Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.c @ 029911d1

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
    }else{
384
        topleft_xy = mb_xy-1 - s->mb_stride;
385
        top_xy     = mb_xy   - s->mb_stride;
386
        topright_xy= mb_xy+1 - s->mb_stride;
387
        left_xy[0]   = mb_xy-1;
388
        left_xy[1]   = mb_xy-1;
389
        left_block[0]= 0;
390
        left_block[1]= 1;
391
        left_block[2]= 2;
392
        left_block[3]= 3;
393
    }
394

    
395
    topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
396
    top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
397
    topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
398
    left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
399
    left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
400

    
401
    if(IS_INTRA(mb_type)){
402
        h->topleft_samples_available= 
403
        h->top_samples_available= 
404
        h->left_samples_available= 0xFFFF;
405
        h->topright_samples_available= 0xEEEA;
406

    
407
        if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
408
            h->topleft_samples_available= 0xB3FF;
409
            h->top_samples_available= 0x33FF;
410
            h->topright_samples_available= 0x26EA;
411
        }
412
        for(i=0; i<2; i++){
413
            if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
414
                h->topleft_samples_available&= 0xDF5F;
415
                h->left_samples_available&= 0x5F5F;
416
            }
417
        }
418
        
419
        if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
420
            h->topleft_samples_available&= 0x7FFF;
421
        
422
        if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
423
            h->topright_samples_available&= 0xFBFF;
424
    
425
        if(IS_INTRA4x4(mb_type)){
426
            if(IS_INTRA4x4(top_type)){
427
                h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
428
                h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
429
                h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
430
                h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
431
            }else{
432
                int pred;
433
                if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
434
                    pred= 2;
435
                else{
436
                    pred= -1;
437
                }
438
                h->intra4x4_pred_mode_cache[4+8*0]=
439
                h->intra4x4_pred_mode_cache[5+8*0]=
440
                h->intra4x4_pred_mode_cache[6+8*0]=
441
                h->intra4x4_pred_mode_cache[7+8*0]= pred;
442
            }
443
            for(i=0; i<2; i++){
444
                if(IS_INTRA4x4(left_type[i])){
445
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
446
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
447
                }else{
448
                    int pred;
449
                    if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
450
                        pred= 2;
451
                    else{
452
                        pred= -1;
453
                    }
454
                    h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
455
                    h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
456
                }
457
            }
458
        }
459
    }
460
    
461
    
462
/*
463
0 . T T. T T T T 
464
1 L . .L . . . . 
465
2 L . .L . . . . 
466
3 . T TL . . . . 
467
4 L . .L . . . . 
468
5 L . .. . . . . 
469
*/
470
//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
471
    if(top_type){
472
        h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
473
        h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
474
        h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
475
        h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
476
    
477
        h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
478
        h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
479
    
480
        h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
481
        h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
482
    }else{
483
        h->non_zero_count_cache[4+8*0]=      
484
        h->non_zero_count_cache[5+8*0]=
485
        h->non_zero_count_cache[6+8*0]=
486
        h->non_zero_count_cache[7+8*0]=
487
    
488
        h->non_zero_count_cache[1+8*0]=
489
        h->non_zero_count_cache[2+8*0]=
490
    
491
        h->non_zero_count_cache[1+8*3]=
492
        h->non_zero_count_cache[2+8*3]= 64;
493
    }
494
    
495
    if(left_type[0]){
496
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
497
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
498
        h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
499
        h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
500
    }else{
501
        h->non_zero_count_cache[3+8*1]= 
502
        h->non_zero_count_cache[3+8*2]= 
503
        h->non_zero_count_cache[0+8*1]= 
504
        h->non_zero_count_cache[0+8*4]= 64;
505
    }
506
    
507
    if(left_type[1]){
508
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
509
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
510
        h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
511
        h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
512
    }else{
513
        h->non_zero_count_cache[3+8*3]= 
514
        h->non_zero_count_cache[3+8*4]= 
515
        h->non_zero_count_cache[0+8*2]= 
516
        h->non_zero_count_cache[0+8*5]= 64;
517
    }
518
    
519
#if 1
520
    if(IS_INTER(mb_type)){
521
        int list;
522
        for(list=0; list<2; list++){
523
            if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
524
                /*if(!h->mv_cache_clean[list]){
525
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
526
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
527
                    h->mv_cache_clean[list]= 1;
528
                }*/
529
                continue; //FIXME direct mode ...
530
            }
531
            h->mv_cache_clean[list]= 0;
532
            
533
            if(IS_INTER(topleft_type)){
534
                const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
535
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
536
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
537
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
538
            }else{
539
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
540
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
541
            }
542
            
