cache[0+8*4]= 64;
508
    }
509
    
510
    if(left_type[1]){
511
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
512
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
513
        h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
514
        h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
515
    }else{
516
        h->non_zero_count_cache[3+8*3]= 
517
        h->non_zero_count_cache[3+8*4]= 
518
        h->non_zero_count_cache[0+8*2]= 
519
        h->non_zero_count_cache[0+8*5]= 64;
520
    }
521
    
522
#if 1
523
    if(IS_INTER(mb_type)){
524
        int list;
525
        for(list=0; list<2; list++){
526
            if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
527
                /*if(!h->mv_cache_clean[list]){
528
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
529
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
530
                    h->mv_cache_clean[list]= 1;
531
                }*/
532
                continue; //FIXME direct mode ...
533
            }
534
            h->mv_cache_clean[list]= 0;
535
            
536
            if(IS_INTER(topleft_type)){
537
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
538
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
539
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
540
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
541
            }else{
542
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
543
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
544
            }
545
            
546
            if(IS_INTER(top_type)){
547
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
548
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
549
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
550
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
551
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
552
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
553
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
554
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
555
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
556
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
557
            }else{
558
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
559
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
560
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
561
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
562
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
563
            }
564

    
565
            if(IS_INTER(topright_type)){
566
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
567
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
568
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
569
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
570
            }else{
571
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
572
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
573
            }
574
            
575
            //FIXME unify cleanup or sth
576
            if(IS_INTER(left_type[0])){
577
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
578
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
579
                *(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]];
580
                *(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]];
581
                h->ref_cache[list][scan8[0] - 1 + 0*8]= 
582
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
583
            }else{
584
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
585
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
586
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
587
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
588
            }
589
            
590
            if(IS_INTER(left_type[1])){
591
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
592
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
593
                *(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]];
594
                *(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]];
595
                h->ref_cache[list][scan8[0] - 1 + 2*8]= 
596
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
597
            }else{
598
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
599
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
600
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
601
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
602
            }
603

    
604
            h->ref_cache[list][scan8[5 ]+1] = 
605
            h->ref_cache[list][scan8[7 ]+1] = 
606
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewher else)
607
            h->ref_cache[list][scan8[4 ]] = 
608
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
609
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
610
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
611
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
612
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
613
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
614
        }
615
//FIXME
616

    
617
    }
618
#endif
619
}
620

    
621
static inline void write_back_intra_pred_mode(H264Context *h){
622
    MpegEncContext * const s = &h->s;
623
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
624

    
625
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
626
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
627
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
628
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
629
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
630
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
631
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
632
}
633

    
634
/**
635
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
636
 */
637
static inline int check_intra4x4_pred_mode(H264Context *h){
638
    MpegEncContext * const s = &h->s;
639
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
640
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
641
    int i;
642
    
643
    if(!(h->top_samples_available&0x8000)){
644
        for(i=0; i<4; i++){
645
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
646
            if(status<0){
647
                fprintf(stderr, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
648
                return -1;
649
            } else if(status){
650
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
651
            }
652
        }
653
    }
654
    
655
    if(!(h->left_samples_available&0x8000)){
656
        for(i=0; i<4; i++){
657
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
658
            if(status<0){
659
                fprintf(stderr, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
660
                return -1;
661
            } else if(status){
662
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
663
            }
664
        }
665
    }
666

    
667
    return 0;
668
} //FIXME cleanup like next
669

    
670
/**
671
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
672
 */
673
static inline int check_intra_pred_mode(H264Context *h, int mode){
674
    MpegEncContext * const s = &h->s;
675
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
676
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
677
    
678
    if(!(h->top_samples_available&0x8000)){
679
        mode= top[ mode ];
680
        if(mode<0){
681
            fprintf(stderr, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
682
            return -1;
683
        }
684
    }
685
    
686
    if(!(h->left_samples_available&0x8000)){
687
        mode= left[ mode ];
688
        if(mode<0){
689
            fprintf(stderr, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
690
            return -1;
691
        } 
692
    }
693

    
694
    return mode;
695
}
696

    
697
/**
698
 * gets the predicted intra4x4 prediction mode.
699
 */
700
static inline int pred_intra_mode(H264Context *h, int n){
701
    const int index8= scan8[n];
702
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
703
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
704
    const int min= FFMIN(left, top);
705

    
706
    tprintf("mode:%d %d min:%d\n", left ,top, min);
707

    
708
    if(min<0) return DC_PRED;
709
    else      return min;
710
}
711

    
712
static inline void write_back_non_zero_count(H264Context *h){
713
    MpegEncContext * const s = &h->s;
714
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
715

    
716
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
717
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
718
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
719
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
720
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
721
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
722
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
723
    
724
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
725
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
726
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
727

    
728
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
729
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
730
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
731
}
732

    
733
/**
734
 * gets the predicted number of non zero coefficients.
735
 * @param n block index
736
 */
737
static inline int pred_non_zero_count(H264Context *h, int n){
738
    const int index8= scan8[n];
739
    const int left= h->non_zero_count_cache[index8 - 1];
740
    const int top = h->non_zero_count_cache[index8 - 8];
741
    int i= left + top;
742
    
743
    if(i<64) i= (i+1)>>1;
744

    
745
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
746

    
747
    return i&31;
748
}
749

    
750
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
751
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
752

    
753
    if(topright_ref != PART_NOT_AVAILABLE){
754
        *C= h->mv_cache[list][ i - 8 + part_width ];
755
        return topright_ref;
756
    }else{
757
        tprintf("topright MV not available\n");
758

    
759
        *C= h->mv_cache[list][ i - 8 - 1 ];
760
        return h->ref_cache[list][ i - 8 - 1 ];
761
    }
762
}
763

    
764
/**
765
 * gets the predicted MV.
766
 * @param n the block index
767
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
768
 * @param mx the x component of the predicted motion vector
769
 * @param my the y component of the predicted motion vector
770
 */
771
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
772
    const int index8= scan8[n];
773
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
774
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
775
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
776
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
777
    const int16_t * C;
778
    int diagonal_ref, match_count;
779

    
780
    assert(part_width==1 || part_width==2 || part_width==4);
781

    
782
/* mv_cache
783
  B . . A T T T T 
784
  U . . L . . , .
785
  U . . L . . . .
786
  U . . L . . , .
787
  . . . L . . . .
788
*/
789

    
790
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
791
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
792
    if(match_count > 1){ //most common
793
        *mx= mid_pred(A[0], B[0], C[0]);
794
        *my= mid_pred(A[1], B[1], C[1]);
795
    }else if(match_count==1){
796
        if(left_ref==ref){
797
            *mx= A[0];
798
            *my= A[1];        
799
        }else if(top_ref==ref){
800
            *mx= B[0];
801
            *my= B[1];        
802
        }else{
803
            *mx= C[0];
804
            *my= C[1];        
805
        }
806
    }else{
807
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
808
            *mx= A[0];
809
            *my= A[1];        
810
        }else{
811
            *mx= mid_pred(A[0], B[0], C[0]);
812
            *my= mid_pred(A[1], B[1], C[1]);
813
        }
814
    }
815
        
816
    tprintf("pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1],                    diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);
817
}
818

    
819
/**
820
 * gets the directionally predicted 16x8 MV.
821
 * @param n the block index
822
 * @param mx the x component of the predicted motion vector
823
 * @param my the y component of the predicted motion vector
824
 */
825
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
826
    if(n==0){
827
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
828
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
829

    
830
        tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
831
        
832
        if(top_ref == ref){
833
            *mx= B[0];
834
            *my= B[1];
835
            return;
836
        }
837
    }else{
838
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
839
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
840
        
841
        tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
842

    
843
        if(left_ref == ref){
844
            *mx= A[0];
845
            *my= A[1];
846
            return;
847
        }
848
    }
849

    
850
    //RARE
851
    pred_motion(h, n, 4, list, ref, mx, my);
852
}
853

    
854
/**
855
 * gets the directionally predicted 8x16 MV.
856
 * @param n the block index
857
 * @param mx the x component of the predicted motion vector
858
 * @param my the y component of the predicted motion vector
859
 */
860
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
861
    if(n==0){
862
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
863
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
864
        
865
        tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
866

    
867
        if(left_ref == ref){
868
            *mx= A[0];
869
            *my= A[1];
870
            return;
871
        }
872
    }else{
873
        const int16_t * C;
874
        int diagonal_ref;
875

    
876
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
877
        
878
        tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
879

    
880
        if(diagonal_ref == ref){ 
881
            *mx= C[0];
882
            *my= C[1];
883
            return;
884
        }
885
    }
886

    
887
    //RARE
888
    pred_motion(h, n, 2, list, ref, mx, my);
889
}
890

    
891
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
892
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
893
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
894

    
895
    tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
896

    
897
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
898
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
899
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
900
       
901
        *mx = *my = 0;
902
        return;
903
    }
904
        
905
    pred_motion(h, 0, 4, 0, 0, mx, my);
906

    
907
    return;
908
}
909

    
910
static inline void write_back_motion(H264Context *h, int mb_type){
911
    MpegEncContext * const s = &h->s;
912
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
913
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
914
    int list;
915

    
916
    for(list=0; list<2; list++){
917
        int y;
918
        if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
919
            if(1){ //FIXME skip or never read if mb_type doesnt use it
920
                for(y=0; y<4; y++){
921
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
922
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
923
                }
924
                for(y=0; y<2; y++){
925
                    *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
926
                }
927
            }
928
            continue; //FIXME direct mode ...
929
        }
930
        
931
        for(y=0; y<4; y++){
932
            *(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];
933
            *(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];
934
        }
935
        for(y=0; y<2; y++){
936
            s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
937
            s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
938
        }
939
    }
940
}
941

    
942
/**
943
 * Decodes a network abstraction layer unit.
944
 * @param consumed is the number of bytes used as input
945
 * @param length is the length of the array
946
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
947
 * @returns decoded bytes, might be src+1 if no escapes 
948
 */
949
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
950
    int i, si, di;
951
    uint8_t *dst;
952

    
953
//    src[0]&0x80;                //forbidden bit
954
    h->nal_ref_idc= src[0]>>5;
955
    h->nal_unit_type= src[0]&0x1F;
956

    
957
    src++; length--;
958
#if 0    
959
    for(i=0; i<length; i++)
960
        printf("%2X ", src[i]);
961
#endif
962
    for(i=0; i+1<length; i+=2){
963
        if(src[i]) continue;
964
        if(i>0 && src[i-1]==0) i--;
965
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
966
            if(src[i+2]!=3){
967
                /* startcode, so we must be past the end */
968
                length=i;
969
            }
970
            break;
971
        }
972
    }
973

    
974
    if(i>=length-1){ //no escaped 0
975
        *dst_length= length;
976
        *consumed= length+1; //+1 for the header
977
        return src; 
978
    }
979

    
980
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
981
    dst= h->rbsp_buffer;
982

    
983
//printf("deoding esc\n");
984
    si=di=0;
985
    while(si<length){ 
986
        //remove escapes (very rare 1:2^22)
987
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
988
            if(src[si+2]==3){ //escape
989
                dst[di++]= 0;
990
                dst[di++]= 0;
991
                si+=3;
992
            }else //next start code
993
                break;
994
        }
995

    
996
        dst[di++]= src[si++];
997
    }
998

    
999
    *dst_length= di;
1000
    *consumed= si + 1;//+1 for the header
1001
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1002
    return dst;
1003
}
1004

    
1005
/**
1006
 * @param src the data which should be escaped
1007
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1008
 * @param length the length of the src data
1009
 * @param dst_length the length of the dst array
1010
 * @returns length of escaped data in bytes or -1 if an error occured
1011
 */
1012
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1013
    int i, escape_count, si, di;
1014
    uint8_t *temp;
1015
    
1016
    assert(length>=0);
1017
    assert(dst_length>0);
1018
    
1019
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1020

    
1021
    if(length==0) return 1;
1022

    
1023
    escape_count= 0;
1024
    for(i=0; i<length; i+=2){
1025
        if(src[i]) continue;
1026
        if(i>0 && src[i-1]==0) 
1027
            i--;
1028
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1029
            escape_count++;
1030
            i+=2;
1031
        }
1032
    }
1033
    
1034
    if(escape_count==0){ 
1035
        if(dst+1 != src)
1036
            memcpy(dst+1, src, length);
1037
        return length + 1;
1038
    }
1039
    
1040
    if(length + escape_count + 1> dst_length)
1041
        return -1;
1042

    
1043
    //this should be damn rare (hopefully)
1044

    
1045
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1046
    temp= h->rbsp_buffer;
1047
//printf("encoding esc\n");
1048
    
1049
    si= 0;
1050
    di= 0;
1051
    while(si < length){
1052
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1053
            temp[di++]= 0; si++;
1054
            temp[di++]= 0; si++;
1055
            temp[di++]= 3; 
1056
            temp[di++]= src[si++];
1057
        }
1058
        else
1059
            temp[di++]= src[si++];
1060
    }
1061
    memcpy(dst+1, temp, length+escape_count);
1062
    
1063
    assert(di == length+escape_count);
1064
    
1065
    return di + 1;
1066
}
1067

    
1068
/**
1069
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1070
 */
1071
static void encode_rbsp_trailing(PutBitContext *pb){
1072
    int length;
1073
    put_bits(pb, 1, 1);
1074
    length= (-get_bit_count(pb))&7;
1075
    if(length) put_bits(pb, length, 0);
1076
}
1077

    
1078
/**
1079
 * identifies the exact end of the bitstream
1080
 * @return the length of the trailing, or 0 if damaged
1081
 */
1082
static int decode_rbsp_trailing(uint8_t *src){
1083
    int v= *src;
1084
    int r;
1085

    
1086
    tprintf("rbsp trailing %X\n", v);
1087

    
1088
    for(r=1; r<9; r++){
1089
        if(v&1) return r;
1090
        v>>=1;
1091
    }
1092
    return 0;
1093
}
1094

    
1095
/**
1096
 * idct tranforms the 16 dc values and dequantize them.
1097
 * @param qp quantization parameter
1098
 */
1099
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1100
    const int qmul= dequant_coeff[qp][0];
1101
#define stride 16
1102
    int i;
1103
    int temp[16]; //FIXME check if this is a good idea
1104
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1105
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1106

    
1107
//memset(block, 64, 2*256);
1108
//return;
1109
    for(i=0; i<4; i++){
1110
        const int offset= y_offset[i];
1111
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1112
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1113
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1114
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1115

    
1116
        temp[4*i+0]= z0+z3;
1117
        temp[4*i+1]= z1+z2;
1118
        temp[4*i+2]= z1-z2;
1119
        temp[4*i+3]= z0-z3;
1120
    }
1121