543
            if(IS_INTER(top_type)){
544
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
545
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
546
                *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
547
                *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
548
                *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
549
                *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
550
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
551
                h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
552
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
553
                h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
554
            }else{
555
                *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]= 
556
                *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]= 
557
                *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]= 
558
                *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
559
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
560
            }
561

    
562
            if(IS_INTER(topright_type)){
563
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
564
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
565
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
566
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
567
            }else{
568
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
569
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
570
            }
571
            
572
            //FIXME unify cleanup or sth
573
            if(IS_INTER(left_type[0])){
574
                const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
575
                const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
576
                *(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]];
577
                *(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]];
578
                h->ref_cache[list][scan8[0] - 1 + 0*8]= 
579
                h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
580
            }else{
581
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
582
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
583
                h->ref_cache[list][scan8[0] - 1 + 0*8]=
584
                h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
585
            }
586
            
587
            if(IS_INTER(left_type[1])){
588
                const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
589
                const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
590
                *(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]];
591
                *(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]];
592
                h->ref_cache[list][scan8[0] - 1 + 2*8]= 
593
                h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
594
            }else{
595
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
596
                *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
597
                h->ref_cache[list][scan8[0] - 1 + 2*8]=
598
                h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
599
            }
600

    
601
            h->ref_cache[list][scan8[5 ]+1] = 
602
            h->ref_cache[list][scan8[7 ]+1] = 
603
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewher else)
604
            h->ref_cache[list][scan8[4 ]] = 
605
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
606
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
607
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
608
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
609
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
610
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
611
        }
612
//FIXME
613

    
614
    }
615
#endif
616
}
617

    
618
static inline void write_back_intra_pred_mode(H264Context *h){
619
    MpegEncContext * const s = &h->s;
620
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
621

    
622
    h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
623
    h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
624
    h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
625
    h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
626
    h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
627
    h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
628
    h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
629
}
630

    
631
/**
632
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
633
 */
634
static inline int check_intra4x4_pred_mode(H264Context *h){
635
    MpegEncContext * const s = &h->s;
636
    static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
637
    static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
638
    int i;
639
    
640
    if(!(h->top_samples_available&0x8000)){
641
        for(i=0; i<4; i++){
642
            int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
643
            if(status<0){
644
                fprintf(stderr, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
645
                return -1;
646
            } else if(status){
647
                h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
648
            }
649
        }
650
    }
651
    
652
    if(!(h->left_samples_available&0x8000)){
653
        for(i=0; i<4; i++){
654
            int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
655
            if(status<0){
656
                fprintf(stderr, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
657
                return -1;
658
            } else if(status){
659
                h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
660
            }
661
        }
662
    }
663

    
664
    return 0;
665
} //FIXME cleanup like next
666

    
667
/**
668
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
669
 */
670
static inline int check_intra_pred_mode(H264Context *h, int mode){
671
    MpegEncContext * const s = &h->s;
672
    static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
673
    static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
674
    
675
    if(!(h->top_samples_available&0x8000)){
676
        mode= top[ mode ];
677
        if(mode<0){
678
            fprintf(stderr, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
679
            return -1;
680
        }
681
    }
682
    
683
    if(!(h->left_samples_available&0x8000)){
684
        mode= left[ mode ];
685
        if(mode<0){
686
            fprintf(stderr, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
687
            return -1;
688
        } 
689
    }
690

    
691
    return mode;
692
}
693

    
694
/**
695
 * gets the predicted intra4x4 prediction mode.
696
 */
697
static inline int pred_intra_mode(H264Context *h, int n){
698
    const int index8= scan8[n];
699
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
700
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
701
    const int min= FFMIN(left, top);
702

    
703
    tprintf("mode:%d %d min:%d\n", left ,top, min);
704

    
705
    if(min<0) return DC_PRED;
706
    else      return min;
707
}
708

    
709
static inline void write_back_non_zero_count(H264Context *h){
710
    MpegEncContext * const s = &h->s;
711
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
712

    
713
    h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
714
    h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
715
    h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
716
    h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
717
    h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
718
    h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
719
    h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
720
    
721
    h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
722
    h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
723
    h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
724

    
725
    h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
726
    h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
727
    h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
728
}
729

    
730
/**
731
 * gets the predicted number of non zero coefficients.
732
 * @param n block index
733
 */
734
static inline int pred_non_zero_count(H264Context *h, int n){
735
    const int index8= scan8[n];
736
    const int left= h->non_zero_count_cache[index8 - 1];
737
    const int top = h->non_zero_count_cache[index8 - 8];
738
    int i= left + top;
739
    
740
    if(i<64) i= (i+1)>>1;
741

    
742
    tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
743

    
744
    return i&31;
745
}
746

    
747
static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
748
    const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
749