    
1122
    for(i=0; i<4; i++){
1123
        const int offset= x_offset[i];
1124
        const int z0= temp[4*0+i] + temp[4*2+i];
1125
        const int z1= temp[4*0+i] - temp[4*2+i];
1126
        const int z2= temp[4*1+i] - temp[4*3+i];
1127
        const int z3= temp[4*1+i] + temp[4*3+i];
1128

    
1129
        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1130
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1131
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1132
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1133
    }
1134
}
1135

    
1136
/**
1137
 * dct tranforms the 16 dc values.
1138
 * @param qp quantization parameter ??? FIXME
1139
 */
1140
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1141
//    const int qmul= dequant_coeff[qp][0];
1142
    int i;
1143
    int temp[16]; //FIXME check if this is a good idea
1144
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1145
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1146

    
1147
    for(i=0; i<4; i++){
1148
        const int offset= y_offset[i];
1149
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1150
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1151
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1152
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1153

    
1154
        temp[4*i+0]= z0+z3;
1155
        temp[4*i+1]= z1+z2;
1156
        temp[4*i+2]= z1-z2;
1157
        temp[4*i+3]= z0-z3;
1158
    }
1159

    
1160
    for(i=0; i<4; i++){
1161
        const int offset= x_offset[i];
1162
        const int z0= temp[4*0+i] + temp[4*2+i];
1163
        const int z1= temp[4*0+i] - temp[4*2+i];
1164
        const int z2= temp[4*1+i] - temp[4*3+i];
1165
        const int z3= temp[4*1+i] + temp[4*3+i];
1166

    
1167
        block[stride*0 +offset]= (z0 + z3)>>1;
1168
        block[stride*2 +offset]= (z1 + z2)>>1;
1169
        block[stride*8 +offset]= (z1 - z2)>>1;
1170
        block[stride*10+offset]= (z0 - z3)>>1;
1171
    }
1172
}
1173
#undef xStride
1174
#undef stride
1175

    
1176
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1177
    const int qmul= dequant_coeff[qp][0];
1178
    const int stride= 16*2;
1179
    const int xStride= 16;
1180
    int a,b,c,d,e;
1181

    
1182
    a= block[stride*0 + xStride*0];
1183
    b= block[stride*0 + xStride*1];
1184
    c= block[stride*1 + xStride*0];
1185
    d= block[stride*1 + xStride*1];
1186

    
1187
    e= a-b;
1188
    a= a+b;
1189
    b= c-d;
1190
    c= c+d;
1191

    
1192
    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1193
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1194
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1195
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1196
}
1197

    
1198
static void chroma_dc_dct_c(DCTELEM *block){
1199
    const int stride= 16*2;
1200
    const int xStride= 16;
1201
    int a,b,c,d,e;
1202

    
1203
    a= block[stride*0 + xStride*0];
1204
    b= block[stride*0 + xStride*1];
1205
    c= block[stride*1 + xStride*0];
1206
    d= block[stride*1 + xStride*1];
1207

    
1208
    e= a-b;
1209
    a= a+b;
1210
    b= c-d;
1211
    c= c+d;
1212

    
1213
    block[stride*0 + xStride*0]= (a+c);
1214
    block[stride*0 + xStride*1]= (e+b);
1215
    block[stride*1 + xStride*0]= (a-c);
1216
    block[stride*1 + xStride*1]= (e-b);
1217
}
1218

    
1219
/**
1220
 * gets the chroma qp.
1221
 */
1222
static inline int get_chroma_qp(H264Context *h, int qscale){
1223
    
1224
    return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1225
}
1226

    
1227

    
1228
/**
1229
 *
1230
 */
1231
static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1232
    int i;
1233
    uint8_t *cm = cropTbl + MAX_NEG_CROP;
1234

    
1235
    block[0] += 32;
1236
#if 1
1237
    for(i=0; i<4; i++){
1238
        const int z0=  block[i + 4*0]     +  block[i + 4*2];
1239
        const int z1=  block[i + 4*0]     -  block[i + 4*2];
1240
        const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
1241
        const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
1242

    
1243
        block[i + 4*0]= z0 + z3;
1244
        block[i + 4*1]= z1 + z2;
1245
        block[i + 4*2]= z1 - z2;
1246
        block[i + 4*3]= z0 - z3;
1247
    }
1248

    
1249
    for(i=0; i<4; i++){
1250
        const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
1251
        const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
1252
        const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
1253
        const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
1254

    
1255
        dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ];
1256
        dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ];
1257
        dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ];
1258
        dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ];
1259
    }
1260
#else
1261
    for(i=0; i<4; i++){
1262
        const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
1263
        const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
1264
        const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
1265
        const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
1266

    
1267
        block[0 + 4*i]= z0 + z3;
1268
        block[1 + 4*i]= z1 + z2;
1269
        block[2 + 4*i]= z1 - z2;
1270
        block[3 + 4*i]= z0 - z3;
1271
    }
1272

    
1273
    for(i=0; i<4; i++){
1274
        const int z0=  block[i + 4*0]     +  block[i + 4*2];
1275
        const int z1=  block[i + 4*0]     -  block[i + 4*2];
1276
        const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
1277
        const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
1278

    
1279
        dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1280
        dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1281
        dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1282
        dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1283
    }
1284
#endif
1285
}
1286

    
1287
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1288
    int i;
1289
    //FIXME try int temp instead of block
1290
    
1291
    for(i=0; i<4; i++){
1292
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1293
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1294
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1295
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1296
        const int z0= d0 + d3;
1297
        const int z3= d0 - d3;
1298
        const int z1= d1 + d2;
1299
        const int z2= d1 - d2;
1300
        
1301
        block[0 + 4*i]=   z0 +   z1;
1302
        block[1 + 4*i]= 2*z3 +   z2;
1303
        block[2 + 4*i]=   z0 -   z1;
1304
        block[3 + 4*i]=   z3 - 2*z2;
1305
    }    
1306

    
1307
    for(i=0; i<4; i++){
1308
        const int z0= block[0*4 + i] + block[3*4 + i];
1309
        const int z3= block[0*4 + i] - block[3*4 + i];
1310
        const int z1= block[1*4 + i] + block[2*4 + i];
1311
        const int z2= block[1*4 + i] - block[2*4 + i];
1312
        
1313
        block[0*4 + i]=   z0 +   z1;
1314
        block[1*4 + i]= 2*z3 +   z2;
1315
        block[2*4 + i]=   z0 -   z1;
1316
        block[3*4 + i]=   z3 - 2*z2;
1317
    }
1318
}
1319

    
1320
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1321
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1322
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1323
    int i;
1324
    const int * const quant_table= quant_coeff[qscale];
1325
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1326
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1327
    const unsigned int threshold2= (threshold1<<1);
1328
    int last_non_zero;
1329

    
1330
    if(seperate_dc){
1331
        if(qscale<=18){
1332
            //avoid overflows
1333
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1334
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1335
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1336

    
1337
            int level= block[0]*quant_coeff[qscale+18][0];
1338
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1339
                if(level>0){
1340
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1341
                    block[0]= level;
1342
                }else{
1343
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1344
                    block[0]= -level;
1345
                }
1346
//                last_non_zero = i;
1347
            }else{
1348
                block[0]=0;
1349
            }
1350
        }else{
1351
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1352
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1353
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1354

    
1355
            int level= block[0]*quant_table[0];
1356
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1357
                if(level>0){
1358
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1359
                    block[0]= level;
1360
                }else{
1361
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1362
                    block[0]= -level;
1363
                }
1364
//                last_non_zero = i;
1365
            }else{
1366
                block[0]=0;
1367
            }
1368
        }
1369
        last_non_zero= 0;
1370
        i=1;
1371
    }else{
1372
        last_non_zero= -1;
1373
        i=0;
1374
    }
1375

    
1376
    for(; i<16; i++){
1377
        const int j= scantable[i];
1378
        int level= block[j]*quant_table[j];
1379

    
1380
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1381
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1382
        if(((unsigned)(level+threshold1))>threshold2){
1383
            if(level>0){
1384
                level= (bias + level)>>QUANT_SHIFT;
1385
                block[j]= level;
1386
            }else{
1387
                level= (bias - level)>>QUANT_SHIFT;
1388
                block[j]= -level;
1389
            }
1390
            last_non_zero = i;
1391
        }else{
1392
            block[j]=0;
1393
        }
1394
    }
1395

    
1396
    return last_non_zero;
1397
}
1398

    
1399
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1400
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1401
    ((uint32_t*)(src+0*stride))[0]= a;
1402
    ((uint32_t*)(src+1*stride))[0]= a;
1403
    ((uint32_t*)(src+2*stride))[0]= a;
1404
    ((uint32_t*)(src+3*stride))[0]= a;
1405
}
1406

    
1407
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1408
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1409
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1410
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1411
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1412
}
1413

    
1414
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1415
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1416
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1417
    
1418
    ((uint32_t*)(src+0*stride))[0]= 
1419
    ((uint32_t*)(src+1*stride))[0]= 
1420
    ((uint32_t*)(src+2*stride))[0]= 
1421
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1422
}
1423

    
1424
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1425
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1426
    
1427
    ((uint32_t*)(src+0*stride))[0]= 
1428
    ((uint32_t*)(src+1*stride))[0]= 
1429
    ((uint32_t*)(src+2*stride))[0]= 
1430
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1431
}
1432

    
1433
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1434
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1435
    
1436
    ((uint32_t*)(src+0*stride))[0]= 
1437
    ((uint32_t*)(src+1*stride))[0]= 
1438
    ((uint32_t*)(src+2*stride))[0]= 
1439
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1440
}
1441

    
1442
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1443
    ((uint32_t*)(src+0*stride))[0]= 
1444
    ((uint32_t*)(src+1*stride))[0]= 
1445
    ((uint32_t*)(src+2*stride))[0]= 
1446
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1447
}
1448

    
1449

    
1450
#define LOAD_TOP_RIGHT_EDGE\
1451
    const int t4= topright[0];\
1452
    const int t5= topright[1];\
1453
    const int t6= topright[2];\
1454
    const int t7= topright[3];\
1455

    
1456
#define LOAD_LEFT_EDGE\
1457
    const int l0= src[-1+0*stride];\
1458
    const int l1= src[-1+1*stride];\
1459
    const int l2= src[-1+2*stride];\
1460
    const int l3= src[-1+3*stride];\
1461

    
1462
#define LOAD_TOP_EDGE\
1463
    const int t0= src[ 0-1*stride];\
1464
    const int t1= src[ 1-1*stride];\
1465
    const int t2= src[ 2-1*stride];\
1466
    const int t3= src[ 3-1*stride];\
1467

    
1468
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1469
    const int lt= src[-1-1*stride];
1470
    LOAD_TOP_EDGE
1471
    LOAD_LEFT_EDGE
1472

    
1473
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1474
    src[0+2*stride]=
1475
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1476
    src[0+1*stride]=
1477
    src[1+2*stride]=
1478
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1479
    src[0+0*stride]=
1480
    src[1+1*stride]=
1481
    src[2+2*stride]=
1482
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1483
    src[1+0*stride]=
1484
    src[2+1*stride]=
1485
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1486
    src[2+0*stride]=
1487
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1488
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1489
}
1490

    
1491
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1492
    LOAD_TOP_EDGE    
1493
    LOAD_TOP_RIGHT_EDGE    
1494
//    LOAD_LEFT_EDGE    
1495

    
1496
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1497
    src[1+0*stride]=
1498
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1499
    src[2+0*stride]=
1500
    src[1+1*stride]=
1501
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1502
    src[3+0*stride]=
1503
    src[2+1*stride]=
1504
    src[1+2*stride]=
1505
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1506
    src[3+1*stride]=
1507
    src[2+2*stride]=
1508
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1509
    src[3+2*stride]=
1510
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1511
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1512
}
1513

    
1514
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1515
    const int lt= src[-1-1*stride];
1516
    LOAD_TOP_EDGE    
1517
    LOAD_LEFT_EDGE    
1518
    const __attribute__((unused)) int unu= l3;
1519

    
1520
    src[0+0*stride]=
1521
    src[1+2*stride]=(lt + t0 + 1)>>1;
1522
    src[1+0*stride]=
1523
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1524
    src[2+0*stride]=
1525
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1526
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1527
    src[0+1*stride]=
1528
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1529
    src[1+1*stride]=
1530
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1531
    src[2+1*stride]=
1532
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1533
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1534
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1535
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1536
}
1537

    
1538
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1539
    LOAD_TOP_EDGE    
1540
    LOAD_TOP_RIGHT_EDGE    
1541
    const __attribute__((unused)) int unu= t7;
1542

    
1543
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1544
    src[1+0*stride]=
1545
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1546
    src[2+0*stride]=
1547
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1548
    src[3+0*stride]=
1549
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1550
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1551
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1552
    src[1+1*stride]=
1553
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1554
    src[2+1*stride]=
1555
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1556
    src[3+1*stride]=
1557
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1558
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1559
}
1560

    
1561
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1562
    LOAD_LEFT_EDGE    
1563

    
1564
    src[0+0*stride]=(l0 + l1 + 1)>>1;
1565
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1566
    src[2+0*stride]=
1567
    src[0+1*stride]=(l1 + l2 + 1)>>1;
1568
    src[3+0*stride]=
1569
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1570
    src[2+1*stride]=
1571
    src[0+2*stride]=(l2 + l3 + 1)>>1;
1572
    src[3+1*stride]=
1573
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1574
    src[3+2*stride]=
1575
    src[1+3*stride]=
1576
    src[0+3*stride]=
1577
    src[2+2*stride]=
1578
    src[2+3*stride]=
1579
    src[3+3*stride]=l3;
1580
}
1581
    
1582
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1583
    const int lt= src[-1-1*stride];
1584
    LOAD_TOP_EDGE    
1585
    LOAD_LEFT_EDGE    
1586
    const __attribute__((unused)) int unu= t3;
1587

    
1588
    src[0+0*stride]=
1589
    src[2+1*stride]=(lt + l0 + 1)>>1;
1590
    src[1+0*stride]=
1591
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1592
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1593
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1594
    src[0+1*stride]=
1595
    src[2+2*stride]=(l0 + l1 + 1)>>1;
1596
    src[1+1*stride]=
1597
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1598
    src[0+2*stride]=
1599
    src[2+3*stride]=(l1 + l2+ 1)>>1;
1600
    src[1+2*stride]=
1601
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1602
    src[0+3*stride]=(l2 + l3 + 1)>>1;
1603
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1604
}
1605

    
1606
static void pred16x16_vertical_c(uint8_t *src, int stride){
1607
    int i;
1608
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1609
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1610
    const uint32_t c= ((uint32_t*)(src-stride))[2];
1611
    const uint32_t d= ((uint32_t*)(src-stride))[3];
1612
    
1613
    for(i=0; i<16; i++){
1614
        ((uint32_t*)(src+i*stride))[0]= a;
1615
        ((uint32_t*)(src+i*stride))[1]= b;
1616
        ((uint32_t*)(src+i*stride))[2]= c;
1617
        ((uint32_t*)(src+i*stride))[3]= d;
1618
    }
1619
}
1620

    
1621
static void pred16x16_horizontal_c(uint8_t *src, int stride){
1622
    int i;
1623

    
1624
    for(i=0; i<16; i++){
1625
        ((uint32_t*)(src+i*stride))[0]=
1626
        ((uint32_t*)(src+i*stride))[1]=
1627
        ((uint32_t*)(src+i*stride))[2]=
1628
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1629
    }
1630
}
1631

    
1632
static void pred16x16_dc_c(uint8_t *src, int stride){
1633
    int i, dc=0;
1634

    
1635
    for(i=0;i<16; i++){
1636
        dc+= src[-1+i*stride];
1637
    }
1638
    
1639
    for(i=0;i<16; i++){
1640
        dc+= src[i-stride];
1641
    }
1642

    
1643
    dc= 0x01010101*((dc + 16)>>5);
1644

    
1645
    for(i=0; i<16; i++){
1646
        ((uint32_t*)(src+i*stride))[0]=
1647
        ((uint32_t*)(src+i*stride))[1]=
1648
        ((uint32_t*)(src+i*stride))[2]=
1649
        ((uint32_t*)(src+i*stride))[3]= dc;
1650
    }
1651
}
1652

    
1653
static void pred16x16_left_dc_c(uint8_t *src, int stride){
1654
    int i, dc=0;
1655

    
1656
    for(i=0;i<16; i++){
1657
        dc+= src[-1+i*stride];
1658
    }
1659
    
1660
    dc= 0x01010101*((dc + 8)>>4);
1661

    
1662
    for(i=0; i<16; i++){
1663
        ((uint32_t*)(src+i*stride))[0]=
1664
        ((uint32_t*)(src+i*stride))[1]=
1665
        ((uint32_t*)(src+i*stride))[2]=
1666
        ((uint32_t*)(src+i*stride))[3]= dc;
1667
    }
1668
}
1669

    
1670
static void pred16x16_top_dc_c(uint8_t *src, int stride){
1671
    int i, dc=0;
1672

    
1673
    for(i=0;i<16; i++){
1674
        dc+= src[i-stride];
1675
    }
1676
    dc= 0x01010101*((dc + 8)>>4);
1677

    
1678
    for(i=0; i<16; i++){
1679
        ((uint32_t*)(src+i*stride))[0]=
1680
        ((uint32_t*)(src+i*stride))[1]=
1681
        ((uint32_t*)(src+i*stride))[2]=
1682
        ((uint32_t*)(src+i*stride))[3]= dc;
1683
    }
1684
}
1685

    
1686
static void pred16x16_128_dc_c(uint8_t *src, int stride){
1687
    int i;
1688

    
1689
    for(i=0; i<16; i++){
1690
        ((uint32_t*)(src+i*stride))[0]=
1691
        ((uint32_t*)(src+i*stride))[1]=
1692
        ((uint32_t*)(src+i*stride))[2]=
1693
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1694
    }
1695
}
1696

    
1697
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1698
  int i, j, k;
1699
  int a;
1700
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
1701
  const uint8_t * const src0 = src+7-stride;
1702
  const uint8_t *src1 = src+8*stride-1;
1703
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
1704
  int H = src0[1] - src0[-1];
1705
  int V = src1[0] - src2[ 0];
1706
  for(k=2; k<=8; ++k) {
1707
    src1 += stride; src2 -= stride;
1708
    H += k*(src0[k] - src0[-k]);
1709
    V += k*(src1[0] - src2[ 0]);
1710
  }
1711
  if(svq3){
1712
    H = ( 5*(H/4) ) / 16;
1713
    V = ( 5*(V/4) ) / 16;
1714

    
1715
    /* required for 100% accuracy */
1716
    i = H; H = V; V = i;
1717
  }else{
1718
    H = ( 5*H+32 ) >> 6;
1719
    V = ( 5*V+32 ) >> 6;
1720
  }
1721

    
1722
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1723
  for(j=16; j>0; --j) {
1724
    int b = a;
1725
    a += V;
1726
    for(i=-16; i<0; i+=4) {
1727
      src[16+i] = cm[ (b    ) >> 5 ];
1728
      src[17+i] = cm[ (b+  H) >> 5 ];
1729
      src[18+i] = cm[ (b+2*H) >> 5 ];
1730
      src[19+i] = cm[ (b+3*H) >> 5 ];
1731
      b += 4*H;
1732
    }
1733
    src += stride;
1734
  }
1735
}
1736

    
1737
static void pred16x16_plane_c(uint8_t *src, int stride){
1738
    pred16x16_plane_compat_c(src, stride, 0);
1739
}
1740

    
1741
static void pred8x8_vertical_c(uint8_t *src, int stride){
1742
    int i;
1743
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1744
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1745
    
1746
    for(i=0; i<8; i++){
1747
        ((uint32_t*)(src+i*stride))[0]= a;
1748
        ((uint32_t*)(src+i*stride))[1]= b;
1749
    }
1750
}
1751

    
1752
static void pred8x8_horizontal_c(uint8_t *src, int stride){
1753
    int i;
1754

    
1755
    for(i=0; i<8; i++){
1756
        ((uint32_t*)(src+i*stride))[0]=
1757
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1758
    }
1759
}
1760

    
1761
static void pred8x8_128_dc_c(uint8_t *src, int stride){
1762
    int i;
1763

    
1764
    for(i=0; i<4; i++){
1765
        ((uint32_t*)(src+i*stride))[0]= 
1766
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1767
    }
1768
    for(i=4; i<8; i++){
1769
        ((uint32_t*)(src+i*stride))[0]= 
1770
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1771
    }
1772
}
1773

    
1774
static void pred8x8_left_dc_c(uint8_t *src, int stride){
1775
    int i;
1776
    int dc0, dc2;
1777

    
1778
    dc0=dc2=0;
1779
    for(i=0;i<4; i++){
1780
        dc0+= src[-1+i*stride];
1781
        dc2+= src[-1+(i+4)*stride];
1782
    }
1783
    dc0= 0x01010101*((dc0 + 2)>>2);
1784
    dc2= 0x01010101*((dc2 + 2)>>2);
1785

    
1786
    for(i=0; i<4; i++){
1787
        ((uint32_t*)(src+i*stride))[0]=
1788
        ((uint32_t*)(src+i*stride))[1]= dc0;
1789
    }
1790
    for(i=4; i<8; i++){
1791
        ((uint32_t*)(src+i*stride))[0]=
1792
        ((uint32_t*)(src+i*stride))[1]= dc2;
1793
    }
1794
}
1795

    
1796
static void pred8x8_top_dc_c(uint8_t *src, int stride){
1797
    int i;
1798
    int dc0, dc1;
1799

    
1800
    dc0=dc1=0;
1801
    for(i=0;i<4; i++){
1802
        dc0+= src[i-stride];
1803
        dc1+= src[4+i-stride];
1804
    }
1805
    dc0= 0x01010101*((dc0 + 2)>>2);
1806
    dc1= 0x01010101*((dc1 + 2)>>2);
1807

    
1808
    for(i=0; i<4; i++){
1809
        ((uint32_t*)(src+i*stride))[0]= dc0;
1810
        ((uint32_t*)(src+i*stride))[1]= dc1;
1811
    }
1812
    for(i=4; i<8; i++){
1813
        ((uint32_t*)(src+i*stride))[0]= dc0;
1814
        ((uint32_t*)(src+i*stride))[1]= dc1;
1815
    }
1816
}
1817

    
1818

    
1819
static void pred8x8_dc_c(uint8_t *src, int stride){
1820
    int i;
1821
    int dc0, dc1, dc2, dc3;
1822

    
1823
    dc0=dc1=dc2=0;
1824
    for(i=0;i<4; i++){
1825
        dc0+= src[-1+i*stride] + src[i-stride];
1826
        dc1+= src[4+i-stride];
1827
        dc2+= src[-1+(i+4)*stride];
1828
    }
1829
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1830
    dc0= 0x01010101*((dc0 + 4)>>3);
1831
    dc1= 0x01010101*((dc1 + 2)>>2);
1832
    dc2= 0x01010101*((dc2 + 2)>>2);
1833

    
1834
    for(i=0; i<4; i++){
1835
        ((uint32_t*)(src+i*stride))[0]= dc0;
1836
        ((uint32_t*)(src+i*stride))[1]= dc1;
1837
    }
1838
    for(i=4; i<8; i++){
1839
        ((uint32_t*)(src+i*stride))[0]= dc2;
1840
        ((uint32_t*)(src+i*stride))[1]= dc3;
1841
    }
1842
}
1843

    
1844
static void pred8x8_plane_c(uint8_t *src, int stride){
1845
  int j, k;
1846
  int a;
1847
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
1848
  const uint8_t * const src0 = src+3-stride;
1849
  const uint8_t *src1 = src+4*stride-1;
1850
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
1851
  int H = src0[1] - src0[-1];
1852
  int V = src1[0] - src2[ 0];
1853
  for(k=2; k<=4; ++k) {
1854
    src1 += stride; src2 -= stride;
1855
    H += k*(src0[k] - src0[-k]);
1856
    V += k*(src1[0] - src2[ 0]);
1857
  }
1858
  H = ( 17*H+16 ) >> 5;
1859
  V = ( 17*V+16 ) >> 5;
1860

    
1861
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1862
  for(j=8; j>0; --j) {
1863
    int b = a;
1864
    a += V;
1865
    src[0] = cm[ (b    ) >> 5 ];
1866
    src[1] = cm[ (b+  H) >> 5 ];
1867
    src[2] = cm[ (b+2*H) >> 5 ];
1868
    src[3] = cm[ (b+3*H) >> 5 ];
1869
    src[4] = cm[ (b+4*H) >> 5 ];
1870
    src[5] = cm[ (b+5*H) >> 5 ];
1871
    src[6] = cm[ (b+6*H) >> 5 ];
1872
    src[7] = cm[ (b+7*H) >> 5 ];
1873
    src += stride;
1874
  }
1875
}
1876

    
1877
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1878
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1879
                           int src_x_offset, int src_y_offset,
1880
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1881
    MpegEncContext * const s = &h->s;
1882
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1883
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1884
    const int luma_xy= (mx&3) + ((my&3)<<2);
1885
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1886
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1887
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1888
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1889
    int extra_height= extra_width;
1890
    int emu=0;
1891
    const int full_mx= mx>>2;
1892
    const int full_my= my>>2;
1893
    
1894
    assert(pic->data[0]);
1895
    
1896
    if(mx&7) extra_width -= 3;
1897
    if(my&7) extra_height -= 3;
1898
    
1899
    if(   full_mx < 0-extra_width 
1900
       || full_my < 0-extra_height 
1901
       || full_mx + 16/*FIXME*/ > s->width + extra_width 
1902
       || full_my + 16/*FIXME*/ > s->height + extra_height){
1903
        ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, s->width, s->height);
1904
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1905
        emu=1;
1906
    }
1907
    
1908
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1909
    if(!square){
1910
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1911
    }
1912
    
1913
    if(s->flags&CODEC_FLAG_GRAY) return;
1914
    
1915
    if(emu){
1916
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
1917
            src_cb= s->edge_emu_buffer;
1918
    }
1919
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1920

    
1921
    if(emu){
1922
        ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
1923
            src_cr= s->edge_emu_buffer;
1924
    }
1925
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1926
}
1927

    
1928
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1929
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1930
                           int x_offset, int y_offset,
1931
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1932
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1933
                           int list0, int list1){
1934
    MpegEncContext * const s = &h->s;
1935
    qpel_mc_func *qpix_op=  qpix_put;
1936
    h264_chroma_mc_func chroma_op= chroma_put;
1937
    
1938
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
1939
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
1940
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
1941
    x_offset += 8*s->mb_x;
1942
    y_offset += 8*s->mb_y;
1943
    
1944
    if(list0){
1945
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1946
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1947
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1948
                           qpix_op, chroma_op);
1949

    
1950
        qpix_op=  qpix_avg;
1951
        chroma_op= chroma_avg;
1952
    }
1953

    
1954
    if(list1){
1955
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1956
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1957
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1958
                           qpix_op, chroma_op);
1959
    }
1960
}
1961

    
1962
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1963
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1964
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
1965
    MpegEncContext * const s = &h->s;
1966
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1967
    const int mb_type= s->current_picture.mb_type[mb_xy];
1968
    
1969
    assert(IS_INTER(mb_type));
1970
    
1971
    if(IS_16X16(mb_type)){
1972
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1973
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1974
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1975
    }else if(IS_16X8(mb_type)){
1976
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1977
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1978
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1979
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1980
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1981
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1982
    }else if(IS_8X16(mb_type)){
1983
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
1984
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1985
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1986
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
1987
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1988
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1989
    }else{
1990
        int i;
1991
        
1992
        assert(IS_8X8(mb_type));
1993

    
1994
        for(i=0; i<4; i++){
1995
            const int sub_mb_type= h->sub_mb_type[i];
1996
            const int n= 4*i;
1997
            int x_offset= (i&1)<<2;
1998
            int y_offset= (i&2)<<1;
1999

    
2000
            if(IS_SUB_8X8(sub_mb_type)){
2001
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2002
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2003
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2004
            }else if(IS_SUB_8X4(sub_mb_type)){
2005
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2006
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2007
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2008
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2009
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2010
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2011
            }else if(IS_SUB_4X8(sub_mb_type)){
2012
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2013
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2014
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2015
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2016
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2017
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2018
            }else{
2019
                int j;
2020
                assert(IS_SUB_4X4(sub_mb_type));
2021
                for(j=0; j<4; j++){
2022
                    int sub_x_offset= x_offset + 2*(j&1);
2023
                    int sub_y_offset= y_offset +   (j&2);
2024
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2025
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2026
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2027
                }
2028
            }
2029
        }
2030
    }
2031
}
2032

    
2033
static void decode_init_vlc(H264Context *h){
2034
    static int done = 0;
2035

    
2036
    if (!done) {
2037
        int i;
2038
        done = 1;
2039

    
2040
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2041
                 &chroma_dc_coeff_token_len [0], 1, 1,
2042
                 &chroma_dc_coeff_token_bits[0], 1, 1);
2043