    
750
    if(topright_ref != PART_NOT_AVAILABLE){
751
        *C= h->mv_cache[list][ i - 8 + part_width ];
752
        return topright_ref;
753
    }else{
754
        tprintf("topright MV not available\n");
755

    
756
        *C= h->mv_cache[list][ i - 8 - 1 ];
757
        return h->ref_cache[list][ i - 8 - 1 ];
758
    }
759
}
760

    
761
/**
762
 * gets the predicted MV.
763
 * @param n the block index
764
 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
765
 * @param mx the x component of the predicted motion vector
766
 * @param my the y component of the predicted motion vector
767
 */
768
static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
769
    const int index8= scan8[n];
770
    const int top_ref=      h->ref_cache[list][ index8 - 8 ];
771
    const int left_ref=     h->ref_cache[list][ index8 - 1 ];
772
    const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
773
    const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
774
    const int16_t * C;
775
    int diagonal_ref, match_count;
776

    
777
    assert(part_width==1 || part_width==2 || part_width==4);
778

    
779
/* mv_cache
780
  B . . A T T T T 
781
  U . . L . . , .
782
  U . . L . . . .
783
  U . . L . . , .
784
  . . . L . . . .
785
*/
786

    
787
    diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
788
    match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
789
    if(match_count > 1){ //most common
790
        *mx= mid_pred(A[0], B[0], C[0]);
791
        *my= mid_pred(A[1], B[1], C[1]);
792
    }else if(match_count==1){
793
        if(left_ref==ref){
794
            *mx= A[0];
795
            *my= A[1];        
796
        }else if(top_ref==ref){
797
            *mx= B[0];
798
            *my= B[1];        
799
        }else{
800
            *mx= C[0];
801
            *my= C[1];        
802
        }
803
    }else{
804
        if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
805
            *mx= A[0];
806
            *my= A[1];        
807
        }else{
808
            *mx= mid_pred(A[0], B[0], C[0]);
809
            *my= mid_pred(A[1], B[1], C[1]);
810
        }
811
    }
812
        
813
    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);
814
}
815

    
816
/**
817
 * gets the directionally predicted 16x8 MV.
818
 * @param n the block index
819
 * @param mx the x component of the predicted motion vector
820
 * @param my the y component of the predicted motion vector
821
 */
822
static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
823
    if(n==0){
824
        const int top_ref=      h->ref_cache[list][ scan8[0] - 8 ];
825
        const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
826

    
827
        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);
828
        
829
        if(top_ref == ref){
830
            *mx= B[0];
831
            *my= B[1];
832
            return;
833
        }
834
    }else{
835
        const int left_ref=     h->ref_cache[list][ scan8[8] - 1 ];
836
        const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
837
        
838
        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);
839

    
840
        if(left_ref == ref){
841
            *mx= A[0];
842
            *my= A[1];
843
            return;
844
        }
845
    }
846

    
847
    //RARE
848
    pred_motion(h, n, 4, list, ref, mx, my);
849
}
850

    
851
/**
852
 * gets the directionally predicted 8x16 MV.
853
 * @param n the block index
854
 * @param mx the x component of the predicted motion vector
855
 * @param my the y component of the predicted motion vector
856
 */
857
static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
858
    if(n==0){
859
        const int left_ref=      h->ref_cache[list][ scan8[0] - 1 ];
860
        const int16_t * const A=  h->mv_cache[list][ scan8[0] - 1 ];
861
        
862
        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);
863

    
864
        if(left_ref == ref){
865
            *mx= A[0];
866
            *my= A[1];
867
            return;
868
        }
869
    }else{
870
        const int16_t * C;
871
        int diagonal_ref;
872

    
873
        diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
874
        
875
        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);
876

    
877
        if(diagonal_ref == ref){ 
878
            *mx= C[0];
879
            *my= C[1];
880
            return;
881
        }
882
    }
883

    
884
    //RARE
885
    pred_motion(h, n, 2, list, ref, mx, my);
886
}
887

    
888
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
889
    const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
890
    const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
891

    
892
    tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
893

    
894
    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
895
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
896
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
897
       
898
        *mx = *my = 0;
899
        return;
900
    }
901
        
902
    pred_motion(h, 0, 4, 0, 0, mx, my);
903

    
904
    return;
905
}
906

    
907
static inline void write_back_motion(H264Context *h, int mb_type){
908
    MpegEncContext * const s = &h->s;
909
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
910
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
911
    int list;
912

    
913
    for(list=0; list<2; list++){
914
        int y;
915
        if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
916
            if(1){ //FIXME skip or never read if mb_type doesnt use it
917
                for(y=0; y<4; y++){
918
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
919
                    *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
920
                }
921
                for(y=0; y<2; y++){
922
                    *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
923
                }
924
            }
925
            continue; //FIXME direct mode ...
926
        }
927
        