    
2044
        for(i=0; i<4; i++){
2045
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2046
                     &coeff_token_len [i][0], 1, 1,
2047
                     &coeff_token_bits[i][0], 1, 1);
2048
        }
2049

    
2050
        for(i=0; i<3; i++){
2051
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2052
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2053
                     &chroma_dc_total_zeros_bits[i][0], 1, 1);
2054
        }
2055
        for(i=0; i<15; i++){
2056
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2057
                     &total_zeros_len [i][0], 1, 1,
2058
                     &total_zeros_bits[i][0], 1, 1);
2059
        }
2060

    
2061
        for(i=0; i<6; i++){
2062
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2063
                     &run_len [i][0], 1, 1,
2064
                     &run_bits[i][0], 1, 1);
2065
        }
2066
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2067
                 &run_len [6][0], 1, 1,
2068
                 &run_bits[6][0], 1, 1);
2069
    }
2070
}
2071

    
2072
/**
2073
 * Sets the intra prediction function pointers.
2074
 */
2075
static void init_pred_ptrs(H264Context *h){
2076
//    MpegEncContext * const s = &h->s;
2077

    
2078
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2079
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2080
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2081
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2082
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2083
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2084
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2085
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2086
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2087
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2088
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2089
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2090

    
2091
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2092
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2093
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2094
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2095
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2096
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2097
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2098

    
2099
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2100
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2101
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2102
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2103
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2104
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2105
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2106
}
2107

    
2108
static void free_tables(H264Context *h){
2109
    av_freep(&h->intra4x4_pred_mode);
2110
    av_freep(&h->non_zero_count);
2111
    av_freep(&h->slice_table_base);
2112
    h->slice_table= NULL;
2113
    
2114
    av_freep(&h->mb2b_xy);
2115
    av_freep(&h->mb2b8_xy);
2116
}
2117

    
2118
/**
2119
 * allocates tables.
2120
 * needs widzh/height
2121
 */
2122
static int alloc_tables(H264Context *h){
2123
    MpegEncContext * const s = &h->s;
2124
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2125
    int x,y;
2126

    
2127
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2128
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2129
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2130

    
2131
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2132
    h->slice_table= h->slice_table_base + s->mb_stride + 1;
2133

    
2134
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint16_t));
2135
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2136
    for(y=0; y<s->mb_height; y++){
2137
        for(x=0; x<s->mb_width; x++){
2138
            const int mb_xy= x + y*s->mb_stride;
2139
            const int b_xy = 4*x + 4*y*h->b_stride;
2140
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2141
        
2142
            h->mb2b_xy [mb_xy]= b_xy;
2143
            h->mb2b8_xy[mb_xy]= b8_xy;
2144
        }
2145
    }
2146
    
2147
    return 0;
2148
fail:
2149
    free_tables(h);
2150
    return -1;
2151
}
2152

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    
2262
                    h->pred4x4[ dir ](ptr, topright, linesize);
2263
                    if(h->non_zero_count_cache[ scan8[i] ]){
2264
                        if(s->codec_id == CODEC_ID_H264)
2265
                            h264_add_idct_c(ptr, h->mb + i*16, linesize);
2266
                        else
2267
                            svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2268
                    }
2269
                }
2270
            }
2271
        }else{
2272
            h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2273
            if(s->codec_id == CODEC_ID_H264)
2274
                h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2275
            else
2276
                svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2277
        }
2278
    }else if(s->codec_id == CODEC_ID_H264){
2279
        hl_motion(h, dest_y, dest_cb, dest_cr,
2280
                  s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
2281
                  s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2282
    }
2283

    
2284

    
2285
    if(!IS_INTRA4x4(mb_type)){
2286
        if(s->codec_id == CODEC_ID_H264){
2287
            for(i=0; i<16; i++){
2288
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2289
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2290
                    h264_add_idct_c(ptr, h->mb + i*16, linesize);
2291
                }
2292
            }
2293
        }else{
2294
            for(i=0; i<16; i++){
2295
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2296
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2297
                    svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2298
                }
2299
            }
2300
        }
2301
    }
2302

    
2303
    if(!(s->flags&CODEC_FLAG_GRAY)){
2304
        chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2305
        chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2306
        if(s->codec_id == CODEC_ID_H264){
2307
            for(i=16; i<16+4; i++){
2308
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2309
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2310
                    h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2311
                }
2312
            }
2313
            for(i=20; i<20+4; i++){
2314
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2315
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2316
                    h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2317
                }
2318
            }
2319
        }else{
2320
            for(i=16; i<16+4; i++){
2321
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2322
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2323
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2324
                }
2325
            }
2326
            for(i=20; i<20+4; i++){
2327
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2328
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2329
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2330
                }
2331
            }
2332
        }
2333
    }
2334
}
2335

    
2336
static void decode_mb_cabac(H264Context *h){
2337
//    MpegEncContext * const s = &h->s;
2338
}
2339

    
2340
/**
2341
 * fills the default_ref_list.
2342
 */
2343
static int fill_default_ref_list(H264Context *h){
2344
    MpegEncContext * const s = &h->s;
2345
    int i;
2346
    Picture sorted_short_ref[16];
2347
    
2348
    if(h->slice_type==B_TYPE){
2349
        int out_i;
2350
        int limit= -1;
2351

    
2352
        for(out_i=0; out_i<h->short_ref_count; out_i++){
2353
            int best_i=-1;
2354
            int best_poc=-1;
2355

    
2356
            for(i=0; i<h->short_ref_count; i++){
2357
                const int poc= h->short_ref[i]->poc;
2358
                if(poc > limit && poc < best_poc){
2359
                    best_poc= poc;
2360
                    best_i= i;
2361
                }
2362
            }
2363
            
2364
            assert(best_i != -1);
2365
            
2366
            limit= best_poc;
2367
            sorted_short_ref[out_i]= *h->short_ref[best_i];
2368
        }
2369
    }
2370

    
2371
    if(s->picture_structure == PICT_FRAME){
2372
        if(h->slice_type==B_TYPE){
2373
            const int current_poc= s->current_picture_ptr->poc;
2374
            int list;
2375

    
2376
            for(list=0; list<2; list++){
2377
                int index=0;
2378

    
2379
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2380
                    const int i2= list ? h->short_ref_count - i - 1 : i;
2381
                    const int poc= sorted_short_ref[i2].poc;
2382
                    
2383
                    if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2384

    
2385
                    if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2386
                        h->default_ref_list[list][index  ]= sorted_short_ref[i2];
2387
                        h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2388
                    }
2389
                }
2390

    
2391
                for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2392
                    if(h->long_ref[i]->reference != 3) continue;
2393

    
2394
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
2395
                    h->default_ref_list[ list ][index++].pic_id= i;;
2396
                }
2397
                
2398
                if(h->long_ref_count > 1 && h->short_ref_count==0){
2399
                    Picture temp= h->default_ref_list[1][0];
2400
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
2401
                    h->default_ref_list[1][0] = temp;
2402
                }
2403

    
2404
                if(index < h->ref_count[ list ])
2405
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2406
            }
2407
        }else{
2408
            int index=0;
2409
            for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2410
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2411
                h->default_ref_list[0][index  ]= *h->short_ref[i];
2412
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2413
            }
2414
            for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2415
                if(h->long_ref[i]->reference != 3) continue;
2416
                h->default_ref_list[0][index  ]= *h->long_ref[i];
2417
                h->default_ref_list[0][index++].pic_id= i;;
2418
            }
2419
            if(index < h->ref_count[0])
2420
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2421
        }
2422
    }else{ //FIELD
2423
        if(h->slice_type==B_TYPE){
2424
        }else{
2425
            //FIXME second field balh
2426
        }
2427
    }
2428
    return 0;
2429
}
2430

    
2431
static int decode_ref_pic_list_reordering(H264Context *h){
2432
    MpegEncContext * const s = &h->s;
2433
    int list;
2434
    
2435
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2436
    
2437
    for(list=0; list<2; list++){
2438
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2439

    
2440
        if(get_bits1(&s->gb)){
2441
            int pred= h->curr_pic_num;
2442
            int index;
2443

    
2444
            for(index=0; ; index++){
2445
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2446
                int pic_id;
2447
                int i;
2448
                
2449
                
2450
                if(index >= h->ref_count[list]){
2451
                    fprintf(stderr, "reference count overflow\n");
2452
                    return -1;
2453
                }
2454
                
2455
                if(reordering_of_pic_nums_idc<3){
2456
                    if(reordering_of_pic_nums_idc<2){
2457
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2458

    
2459
                        if(abs_diff_pic_num >= h->max_pic_num){
2460
                            fprintf(stderr, "abs_diff_pic_num overflow\n");
2461
                            return -1;
2462
                        }
2463

    
2464
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2465
                        else                                pred+= abs_diff_pic_num;
2466
                        pred &= h->max_pic_num - 1;
2467
                    
2468
                        for(i= h->ref_count[list]-1; i>=index; i--){
2469
                            if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2470
                                break;
2471
                        }
2472
                    }else{
2473
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2474

    
2475
                        for(i= h->ref_count[list]-1; i>=index; i--){
2476
                            if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2477
                                break;
2478
                        }
2479
                    }
2480

    
2481
                    if(i < index){
2482
                        fprintf(stderr, "reference picture missing during reorder\n");
2483
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2484
                    }else if(i > index){
2485
                        Picture tmp= h->ref_list[list][i];
2486
                        for(; i>index; i--){
2487
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2488
                        }
2489
                        h->ref_list[list][index]= tmp;
2490
                    }
2491
                }else if(reordering_of_pic_nums_idc==3) 
2492
                    break;
2493
                else{
2494
                    fprintf(stderr, "illegal reordering_of_pic_nums_idc\n");
2495
                    return -1;
2496
                }
2497
            }
2498
        }
2499

    
2500
        if(h->slice_type!=B_TYPE) break;
2501
    }
2502
    return 0;    
2503
}
2504

    
2505
static int pred_weight_table(H264Context *h){
2506
    MpegEncContext * const s = &h->s;
2507
    int list, i;
2508
    
2509
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2510
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2511

    
2512
    for(list=0; list<2; list++){
2513
        for(i=0; i<h->ref_count[list]; i++){
2514
            int luma_weight_flag, chroma_weight_flag;
2515
            
2516
            luma_weight_flag= get_bits1(&s->gb);
2517
            if(luma_weight_flag){
2518
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
2519
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
2520
            }
2521

    
2522
            chroma_weight_flag= get_bits1(&s->gb);
2523
            if(chroma_weight_flag){
2524
                int j;
2525
                for(j=0; j<2; j++){
2526
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2527
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2528
                }
2529
            }
2530
        }
2531
        if(h->slice_type != B_TYPE) break;
2532
    }
2533
    return 0;
2534
}
2535

    
2536
/**
2537
 * instantaneos decoder refresh.
2538
 */
2539
static void idr(H264Context *h){
2540
    int i;
2541

    
2542
    for(i=0; i<h->long_ref_count; i++){
2543
        h->long_ref[i]->reference=0;
2544
        h->long_ref[i]= NULL;
2545
    }
2546
    h->long_ref_count=0;
2547

    
2548
    for(i=0; i<h->short_ref_count; i++){
2549
        h->short_ref[i]->reference=0;
2550
        h->short_ref[i]= NULL;
2551
    }
2552
    h->short_ref_count=0;
2553
}
2554

    
2555
/**
2556
 *
2557
 * @return the removed picture or NULL if an error occures
2558
 */
2559
static Picture * remove_short(H264Context *h, int frame_num){
2560
    MpegEncContext * const s = &h->s;
2561
    int i;
2562
    
2563
    if(s->avctx->debug&FF_DEBUG_MMCO)
2564
        printf("remove short %d count %d\n", frame_num, h->short_ref_count);
2565
    
2566
    for(i=0; i<h->short_ref_count; i++){
2567
        Picture *pic= h->short_ref[i];
2568
        if(s->avctx->debug&FF_DEBUG_MMCO)
2569
            printf("%d %d %p\n", i, pic->frame_num, pic);
2570
        if(pic->frame_num == frame_num){
2571
            h->short_ref[i]= NULL;
2572
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2573
            h->short_ref_count--;
2574
            return pic;
2575
        }
2576
    }
2577
    return NULL;
2578
}
2579

    
2580
/**
2581
 *
2582
 * @return the removed picture or NULL if an error occures
2583
 */
2584
static Picture * remove_long(H264Context *h, int i){
2585
    Picture *pic;
2586

    
2587
    if(i >= h->long_ref_count) return NULL;
2588
    pic= h->long_ref[i];
2589
    if(pic==NULL) return NULL;
2590
    
2591
    h->long_ref[i]= NULL;
2592
    memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2593
    h->long_ref_count--;
2594

    
2595
    return pic;
2596
}
2597

    
2598
/**
2599
 * Executes the reference picture marking (memory management control operations).
2600
 */
2601
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2602
    MpegEncContext * const s = &h->s;
2603
    int i;
2604
    int current_is_long=0;
2605
    Picture *pic;
2606
    
2607
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2608
        printf("no mmco here\n");
2609
        
2610
    for(i=0; i<mmco_count; i++){
2611
        if(s->avctx->debug&FF_DEBUG_MMCO)
2612
            printf("mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
2613

    
2614
        switch(mmco[i].opcode){
2615
        case MMCO_SHORT2UNUSED:
2616
            pic= remove_short(h, mmco[i].short_frame_num);
2617
            if(pic==NULL) return -1;
2618
            pic->reference= 0;
2619
            break;
2620
        case MMCO_SHORT2LONG:
2621
            pic= remove_long(h, mmco[i].long_index);
2622
            if(pic) pic->reference=0;
2623
            
2624
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2625
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
2626
            break;
2627
        case MMCO_LONG2UNUSED:
2628
            pic= remove_long(h, mmco[i].long_index);
2629
            if(pic==NULL) return -1;
2630
            pic->reference= 0;
2631
            break;
2632
        case MMCO_LONG:
2633
            pic= remove_long(h, mmco[i].long_index);
2634
            if(pic) pic->reference=0;
2635
            
2636
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2637
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
2638
            h->long_ref_count++;
2639
            
2640
            current_is_long=1;
2641
            break;
2642
        case MMCO_SET_MAX_LONG:
2643
            assert(mmco[i].long_index <= 16);
2644
            while(mmco[i].long_index < h->long_ref_count){
2645
                pic= remove_long(h, mmco[i].long_index);
2646
                pic->reference=0;
2647
            }
2648
            while(mmco[i].long_index > h->long_ref_count){
2649
                h->long_ref[ h->long_ref_count++ ]= NULL;
2650
            }
2651
            break;
2652
        case MMCO_RESET:
2653
            while(h->short_ref_count){
2654
                pic= remove_short(h, h->short_ref[0]->frame_num);
2655
                pic->reference=0;
2656
            }
2657
            while(h->long_ref_count){
2658
                pic= remove_long(h, h->long_ref_count-1);
2659
                pic->reference=0;
2660
            }
2661
            break;
2662
        default: assert(0);
2663
        }
2664
    }
2665
    
2666
    if(!current_is_long){
2667
        pic= remove_short(h, s->current_picture_ptr->frame_num);
2668
        if(pic){
2669
            pic->reference=0;
2670
            fprintf(stderr, "illegal short term buffer state detected\n");
2671
        }
2672
        
2673
        if(h->short_ref_count)
2674
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2675

    
2676
        h->short_ref[0]= s->current_picture_ptr;
2677
        h->short_ref[0]->long_ref=0;
2678
        h->short_ref_count++;
2679
    }
2680
    
2681
    return 0; 
2682
}
2683

    
2684
static int decode_ref_pic_marking(H264Context *h){
2685
    MpegEncContext * const s = &h->s;
2686
    int i;
2687
    
2688
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2689
        s->broken_link= get_bits1(&s->gb) -1;
2690
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2691
        if(h->mmco[0].long_index == -1)
2692
            h->mmco_index= 0;
2693
        else{
2694
            h->mmco[0].opcode= MMCO_LONG;
2695
            h->mmco_index= 1;
2696
        } 
2697
    }else{
2698
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2699
            for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) { 
2700
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2701

    
2702
                h->mmco[i].opcode= opcode;
2703
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2704
                    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
2705
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2706
                        fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2707
                        return -1;
2708
                    }*/
2709
                }
2710
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2711
                    h->mmco[i].long_index= get_ue_golomb(&s->gb);
2712
                    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){
2713
                        fprintf(stderr, "illegal long ref in memory management control operation %d\n", opcode);
2714
                        return -1;
2715
                    }
2716
                }
2717
                    
2718
                if(opcode > MMCO_LONG){
2719
                    fprintf(stderr, "illegal memory management control operation %d\n", opcode);
2720
                    return -1;
2721
                }
2722
            }
2723
            h->mmco_index= i;
2724
        }else{
2725
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2726

    
2727
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2728
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2729
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2730
                h->mmco_index= 1;
2731
            }else
2732
                h->mmco_index= 0;
2733
        }
2734
    }
2735
    
2736
    return 0; 
2737
}
2738

    
2739
static int init_poc(H264Context *h){
2740
    MpegEncContext * const s = &h->s;
2741
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2742
    int field_poc[2];
2743

    
2744
    if(h->nal_unit_type == NAL_IDR_SLICE){
2745
        h->frame_num_offset= 0;
2746
    }else{
2747
        if(h->frame_num < h->prev_frame_num)
2748
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2749
        else
2750
            h->frame_num_offset= h->prev_frame_num_offset;
2751
    }
2752

    
2753
    if(h->sps.poc_type==0){
2754
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2755

    
2756
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2757
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2758
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2759
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2760
        else
2761
            h->poc_msb = h->prev_poc_msb;
2762
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2763
        field_poc[0] = 
2764
        field_poc[1] = h->poc_msb + h->poc_lsb;
2765
        if(s->picture_structure == PICT_FRAME) 
2766
            field_poc[1] += h->delta_poc_bottom;
2767
    }else if(h->sps.poc_type==1){
2768
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2769
        int i;
2770

    
2771
        if(h->sps.poc_cycle_length != 0)
2772
            abs_frame_num = h->frame_num_offset + h->frame_num;
2773
        else
2774
            abs_frame_num = 0;
2775

    
2776
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
2777
            abs_frame_num--;
2778
            
2779
        expected_delta_per_poc_cycle = 0;
2780
        for(i=0; i < h->sps.poc_cycle_length; i++)
2781
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2782

    
2783
        if(abs_frame_num > 0){
2784
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2785
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2786

    
2787
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2788
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
2789
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2790
        } else
2791
            expectedpoc = 0;
2792

    
2793
        if(h->nal_ref_idc == 0) 
2794
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2795
        
2796
        field_poc[0] = expectedpoc + h->delta_poc[0];
2797
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2798

    
2799
        if(s->picture_structure == PICT_FRAME)
2800
            field_poc[1] += h->delta_poc[1];
2801
    }else{
2802
        int poc;
2803
        if(h->nal_unit_type == NAL_IDR_SLICE){
2804
            poc= 0;
2805
        }else{
2806
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2807
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2808
        }
2809
        field_poc[0]= poc;
2810
        field_poc[1]= poc;
2811
    }
2812
    
2813
    if(s->picture_structure != PICT_BOTTOM_FIELD)
2814
        s->current_picture_ptr->field_poc[0]= field_poc[0];
2815
    if(s->picture_structure != PICT_TOP_FIELD)
2816
        s->current_picture_ptr->field_poc[1]= field_poc[1];
2817
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2818
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2819

    
2820
    return 0;
2821
}
2822

    
2823
/**
2824
 * decodes a slice header.
2825
 * this will allso call MPV_common_init() and frame_start() as needed
2826
 */
2827
static int decode_slice_header(H264Context *h){
2828
    MpegEncContext * const s = &h->s;
2829
    int first_mb_in_slice, pps_id;
2830
    int num_ref_idx_active_override_flag;
2831
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2832
    float new_aspect;
2833

    
2834
    s->current_picture.reference= h->nal_ref_idc != 0;
2835

    
2836
    first_mb_in_slice= get_ue_golomb(&s->gb);
2837

    
2838
    h->slice_type= get_ue_golomb(&s->gb);
2839
    if(h->slice_type > 9){
2840
        fprintf(stderr, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
2841
    }
2842
    if(h->slice_type > 4){
2843
        h->slice_type -= 5;
2844
        h->slice_type_fixed=1;
2845
    }else
2846
        h->slice_type_fixed=0;
2847
    
2848
    h->slice_type= slice_type_map[ h->slice_type ];
2849
    
2850
    s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
2851
        
2852
    pps_id= get_ue_golomb(&s->gb);
2853
    if(pps_id>255){
2854
        fprintf(stderr, "pps_id out of range\n");
2855
        return -1;
2856
    }
2857
    h->pps= h->pps_buffer[pps_id];
2858
    if(h->pps.slice_group_count == 0){
2859
        fprintf(stderr, "non existing PPS referenced\n");
2860
        return -1;
2861
    }
2862

    
2863
    h->sps= h->sps_buffer[ h->pps.sps_id ];
2864
    if(h->sps.log2_max_frame_num == 0){
2865
        fprintf(stderr, "non existing SPS referenced\n");
2866
        return -1;
2867
    }
2868
    
2869
    s->mb_width= h->sps.mb_width;
2870
    s->mb_height= h->sps.mb_height;
2871
    
2872
    h->b_stride=  s->mb_width*4;
2873
    h->b8_stride= s->mb_width*2;
2874

    
2875
    s->mb_x = first_mb_in_slice % s->mb_width;
2876
    s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
2877
    
2878
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
2879
    if(h->sps.frame_mbs_only_flag)
2880
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
2881
    else
2882
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
2883
    
2884
    if(s->aspected_height) //FIXME emms at end of slice ?
2885
        new_aspect= h->sps.sar_width*s->width / (float)(s->height*h->sps.sar_height);
2886
    else
2887
        new_aspect=0;
2888

    
2889
    if (s->context_initialized 
2890
        && (   s->width != s->avctx->width || s->height != s->avctx->height 
2891
            || ABS(new_aspect - s->avctx->aspect_ratio) > 0.001)) {
2892
        free_tables(h);
2893
        MPV_common_end(s);
2894
    }
2895
    if (!s->context_initialized) {
2896
        if (MPV_common_init(s) < 0)
2897
            return -1;
2898

    
2899
        alloc_tables(h);
2900

    
2901
        s->avctx->width = s->width;
2902
        s->avctx->height = s->height;
2903
        s->avctx->aspect_ratio= new_aspect;
2904
    }
2905

    
2906
    if(first_mb_in_slice == 0){
2907
        frame_start(h);
2908
    }
2909

    
2910
    s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
2911
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2912

    
2913
    if(h->sps.frame_mbs_only_flag){
2914
        s->picture_structure= PICT_FRAME;
2915
    }else{
2916
        if(get_bits1(&s->gb)) //field_pic_flag
2917
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2918
        else
2919
            s->picture_structure= PICT_FRAME;
2920
    }
2921

    
2922
    if(s->picture_structure==PICT_FRAME){
2923
        h->curr_pic_num=   h->frame_num;
2924
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2925
    }else{
2926
        h->curr_pic_num= 2*h->frame_num;
2927
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2928
    }
2929
        
2930
    if(h->nal_unit_type == NAL_IDR_SLICE){
2931
        get_ue_golomb(&s->gb); /* idr_pic_id */
2932
    }
2933
   
2934
    if(h->sps.poc_type==0){
2935
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2936
        
2937
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2938
            h->delta_poc_bottom= get_se_golomb(&s->gb);
2939
        }
2940
    }
2941
    
2942
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2943
        h->delta_poc[0]= get_se_golomb(&s->gb);
2944
        
2945
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2946
            h->delta_poc[1]= get_se_golomb(&s->gb);
2947
    }
2948
    
2949
    init_poc(h);
2950
    
2951
    if(h->pps.redundant_pic_cnt_present){
2952
        h->redundant_pic_count= get_ue_golomb(&s->gb);
2953
    }
2954

    
2955
    //set defaults, might be overriden a few line later
2956
    h->ref_count[0]= h->pps.ref_count[0];
2957
    h->ref_count[1]= h->pps.ref_count[1];
2958

    
2959
    if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
2960
        if(h->slice_type == B_TYPE){
2961
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
2962
        }
2963
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
2964
    
2965
        if(num_ref_idx_active_override_flag){
2966
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2967
            if(h->slice_type==B_TYPE)
2968
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2969

    
2970
            if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
2971
                fprintf(stderr, "reference overflow\n");
2972
                return -1;
2973
            }
2974
        }
2975
    }
2976

    
2977
    if(first_mb_in_slice == 0){
2978
        fill_default_ref_list(h);
2979
    }
2980

    
2981
    decode_ref_pic_list_reordering(h);
2982

    
2983
    if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) 
2984
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
2985
        pred_weight_table(h);
2986
    
2987
    if(s->current_picture.reference)
2988
        decode_ref_pic_marking(h);
2989
    //FIXME CABAC stuff
2990

    
2991
    s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta
2992
    //FIXME qscale / qp ... stuff
2993
    if(h->slice_type == SP_TYPE){
2994
        get_bits1(&s->gb); /* sp_for_switch_flag */
2995
    }
2996
    if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
2997
        get_se_golomb(&s->gb); /* slice_qs_delta */
2998
    }
2999

    
3000
    if( h->pps.deblocking_filter_parameters_present ) {
3001
        h->disable_deblocking_filter_idc= get_ue_golomb(&s->gb);
3002
        if( h->disable_deblocking_filter_idc  !=  1 ) {
3003
            h->slice_alpha_c0_offset_div2= get_se_golomb(&s->gb);
3004
            h->slice_beta_offset_div2= get_se_golomb(&s->gb);
3005
        }
3006
    }else
3007
        h->disable_deblocking_filter_idc= 0;
3008

    
3009
#if 0 //FMO
3010
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3011
        slice_group_change_cycle= get_bits(&s->gb, ?);
3012
#endif
3013

    
3014
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3015
        printf("mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n", 
3016
               first_mb_in_slice, 
3017
               av_get_pict_type_char(h->slice_type),
3018
               pps_id, h->frame_num,
3019
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3020
               h->ref_count[0], h->ref_count[1],
3021
               s->qscale,
3022
               h->disable_deblocking_filter_idc
3023
               );
3024
    }
3025

    
3026
    return 0;
3027
}
3028

    
3029
/**
3030
 *
3031
 */
3032
static inline int get_level_prefix(GetBitContext *gb){
3033
    unsigned int buf;
3034
    int log;
3035
    
3036
    OPEN_READER(re, gb);
3037
    UPDATE_CACHE(re, gb);
3038
    buf=GET_CACHE(re, gb);
3039
    
3040
    log= 32 - av_log2(buf);
3041
#ifdef TRACE
3042
    print_bin(buf>>(32-log), log);
3043
    printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3044
#endif
3045

    
3046
    LAST_SKIP_BITS(re, gb, log);
3047
    CLOSE_READER(re, gb);
3048

    
3049
    return log-1;
3050
}
3051

    
3052
/**
3053
 * decodes a residual block.
3054
 * @param n block index
3055
 * @param scantable scantable
3056
 * @param max_coeff number of coefficients in the block
3057
 * @return <0 if an error occured
3058
 */
3059
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3060
    MpegEncContext * const s = &h->s;
3061
    const uint16_t *qmul= dequant_coeff[qp];
3062
    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};
3063
    int level[16], run[16];
3064
    int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3065

    
3066
    //FIXME put trailing_onex into the context
3067

    
3068
    if(n == CHROMA_DC_BLOCK_INDEX){
3069
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3070
        total_coeff= coeff_token>>2;
3071
    }else{    
3072
        if(n == LUMA_DC_BLOCK_INDEX){
3073
            total_coeff= pred_non_zero_count(h, 0);
3074
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3075
            total_coeff= coeff_token>>2;
3076
        }else{
3077
            total_coeff= pred_non_zero_count(h, n);
3078
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3079
            total_coeff= coeff_token>>2;
3080
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3081
        }
3082
    }
3083

    
3084
    //FIXME set last_non_zero?
3085

    
3086
    if(total_coeff==0)
3087
        return 0;
3088
        
3089
    trailing_ones= coeff_token&3;
3090
    tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3091
    assert(total_coeff<=16);
3092
    
3093
    for(i=0; i<trailing_ones; i++){
3094
        level[i]= 1 - 2*get_bits1(gb);
3095
    }
3096

    
3097
    suffix_length= total_coeff > 10 && trailing_ones < 3;
3098

    
3099
    for(; i<total_coeff; i++){
3100
        const int prefix= get_level_prefix(gb);
3101
        int level_code, mask;
3102

    
3103
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3104
            if(suffix_length)
3105
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3106
            else
3107
                level_code= (prefix<<suffix_length); //part
3108
        }else if(prefix==14){
3109
            if(suffix_length)
3110
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3111
            else
3112
                level_code= prefix + get_bits(gb, 4); //part
3113
        }else if(prefix==15){
3114
            level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3115
            if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3116
        }else{
3117
            fprintf(stderr, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3118
            return -1;
3119
        }
3120

    
3121
        if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3122

    
3123
        mask= -(level_code&1);
3124
        level[i]= (((2+level_code)>>1) ^ mask) - mask;
3125

    
3126
        if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3127

    
3128
#if 1
3129
        if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3130
#else        
3131
        if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3132
        ? == prefix > 2 or sth
3133
#endif
3134
        tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3135
    }
3136

    
3137
    if(total_coeff == max_coeff)
3138
        zeros_left=0;
3139
    else{
3140
        if(n == CHROMA_DC_BLOCK_INDEX)
3141
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3142
        else
3143
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3144
    }
3145
    
3146
    for(i=0; i<total_coeff-1; i++){
3147
        if(zeros_left <=0)
3148
            break;
3149
        else if(zeros_left < 7){
3150
            run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3151
        }else{
3152
            run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3153
        }
3154
        zeros_left -= run[i];
3155
    }
3156

    
3157
    if(zeros_left<0){
3158
        fprintf(stderr, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3159
        return -1;
3160
    }
3161
    
3162
    for(; i<total_coeff-1; i++){
3163
        run[i]= 0;
3164
    }
3165

    
3166
    run[i]= zeros_left;
3167

    
3168
    coeff_num=-1;
3169
    if(n > 24){
3170
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3171
            int j;
3172

    
3173
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3174
            j= scantable[ coeff_num ];
3175

    
3176
            block[j]= level[i];
3177
        }
3178
    }else{
3179
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into  rundecode?
3180
            int j;
3181

    
3182
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3183
            j= scantable[ coeff_num ];
3184

    
3185
            block[j]= level[i] * qmul[j];
3186
//            printf("%d %d  ", block[j], qmul[j]);
3187
        }
3188
    }
3189
    return 0;
3190
}
3191

    
3192
/**
3193
 * decodes a macroblock
3194
 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3195
 */
3196
static int decode_mb(H264Context *h){
3197
    MpegEncContext * const s = &h->s;
3198
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3199
    int mb_type, partition_count, cbp;
3200

    
3201
    s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?    
3202

    
3203
    tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3204
    cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3205
                down the code */
3206
    if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3207
        if(s->mb_skip_run==-1)
3208
            s->mb_skip_run= get_ue_golomb(&s->gb);
3209
        
3210
        if (s->mb_skip_run--) {
3211
            int mx, my;
3212
            /* skip mb */
3213
//FIXME b frame
3214
            mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3215

    
3216
            memset(h->non_zero_count[mb_xy], 0, 16);
3217
            memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3218

    
3219
            if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3220
                h->mb_field_decoding_flag= get_bits1(&s->gb);
3221
            }
3222

    
3223
            if(h->mb_field_decoding_flag)
3224
                mb_type|= MB_TYPE_INTERLACED;
3225
            
3226
            fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3227
            pred_pskip_motion(h, &mx, &my);
3228
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3229
            fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3230
            write_back_motion(h, mb_type);
3231

    
3232
            s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3233
            h->slice_table[ mb_xy ]= h->slice_num;
3234

    
3235
            h->prev_mb_skiped= 1;
3236
            return 0;
3237
        }
3238
    }
3239
    if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3240
        if((s->mb_y&1)==0)
3241
            h->mb_field_decoding_flag = get_bits1(&s->gb);
3242
    }else
3243
        h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3244
    
3245
    h->prev_mb_skiped= 0;
3246
    
3247
    mb_type= get_ue_golomb(&s->gb);
3248
    if(h->slice_type == B_TYPE){
3249
        if(mb_type < 23){
3250
            partition_count= b_mb_type_info[mb_type].partition_count;
3251
            mb_type=         b_mb_type_info[mb_type].type;
3252
        }else{
3253
            mb_type -= 23;
3254
            goto decode_intra_mb;
3255
        }
3256
    }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3257
        if(mb_type < 5){
3258
            partition_count= p_mb_type_info[mb_type].partition_count;
3259
            mb_type=         p_mb_type_info[mb_type].type;
3260
        }else{
3261
            mb_type -= 5;
3262
            goto decode_intra_mb;
3263
        }
3264
    }else{
3265
       assert(h->slice_type == I_TYPE);
3266
decode_intra_mb:
3267
        if(mb_type > 25){
3268
            fprintf(stderr, "mb_type %d in %c slice to large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);
3269
            return -1;
3270
        }
3271
        partition_count=0;
3272
        cbp= i_mb_type_info[mb_type].cbp;
3273
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3274
        mb_type= i_mb_type_info[mb_type].type;
3275
    }
3276

    
3277
    if(h->mb_field_decoding_flag)
3278
        mb_type |= MB_TYPE_INTERLACED;
3279

    
3280
    s->current_picture.mb_type[mb_xy]= mb_type;
3281
    h->slice_table[ mb_xy ]= h->slice_num;
3282
    
3283
    if(IS_INTRA_PCM(mb_type)){
3284
        const uint8_t *ptr;
3285
        int x, y;
3286
        
3287
        // we assume these blocks are very rare so we dont optimize it
3288
        align_get_bits(&s->gb);
3289
        
3290
        ptr= s->gb.buffer + get_bits_count(&s->gb);
3291
    
3292
        for(y=0; y<16; y++){
3293
            const int index= 4*(y&3) + 64*(y>>2);
3294
            for(x=0; x<16; x++){
3295
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3296
            }
3297
        }
3298
        for(y=0; y<8; y++){
3299
            const int index= 256 + 4*(y&3) + 32*(y>>2);
3300
            for(x=0; x<8; x++){
3301
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3302
            }
3303
        }
3304
        for(y=0; y<8; y++){
3305
            const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3306
            for(x=0; x<8; x++){
3307
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3308
            }
3309
        }
3310
    
3311
        skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3312
        
3313
        memset(h->non_zero_count[mb_xy], 16, 16);
3314
        
3315
        return 0;
3316
    }
3317
        
3318
    fill_caches(h, mb_type);
3319

    
3320
    //mb_pred
3321
    if(IS_INTRA(mb_type)){
3322
//            init_top_left_availability(h);
3323
            if(IS_INTRA4x4(mb_type)){
3324
                int i;
3325

    
3326
//                fill_intra4x4_pred_table(h);
3327
                for(i=0; i<16; i++){
3328
                    const int mode_coded= !get_bits1(&s->gb);
3329
                    const int predicted_mode=  pred_intra_mode(h, i);
3330
                    int mode;
3331

    
3332
                    if(mode_coded){
3333
                        const int rem_mode= get_bits(&s->gb, 3);
3334
                        if(rem_mode<predicted_mode)
3335
                            mode= rem_mode;
3336
                        else
3337
                            mode= rem_mode + 1;
3338
                    }else{
3339
                        mode= predicted_mode;
3340
                    }
3341
                    
3342
                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3343
                }
3344
                write_back_intra_pred_mode(h);
3345
                if( check_intra4x4_pred_mode(h) < 0)
3346
                    return -1;
3347
            }else{
3348
                h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3349
                if(h->intra16x16_pred_mode < 0)
3350
                    return -1;
3351
            }
3352
            h->chroma_pred_mode= get_ue_golomb(&s->gb);
3353

    
3354
            h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3355
            if(h->chroma_pred_mode < 0)
3356
                return -1;
3357
    }else if(partition_count==4){
3358
        int i, j, sub_partition_count[4], list, ref[2][4];
3359
        
3360
        if(h->slice_type == B_TYPE){
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] >=13){
3364
                    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);
3365
                    return -1;
3366
                }
3367
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3368
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3369
            }
3370
        }else{
3371
            assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3372
            for(i=0; i<4; i++){
3373
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3374
                if(h->sub_mb_type[i] >=4){
3375
                    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);
3376
                    return -1;
3377
                }
3378
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3379
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3380
            }
3381
        }
3382
        
3383
        for(list=0; list<2; list++){
3384
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3385
            if(ref_count == 0) continue;
3386
            for(i=0; i<4; i++){
3387
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3388
                    ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3389
                }else{
3390
                 //FIXME
3391
                    ref[list][i] = -1;
3392
                }
3393
            }
3394
        }
3395
        
3396
        for(list=0; list<2; list++){
3397
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3398
            if(ref_count == 0) continue;
3399

    
3400
            for(i=0; i<4; i++){
3401
                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3402
                h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3403

    
3404
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3405
                    const int sub_mb_type= h->sub_mb_type[i];
3406
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3407
                    for(j=0; j<sub_partition_count[i]; j++){
3408
                        int mx, my;
3409
                        const int index= 4*i + block_width*j;
3410
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3411
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3412
                        mx += get_se_golomb(&s->gb);
3413
                        my += get_se_golomb(&s->gb);
3414
                        tprintf("final mv:%d %d\n", mx, my);
3415

    
3416
                        if(IS_SUB_8X8(sub_mb_type)){
3417
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= 
3418
                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3419
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= 
3420
                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3421
                        }else if(IS_SUB_8X4(sub_mb_type)){
3422
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3423
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3424
                        }else if(IS_SUB_4X8(sub_mb_type)){
3425
                            mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3426
                            mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3427
                        }else{
3428
                            assert(IS_SUB_4X4(sub_mb_type));
3429
                            mv_cache[ 0 ][0]= mx;
3430
                            mv_cache[ 0 ][1]= my;
3431
                        }
3432
                    }
3433
                }else{
3434
                    uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3435
                    p[0] = p[1]=
3436
                    p[8] = p[9]= 0;
3437
                }
3438
            }
3439
        }
3440
    }else if(!IS_DIRECT(mb_type)){
3441
        int list, mx, my, i;
3442
         //FIXME we should set ref_idx_l? to 0 if we use that later ...
3443
        if(IS_16X16(mb_type)){
3444
            for(list=0; list<2; list++){
3445
                if(h->ref_count[0]>0){
3446
                    if(IS_DIR(mb_type, 0, list)){
3447
                        const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3448
                        fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3449
                    }
3450
                }
3451
            }
3452
            for(list=0; list<2; list++){
3453
                if(IS_DIR(mb_type, 0, list)){
3454
                    pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3455
                    mx += get_se_golomb(&s->gb);
3456
                    my += get_se_golomb(&s->gb);
3457
                    tprintf("final mv:%d %d\n", mx, my);
3458

    
3459
                    fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
3460
                }
3461
            }
3462
        }
3463
        else if(IS_16X8(mb_type)){
3464
            for(list=0; list<2; list++){
3465
                if(h->ref_count[list]>0){
3466
                    for(i=0; i<2; i++){
3467
                        if(IS_DIR(mb_type, i, list)){
3468
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3469
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3470
                        }
3471
                    }
3472
                }
3473
            }
3474
            for(list=0; list<2; list++){
3475
                for(i=0; i<2; i++){
3476
                    if(IS_DIR(mb_type, i, list)){
3477
                        pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3478
                        mx += get_se_golomb(&s->gb);
3479
                        my += get_se_golomb(&s->gb);
3480
                        tprintf("final mv:%d %d\n", mx, my);
3481

    
3482
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
3483
                    }
3484
                }
3485
            }
3486
        }else{
3487
            assert(IS_8X16(mb_type));
3488
            for(list=0; list<2; list++){
3489
                if(h->ref_count[list]>0){
3490
                    for(i=0; i<2; i++){
3491
                        if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3492
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3493
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3494
                        }
3495
                    }
3496
                }
3497
            }
3498
            for(list=0; list<2; list++){
3499
                for(i=0; i<2; i++){
3500
                    if(IS_DIR(mb_type, i, list)){
3501
                        pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3502
                        mx += get_se_golomb(&s->gb);
3503
                        my += get_se_golomb(&s->gb);
3504
                        tprintf("final mv:%d %d\n", mx, my);
3505

    
3506
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
3507
                    }
3508
                }
3509
            }
3510
        }
3511
    }
3512
    
3513
    if(IS_INTER(mb_type))
3514
        write_back_motion(h, mb_type);
3515
    
3516
    if(!IS_INTRA16x16(mb_type)){
3517
        cbp= get_ue_golomb(&s->gb);
3518
        if(cbp > 47){
3519
            fprintf(stderr, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3520
            return -1;
3521
        }
3522
        
3523
        if(IS_INTRA4x4(mb_type))
3524
            cbp= golomb_to_intra4x4_cbp[cbp];
3525
        else
3526
            cbp= golomb_to_inter_cbp[cbp];
3527
    }
3528

    
3529
    if(cbp || IS_INTRA16x16(mb_type)){
3530
        int i8x8, i4x4, chroma_idx;
3531
        int chroma_qp, dquant;
3532
        GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3533
        const uint8_t *scan, *dc_scan;
3534
        
3535
//        fill_non_zero_count_cache(h);
3536

    
3537
        if(IS_INTERLACED(mb_type)){
3538
            scan= field_scan;
3539
            dc_scan= luma_dc_field_scan;
3540
        }else{
3541
            scan= zigzag_scan;
3542
            dc_scan= luma_dc_zigzag_scan;
3543
        }
3544

    
3545
        dquant= get_se_golomb(&s->gb);
3546

    
3547
        if( dquant > 25 || dquant < -26 ){
3548
            fprintf(stderr, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3549
            return -1;
3550
        }
3551
        
3552
        s->qscale += dquant;
3553
        if(((unsigned)s->qscale) > 51){
3554
            if(s->qscale<0) s->qscale+= 52;
3555
            else            s->qscale-= 52;
3556
        }
3557
        
3558
        h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3559
        if(IS_INTRA16x16(mb_type)){
3560
            if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3561
                return -1; //FIXME continue if partotioned and other retirn -1 too
3562
            }
3563

    
3564
            assert((cbp&15) == 0 || (cbp&15) == 15);
3565

    
3566
            if(cbp&15){
3567
                for(i8x8=0; i8x8<4; i8x8++){
3568
                    for(i4x4=0; i4x4<4; i4x4++){
3569
                        const int index= i4x4 + 4*i8x8;
3570
                        if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3571
                            return -1;
3572
                        }
3573
                    }
3574
                }
3575
            }else{
3576
                memset(&h->non_zero_count_cache[8], 0, 8*4); //FIXME stupid & slow
3577
            }
3578
        }else{
3579
            for(i8x8=0; i8x8<4; i8x8++){
3580
                if(cbp & (1<<i8x8)){
3581
                    for(i4x4=0; i4x4<4; i4x4++){
3582
                        const int index= i4x4 + 4*i8x8;
3583
                        
3584
                        if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3585
                            return -1;
3586
                        }
3587
                    }
3588
                }else{
3589
                    uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3590
                    nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3591
                }
3592
            }
3593
        }
3594
        
3595
        if(cbp&0x30){
3596
            for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3597
                if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3598
                    return -1;
3599
                }
3600
        }
3601

    
3602
        if(cbp&0x20){
3603
            for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3604
                for(i4x4=0; i4x4<4; i4x4++){
3605
                    const int index= 16 + 4*chroma_idx + i4x4;
3606
                    if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3607
                        return -1;
3608
                    }
3609
                }
3610
            }
3611
        }else{
3612
            uint8_t * const nnz= &h->non_zero_count_cache[0];
3613
            nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3614
            nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3615
        }
3616
    }else{
3617
        memset(&h->non_zero_count_cache[8], 0, 8*5);
3618
    }
3619
    write_back_non_zero_count(h);
3620

    
3621
    return 0;
3622
}
3623

    
3624
static int decode_slice(H264Context *h){
3625
    MpegEncContext * const s = &h->s;
3626
    const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3627

    
3628
    s->mb_skip_run= -1;
3629
    
3630
#if 1
3631
    for(;;){
3632
        int ret= decode_mb(h);
3633
            
3634
        hl_decode_mb(h);
3635
        
3636
        if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
3637
            s->mb_y++;
3638
            ret= decode_mb(h);
3639
            
3640
            hl_decode_mb(h);
3641
            s->mb_y--;
3642
        }
3643

    
3644
        if(ret<0){
3645
            fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3646
            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);
3647

    
3648
            return -1;
3649
        }
3650
        
3651
        if(++s->mb_x >= s->mb_width){
3652
            s->mb_x=0;
3653
            ff_draw_horiz_band(s, 16*s->mb_y, 16);
3654
            if(++s->mb_y >= s->mb_height){
3655
                tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3656

    
3657
                if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3658
                    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);
3659

    
3660
                    return 0;
3661
                }else{
3662
                    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);
3663

    
3664
                    return -1;
3665
                }
3666
            }
3667
        }
3668
        
3669
        if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3670
            if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3671
                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);
3672

    
3673
                return 0;
3674
            }else{
3675
                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);
3676

    
3677
                return -1;
3678
            }
3679
        }
3680
    }
3681
#endif
3682
#if 0
3683
    for(;s->mb_y < s->mb_height; s->mb_y++){
3684
        for(;s->mb_x < s->mb_width; s->mb_x++){
3685
            int ret= decode_mb(h);
3686
            
3687
            hl_decode_mb(h);
3688

3689
            if(ret<0){
3690
                fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3691
                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);
3692

3693
                return -1;
3694
            }
3695
        
3696
            if(++s->mb_x >= s->mb_width){
3697
                s->mb_x=0;
3698
                if(++s->mb_y >= s->mb_height){
3699
                    if(get_bits_count(s->gb) == s->gb.size_in_bits){
3700
                        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);
3701

3702
                        return 0;
3703
                    }else{
3704
                        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);
3705

3706
                        return -1;
3707
                    }
3708
                }
3709
            }
3710
        
3711
            if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
3712
                if(get_bits_count(s->gb) == s->gb.size_in_bits){
3713
                    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);
3714

3715
                    return 0;
3716
                }else{
3717
                    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);
3718

3719
                    return -1;
3720
                }
3721
            }
3722
        }
3723
        s->mb_x=0;
3724
        ff_draw_horiz_band(s, 16*s->mb_y, 16);
3725
    }
3726
#endif
3727
    return -1; //not reached
3728
}
3729

    
3730
static inline int decode_vui_parameters(H264Context *h, SPS *sps){
3731
    MpegEncContext * const s = &h->s;
3732
    int aspect_ratio_info_present_flag, aspect_ratio_idc;
3733

    
3734
    aspect_ratio_info_present_flag= get_bits1(&s->gb);
3735
    
3736
    if( aspect_ratio_info_present_flag ) {
3737
        aspect_ratio_idc= get_bits(&s->gb, 8);
3738
        if( aspect_ratio_idc == EXTENDED_SAR ) {
3739
            sps->sar_width= get_bits(&s->gb, 16);
3740
            sps->sar_height= get_bits(&s->gb, 16);
3741
        }else if(aspect_ratio_idc < 16){
3742
            sps->sar_width=  pixel_aspect[aspect_ratio_idc][0];
3743
            sps->sar_height= pixel_aspect[aspect_ratio_idc][1];
3744
        }else{
3745
            fprintf(stderr, "illegal aspect ratio\n");
3746
            return -1;
3747
        }
3748
    }else{
3749
        sps->sar_width= 
3750
        sps->sar_height= 0;
3751
    }
3752
//            s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
3753
#if 0
3754
| overscan_info_present_flag                        |0  |u(1)    |
3755
| if( overscan_info_present_flag )                  |   |        |
3756
|  overscan_appropriate_flag                        |0  |u(1)    |
3757
| video_signal_type_present_flag                    |0  |u(1)    |
3758
| if( video_signal_type_present_flag ) {            |   |        |
3759
|  video_format                                     |0  |u(3)    |
3760
|  video_full_range_flag                            |0  |u(1)    |
3761
|  colour_description_present_flag                  |0  |u(1)    |
3762
|  if( colour_description_present_flag ) {          |   |        |
3763
|   colour_primaries                                |0  |u(8)    |
3764
|   transfer_characteristics                        |0  |u(8)    |
3765
|   matrix_coefficients                             |0  |u(8)    |
3766
|  }                                                |   |        |
3767
| }                                                 |   |        |
3768
| chroma_location_info_present_flag                 |0  |u(1)    |
3769
| if ( chroma_location_info_present_flag ) {        |   |        |
3770
|  chroma_sample_location_type_top_field            |0  |ue(v)   |
3771
|  chroma_sample_location_type_bottom_field         |0  |ue(v)   |
3772
| }                                                 |   |        |
3773
| timing_info_present_flag                          |0  |u(1)    |
3774
| if( timing_info_present_flag ) {                  |   |        |
3775
|  num_units_in_tick                                |0  |u(32)   |
3776
|  time_scale                                       |0  |u(32)   |
3777
|  fixed_frame_rate_flag                            |0  |u(1)    |
3778
| }                                                 |   |        |
3779
| nal_hrd_parameters_present_flag                   |0  |u(1)    |
3780
| if( nal_hrd_parameters_present_flag  = =  1)      |   |        |
3781
|  hrd_parameters( )                                |   |        |
3782
| vcl_hrd_parameters_present_flag                   |0  |u(1)    |
3783
| if( vcl_hrd_parameters_present_flag  = =  1)      |   |        |
3784
|  hrd_parameters( )                                |   |        |
3785
| if( ( nal_hrd_parameters_present_flag  = =  1  | ||   |        |
3786
|                                                   |   |        |
3787
|( vcl_hrd_parameters_present_flag  = =  1 ) )      |   |        |
3788
|  low_delay_hrd_flag                               |0  |u(1)    |
3789
| bitstream_restriction_flag                        |0  |u(1)    |
3790
| if( bitstream_restriction_flag ) {                |0  |u(1)    |
3791
|  motion_vectors_over_pic_boundaries_flag          |0  |u(1)    |
3792
|  max_bytes_per_pic_denom                          |0  |ue(v)   |
3793
|  max_bits_per_mb_denom                            |0  |ue(v)   |
3794
|  log2_max_mv_length_horizontal                    |0  |ue(v)   |
3795
|  log2_max_mv_length_vertical                      |0  |ue(v)   |
3796
|  num_reorder_frames                               |0  |ue(v)   |
3797
|  max_dec_frame_buffering                          |0  |ue(v)   |
3798
| }                                                 |   |        |
3799
|}                                                  |   |        |
3800
#endif
3801
    return 0;
3802
}
3803

    
3804
static inline int decode_seq_parameter_set(H264Context *h){
3805
    MpegEncContext * const s = &h->s;
3806
    int profile_idc, level_idc;
3807
    int sps_id, i;
3808
    SPS *sps;
3809
    
3810
    profile_idc= get_bits(&s->gb, 8);
3811
    get_bits1(&s->gb);   //constraint_set0_flag
3812
    get_bits1(&s->gb);   //constraint_set1_flag
3813
    get_bits1(&s->gb);   //constraint_set2_flag
3814
    get_bits(&s->gb, 5); // reserved
3815
    level_idc= get_bits(&s->gb, 8);
3816
    sps_id= get_ue_golomb(&s->gb);
3817
    
3818
    sps= &h->sps_buffer[ sps_id ];
3819
    sps->profile_idc= profile_idc;
3820
    sps->level_idc= level_idc;
3821
    
3822
    sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
3823
    sps->poc_type= get_ue_golomb(&s->gb);
3824
    
3825
    if(sps->poc_type == 0){ //FIXME #define
3826
        sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
3827
    } else if(sps->poc_type == 1){//FIXME #define
3828
        sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
3829
        sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
3830
        sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
3831
        sps->poc_cycle_length= get_ue_golomb(&s->gb);
3832
        
3833
        for(i=0; i<sps->poc_cycle_length; i++)
3834
            sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
3835
    }
3836
    if(sps->poc_type > 2){
3837
        fprintf(stderr, "illegal POC type %d\n", sps->poc_type);
3838
        return -1;
3839
    }
3840

    
3841
    sps->ref_frame_count= get_ue_golomb(&s->gb);
3842
    sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
3843
    sps->mb_width= get_ue_golomb(&s->gb) + 1;
3844
    sps->mb_height= get_ue_golomb(&s->gb) + 1;
3845
    sps->frame_mbs_only_flag= get_bits1(&s->gb);
3846
    if(!sps->frame_mbs_only_flag)
3847
        sps->mb_aff= get_bits1(&s->gb);
3848
    else
3849
        sps->mb_aff= 0;
3850

    
3851
    sps->direct_8x8_inference_flag= get_bits1(&s->gb);
3852

    
3853
    sps->crop= get_bits1(&s->gb);
3854
    if(sps->crop){
3855
        sps->crop_left  = get_ue_golomb(&s->gb);
3856
        sps->crop_right = get_ue_golomb(&s->gb);
3857
        sps->crop_top   = get_ue_golomb(&s->gb);
3858
        sps->crop_bottom= get_ue_golomb(&s->gb);
3859
        if(sps->crop_left || sps->crop_top){
3860
            fprintf(stderr, "insane croping not completly supported, this could look slightly wrong ...\n");
3861
        }
3862
    }else{
3863
        sps->crop_left  = 
3864
        sps->crop_right = 
3865
        sps->crop_top   = 
3866
        sps->crop_bottom= 0;
3867
    }
3868

    
3869
    sps->vui_parameters_present_flag= get_bits1(&s->gb);
3870
    if( sps->vui_parameters_present_flag )
3871
        decode_vui_parameters(h, sps);
3872
    
3873
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3874
        printf("sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s crop:%d/%d/%d/%d %s\n", 
3875
               sps_id, sps->profile_idc, sps->level_idc,
3876
               sps->poc_type,
3877
               sps->ref_frame_count,
3878
               sps->mb_width, sps->mb_height,
3879
               sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
3880
               sps->direct_8x8_inference_flag ? "8B8" : "",
3881
               sps->crop_left, sps->crop_right, 
3882
               sps->crop_top, sps->crop_bottom, 
3883
               sps->vui_parameters_present_flag ? "VUI" : ""
3884
               );
3885
    }
3886
    return 0;
3887
}
3888

    
3889
static inline int decode_picture_parameter_set(H264Context *h){
3890
    MpegEncContext * const s = &h->s;
3891
    int pps_id= get_ue_golomb(&s->gb);
3892
    PPS *pps= &h->pps_buffer[pps_id];
3893
    
3894
    pps->sps_id= get_ue_golomb(&s->gb);
3895
    pps->cabac= get_bits1(&s->gb);
3896
    pps->pic_order_present= get_bits1(&s->gb);
3897
    pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
3898
    if(pps->slice_group_count > 1 ){
3899
        pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
3900
fprintf(stderr, "FMO not supported\n");
3901
        switch(pps->mb_slice_group_map_type){
3902
        case 0:
3903
#if 0
3904
|   for( i = 0; i <= num_slice_groups_minus1; i++ ) |   |        |
3905
|    run_length[ i ]                                |1  |ue(v)   |
3906
#endif
3907
            break;
3908
        case 2:
3909
#if 0
3910
|   for( i = 0; i < num_slice_groups_minus1; i++ )  |   |        |
3911
|{                                                  |   |        |
3912
|    top_left_mb[ i ]                               |1  |ue(v)   |
3913
|    bottom_right_mb[ i ]                           |1  |ue(v)   |
3914
|   }                                               |   |        |
3915
#endif
3916
            break;
3917
        case 3:
3918
        case 4:
3919
        case 5:
3920
#if 0
3921
|   slice_group_change_direction_flag               |1  |u(1)    |
3922
|   slice_group_change_rate_minus1                  |1  |ue(v)   |
3923
#endif
3924
            break;
3925
        case 6:
3926
#if 0
3927
|   slice_group_id_cnt_minus1                       |1  |ue(v)   |
3928
|   for( i = 0; i <= slice_group_id_cnt_minus1; i++ |   |        |
3929
|)                                                  |   |        |
3930
|    slice_group_id[ i ]                            |1  |u(v)    |
3931
#endif
3932
            break;
3933
        }
3934
    }
3935
    pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3936
    pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3937
    if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
3938
        fprintf(stderr, "reference overflow (pps)\n");
3939
        return -1;
3940
    }
3941
    
3942
    pps->weighted_pred= get_bits1(&s->gb);
3943
    pps->weighted_bipred_idc= get_bits(&s->gb, 2);
3944
    pps->init_qp= get_se_golomb(&s->gb) + 26;
3945
    pps->init_qs= get_se_golomb(&s->gb) + 26;
3946
    pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
3947
    pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
3948
    pps->constrained_intra_pred= get_bits1(&s->gb);
3949
    pps->redundant_pic_cnt_present = get_bits1(&s->gb);
3950
    
3951
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3952
        printf("pps:%d sps:%d %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s\n", 
3953
               pps_id, pps->sps_id,
3954
               pps->cabac ? "CABAC" : "CAVLC",
3955
               pps->slice_group_count,
3956
               pps->ref_count[0], pps->ref_count[1],
3957
               pps->weighted_pred ? "weighted" : "",
3958
               pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
3959
               pps->deblocking_filter_parameters_present ? "LPAR" : "",
3960
               pps->constrained_intra_pred ? "CONSTR" : "",
3961
               pps->redundant_pic_cnt_present ? "REDU" : ""
3962
               );
3963
    }
3964
    
3965
    return 0;
3966
}
3967

    
3968
/**
3969
 * finds the end of the current frame in the bitstream.
3970
 * @return the position of the first byte of the next frame, or -1
3971
 */
3972
static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){
3973
    ParseContext *pc= &s->parse_context;
3974
    int i;
3975
    uint32_t state;
3976
//printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
3977
//    mb_addr= pc->mb_addr - 1;
3978
    state= pc->state;
3979
    //FIXME this will fail with slices
3980
    for(i=0; i<buf_size; i++){
3981
        state= (state<<8) | buf[i];
3982
        if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
3983
            if(pc->frame_start_found){
3984
                pc->state=-1; 
3985
                pc->frame_start_found= 0;
3986
                return i-3;
3987
            }
3988
            pc->frame_start_found= 1;
3989
        }
3990
    }
3991
    
3992
    pc->state= state;
3993
    return END_NOT_FOUND;
3994
}
3995

    
3996
static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
3997
    MpegEncContext * const s = &h->s;
3998
    AVCodecContext * const avctx= s->avctx;
3999
    int buf_index=0;
4000
#if 0
4001
    int i;
4002
    for(i=0; i<32; i++){
4003
        printf("%X ", buf[i]);
4004
    }
4005
#endif
4006
    for(;;){
4007
        int consumed;
4008
        int dst_length;
4009
        int bit_length;
4010
        uint8_t *ptr;
4011
        
4012
        // start code prefix search
4013
        for(; buf_index + 3 < buf_size; buf_index++){
4014
            // this should allways succeed in the first iteration
4015
            if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
4016
                break;
4017
        }
4018
        
4019
        if(buf_index+3 >= buf_size) break;
4020
        
4021
        buf_index+=3;
4022
        
4023
        ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index);
4024
        if(ptr[dst_length - 1] == 0) dst_length--;
4025
        bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
4026

    
4027
        if(s->avctx->debug&FF_DEBUG_STARTCODE){
4028
            printf("NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
4029
        }
4030
        
4031
        buf_index += consumed;
4032

    
4033
        if(h->nal_ref_idc < s->hurry_up)
4034
            continue;
4035
        
4036
        switch(h->nal_unit_type){
4037
        case NAL_IDR_SLICE:
4038
            idr(h); //FIXME ensure we dont loose some frames if there is reordering
4039
        case NAL_SLICE:
4040
            init_get_bits(&s->gb, ptr, bit_length);
4041
            h->intra_gb_ptr=
4042
            h->inter_gb_ptr= &s->gb;
4043
            s->data_partitioning = 0;
4044
            
4045
            if(decode_slice_header(h) < 0) return -1;
4046
            if(h->redundant_pic_count==0)
4047
                decode_slice(h);
4048
            break;
4049
        case NAL_DPA:
4050
            init_get_bits(&s->gb, ptr, bit_length);
4051
            h->intra_gb_ptr=
4052
            h->inter_gb_ptr= NULL;
4053
            s->data_partitioning = 1;
4054
            
4055
            if(decode_slice_header(h) < 0) return -1;
4056
            break;
4057
        case NAL_DPB:
4058
            init_get_bits(&h->intra_gb, ptr, bit_length);
4059
            h->intra_gb_ptr= &h->intra_gb;
4060
            break;
4061
        case NAL_DPC:
4062
            init_get_bits(&h->inter_gb, ptr, bit_length);
4063
            h->inter_gb_ptr= &h->inter_gb;
4064

    
4065
            if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning)
4066
                decode_slice(h);
4067
            break;
4068
        case NAL_SEI:
4069
            break;
4070
        case NAL_SPS:
4071
            init_get_bits(&s->gb, ptr, bit_length);
4072
            decode_seq_parameter_set(h);
4073
            
4074
            if(s->flags& CODEC_FLAG_LOW_DELAY)
4075
                s->low_delay=1;
4076
      
4077
            avctx->has_b_frames= !s->low_delay;
4078
            break;
4079
        case NAL_PPS:
4080
            init_get_bits(&s->gb, ptr, bit_length);
4081
            
4082
            decode_picture_parameter_set(h);
4083

    
4084
            break;
4085
        case NAL_PICTURE_DELIMITER:
4086
            break;
4087
        case NAL_FILTER_DATA:
4088
            break;
4089
        }        
4090

    
4091
        //FIXME move after where irt is set
4092
        s->current_picture.pict_type= s->pict_type;
4093
        s->current_picture.key_frame= s->pict_type == I_TYPE;
4094
    }
4095
    
4096
    if(!s->current_picture_ptr) return buf_index; //no frame
4097
    
4098
    h->prev_frame_num_offset= h->frame_num_offset;
4099
    h->prev_frame_num= h->frame_num;
4100
    if(s->current_picture_ptr->reference){
4101
        h->prev_poc_msb= h->poc_msb;
4102
        h->prev_poc_lsb= h->poc_lsb;
4103
    }
4104
    if(s->current_picture_ptr->reference)
4105
        execute_ref_pic_marking(h, h->mmco, h->mmco_index);
4106
    else
4107
        assert(h->mmco_index==0);
4108

    
4109
    ff_er_frame_end(s);
4110
    MPV_frame_end(s);
4111

    
4112
    return buf_index;
4113
}
4114

    
4115
/**
4116
 * retunrs the number of bytes consumed for building the current frame
4117
 */
4118
static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
4119
    if(s->flags&CODEC_FLAG_TRUNCATED){
4120
        pos -= s->parse_context.last_index;
4121
        if(pos<0) pos=0; // FIXME remove (uneeded?)
4122
        
4123
        return pos;
4124
    }else{
4125
        if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
4126
        if(pos+10>buf_size) pos=buf_size; // oops ;)
4127

    
4128
        return pos;
4129
    }
4130
}
4131

    
4132
static int decode_frame(AVCodecContext *avctx, 
4133
                             void *data, int *data_size,
4134
                             uint8_t *buf, int buf_size)
4135
{
4136
    H264Context *h = avctx->priv_data;
4137
    MpegEncContext *s = &h->s;
4138
    AVFrame *pict = data; 
4139
    int buf_index;
4140
    
4141
    s->flags= avctx->flags;
4142

    
4143
    *data_size = 0;
4144
   
4145
   /* no supplementary picture */
4146
    if (buf_size == 0) {
4147
        return 0;
4148
    }
4149
    
4150
    if(s->flags&CODEC_FLAG_TRUNCATED){
4151
        int next= find_frame_end(s, buf, buf_size);
4152
        
4153
        if( ff_combine_frame(s, next, &buf, &buf_size) < 0 )
4154
            return buf_size;
4155
//printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
4156
    }
4157

    
4158
    if(s->avctx->extradata_size && s->picture_number==0){
4159
        if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) ) 
4160
            return -1;
4161
    }
4162

    
4163
    buf_index=decode_nal_units(h, buf, buf_size);
4164
    if(buf_index < 0) 
4165
        return -1;
4166

    
4167
    //FIXME do something with unavailable reference frames    
4168
 
4169
//    if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
4170
#if 0
4171
    if(s->pict_type==B_TYPE || s->low_delay){
4172
        *pict= *(AVFrame*)&s->current_picture;
4173
    } else {
4174
        *pict= *(AVFrame*)&s->last_picture;
4175
    }
4176
#endif
4177
    if(!s->current_picture_ptr){
4178
        fprintf(stderr, "error, NO frame\n");
4179
        return -1;
4180
    }
4181

    
4182
    *pict= *(AVFrame*)&s->current_picture; //FIXME 
4183
    ff_print_debug_info(s, s->current_picture_ptr);
4184
    assert(pict->data[0]);
4185
//printf("out %d\n", (int)pict->data[0]);
4186
#if 0 //?
4187

4188
    /* Return the Picture timestamp as the frame number */
4189
    /* we substract 1 because it is added on utils.c    */
4190
    avctx->frame_number = s->picture_number - 1;
4191
#endif
4192
#if 0
4193
    /* dont output the last pic after seeking */
4194
    if(s->last_picture_ptr || s->low_delay)
4195
    //Note this isnt a issue as a IDR pic should flush teh buffers
4196
#endif
4197
        *data_size = sizeof(AVFrame);
4198
    return get_consumed_bytes(s, buf_index, buf_size);
4199
}
4200
#if 0
4201
static inline void fill_mb_avail(H264Context *h){
4202
    MpegEncContext * const s = &h->s;
4203
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4204

4205
    if(s->mb_y){
4206
        h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
4207
        h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
4208
        h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
4209
    }else{
4210
        h->mb_avail[0]=
4211
        h->mb_avail[1]=
4212
        h->mb_avail[2]= 0;
4213
    }
4214
    h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
4215
    h->mb_avail[4]= 1; //FIXME move out
4216
    h->mb_avail[5]= 0; //FIXME move out
4217
}
4218
#endif
4219

    
4220
#if 0 //selftest
4221
#define COUNT 8000
4222
#define SIZE (COUNT*40)
4223
int main(){
4224
    int i;
4225
    uint8_t temp[SIZE];
4226
    PutBitContext pb;
4227
    GetBitContext gb;
4228
//    int int_temp[10000];
4229
    DSPContext dsp;
4230
    AVCodecContext avctx;
4231
    
4232
    dsputil_init(&dsp, &avctx);
4233

4234
    init_put_bits(&pb, temp, SIZE);
4235
    printf("testing unsigned exp golomb\n");
4236
    for(i=0; i<COUNT; i++){
4237
        START_TIMER
4238
        set_ue_golomb(&pb, i);
4239
        STOP_TIMER("set_ue_golomb");
4240
    }
4241
    flush_put_bits(&pb);
4242
    
4243
    init_get_bits(&gb, temp, 8*SIZE);
4244
    for(i=0; i<COUNT; i++){
4245
        int j, s;
4246
        
4247
        s= show_bits(&gb, 24);
4248
        
4249
        START_TIMER
4250
        j= get_ue_golomb(&gb);
4251
        if(j != i){
4252
            printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4253
//            return -1;
4254
        }
4255
        STOP_TIMER("get_ue_golomb");
4256
    }
4257
    
4258
    
4259
    init_put_bits(&pb, temp, SIZE);
4260
    printf("testing signed exp golomb\n");
4261
    for(i=0; i<COUNT; i++){
4262
        START_TIMER
4263
        set_se_golomb(&pb, i - COUNT/2);
4264
        STOP_TIMER("set_se_golomb");
4265
    }
4266
    flush_put_bits(&pb);
4267
    
4268
    init_get_bits(&gb, temp, 8*SIZE);
4269
    for(i=0; i<COUNT; i++){
4270
        int j, s;
4271
        
4272
        s= show_bits(&gb, 24);
4273
        
4274
        START_TIMER
4275
        j= get_se_golomb(&gb);
4276
        if(j != i - COUNT/2){
4277
            printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4278
//            return -1;
4279
        }
4280
        STOP_TIMER("get_se_golomb");
4281
    }
4282

4283
    printf("testing 4x4 (I)DCT\n");
4284
    
4285
    DCTELEM block[16];
4286
    uint8_t src[16], ref[16];
4287
    uint64_t error= 0, max_error=0;
4288

4289
    for(i=0; i<COUNT; i++){
4290
        int j;
4291
//        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
4292
        for(j=0; j<16; j++){
4293
            ref[j]= random()%255;
4294
            src[j]= random()%255;
4295
        }
4296

4297
        h264_diff_dct_c(block, src, ref, 4);
4298
        
4299
        //normalize
4300
        for(j=0; j<16; j++){
4301
//            printf("%d ", block[j]);
4302
            block[j]= block[j]*4;
4303
            if(j&1) block[j]= (block[j]*4 + 2)/5;
4304
            if(j&4) block[j]= (block[j]*4 + 2)/5;
4305
        }
4306
//        printf("\n");
4307
        
4308
        h264_add_idct_c(ref, block, 4);
4309
/*        for(j=0; j<16; j++){
4310
            printf("%d ", ref[j]);
4311
        }
4312
        printf("\n");*/
4313
            
4314
        for(j=0; j<16; j++){
4315
            int diff= ABS(src[j] - ref[j]);
4316
            
4317
            error+= diff*diff;
4318
            max_error= FFMAX(max_error, diff);
4319
        }
4320
    }
4321
    printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
4322
#if 0
4323
    printf("testing quantizer\n");
4324
    for(qp=0; qp<52; qp++){
4325
        for(i=0; i<16; i++)
4326
            src1_block[i]= src2_block[i]= random()%255;
4327
        
4328
    }
4329
#endif
4330
    printf("Testing NAL layer\n");
4331
    
4332
    uint8_t bitstream[COUNT];
4333
    uint8_t nal[COUNT*2];
4334
    H264Context h;
4335
    memset(&h, 0, sizeof(H264Context));
4336
    
4337
    for(i=0; i<COUNT; i++){
4338
        int zeros= i;
4339
        int nal_length;
4340
        int consumed;
4341
        int out_length;
4342
        uint8_t *out;
4343
        int j;
4344
        
4345
        for(j=0; j<COUNT; j++){
4346
            bitstream[j]= (random() % 255) + 1;
4347
        }
4348
        
4349
        for(j=0; j<zeros; j++){
4350
            int pos= random() % COUNT;
4351
            while(bitstream[pos] == 0){
4352
                pos++;
4353
                pos %= COUNT;
4354
            }
4355
            bitstream[pos]=0;
4356
        }
4357
        
4358
        START_TIMER
4359
        
4360
        nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
4361
        if(nal_length<0){
4362
            printf("encoding failed\n");
4363
            return -1;
4364
        }
4365
        
4366
        out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
4367

    
4368
        STOP_TIMER("NAL")
4369
        
4370
        if(out_length != COUNT){
4371
            printf("incorrect length %d %d\n", out_length, COUNT);
4372
            return -1;
4373
        }
4374
        
4375
        if(consumed != nal_length){
4376
            printf("incorrect consumed length %d %d\n", nal_length, consumed);
4377
            return -1;
4378
        }
4379
        
4380
        if(memcmp(bitstream, out, COUNT)){
4381
            printf("missmatch\n");
4382
            return -1;
4383
        }
4384
    }
4385
    
4386
    printf("Testing RBSP\n");
4387
    
4388
    
4389
    return 0;
4390
}
4391
#endif
4392

    
4393

    
4394
static int decode_end(AVCodecContext *avctx)
4395
{
4396
    H264Context *h = avctx->priv_data;
4397
    MpegEncContext *s = &h->s;
4398
    
4399
    free_tables(h); //FIXME cleanup init stuff perhaps
4400
    MPV_common_end(s);
4401

    
4402
//    memset(h, 0, sizeof(H264Context));
4403
        
4404
    return 0;
4405
}
4406

    
4407

    
4408
AVCodec h264_decoder = {
4409
    "h264",
4410
    CODEC_TYPE_VIDEO,
4411
    CODEC_ID_H264,
4412
    sizeof(H264Context),
4413
    decode_init,
4414
    NULL,
4415
    decode_end,
4416
    decode_frame,
4417
    /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,
4418
};
4419

    
4420
#include "svq3.c"