928
        for(y=0; y<4; y++){
929
            *(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];
930
            *(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];
931
        }
932
        for(y=0; y<2; y++){
933
            s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
934
            s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
935
        }
936
    }
937
}
938

    
939
/**
940
 * Decodes a network abstraction layer unit.
941
 * @param consumed is the number of bytes used as input
942
 * @param length is the length of the array
943
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
944
 * @returns decoded bytes, might be src+1 if no escapes 
945
 */
946
static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
947
    int i, si, di;
948
    uint8_t *dst;
949

    
950
//    src[0]&0x80;                //forbidden bit
951
    h->nal_ref_idc= src[0]>>5;
952
    h->nal_unit_type= src[0]&0x1F;
953

    
954
    src++; length--;
955
#if 0    
956
    for(i=0; i<length; i++)
957
        printf("%2X ", src[i]);
958
#endif
959
    for(i=0; i+1<length; i+=2){
960
        if(src[i]) continue;
961
        if(i>0 && src[i-1]==0) i--;
962
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
963
            if(src[i+2]!=3){
964
                /* startcode, so we must be past the end */
965
                length=i;
966
            }
967
            break;
968
        }
969
    }
970

    
971
    if(i>=length-1){ //no escaped 0
972
        *dst_length= length;
973
        *consumed= length+1; //+1 for the header
974
        return src; 
975
    }
976

    
977
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
978
    dst= h->rbsp_buffer;
979

    
980
//printf("deoding esc\n");
981
    si=di=0;
982
    while(si<length){ 
983
        //remove escapes (very rare 1:2^22)
984
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
985
            if(src[si+2]==3){ //escape
986
                dst[di++]= 0;
987
                dst[di++]= 0;
988
                si+=3;
989
            }else //next start code
990
                break;
991
        }
992

    
993
        dst[di++]= src[si++];
994
    }
995

    
996
    *dst_length= di;
997
    *consumed= si + 1;//+1 for the header
998
//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
999
    return dst;
1000
}
1001

    
1002
/**
1003
 * @param src the data which should be escaped
1004
 * @param dst the target buffer, dst+1 == src is allowed as a special case
1005
 * @param length the length of the src data
1006
 * @param dst_length the length of the dst array
1007
 * @returns length of escaped data in bytes or -1 if an error occured
1008
 */
1009
static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1010
    int i, escape_count, si, di;
1011
    uint8_t *temp;
1012
    
1013
    assert(length>=0);
1014
    assert(dst_length>0);
1015
    
1016
    dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1017

    
1018
    if(length==0) return 1;
1019

    
1020
    escape_count= 0;
1021
    for(i=0; i<length; i+=2){
1022
        if(src[i]) continue;
1023
        if(i>0 && src[i-1]==0) 
1024
            i--;
1025
        if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1026
            escape_count++;
1027
            i+=2;
1028
        }
1029
    }
1030
    
1031
    if(escape_count==0){ 
1032
        if(dst+1 != src)
1033
            memcpy(dst+1, src, length);
1034
        return length + 1;
1035
    }
1036
    
1037
    if(length + escape_count + 1> dst_length)
1038
        return -1;
1039

    
1040
    //this should be damn rare (hopefully)
1041

    
1042
    h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1043
    temp= h->rbsp_buffer;
1044
//printf("encoding esc\n");
1045
    
1046
    si= 0;
1047
    di= 0;
1048
    while(si < length){
1049
        if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1050
            temp[di++]= 0; si++;
1051
            temp[di++]= 0; si++;
1052
            temp[di++]= 3; 
1053
            temp[di++]= src[si++];
1054
        }
1055
        else
1056
            temp[di++]= src[si++];
1057
    }
1058
    memcpy(dst+1, temp, length+escape_count);
1059
    
1060
    assert(di == length+escape_count);
1061
    
1062
    return di + 1;
1063
}
1064

    
1065
/**
1066
 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1067
 */
1068
static void encode_rbsp_trailing(PutBitContext *pb){
1069
    int length;
1070
    put_bits(pb, 1, 1);
1071
    length= (-get_bit_count(pb))&7;
1072
    if(length) put_bits(pb, length, 0);
1073
}
1074

    
1075
/**
1076
 * identifies the exact end of the bitstream
1077
 * @return the length of the trailing, or 0 if damaged
1078
 */
1079
static int decode_rbsp_trailing(uint8_t *src){
1080
    int v= *src;
1081
    int r;
1082

    
1083
    tprintf("rbsp trailing %X\n", v);
1084

    
1085
    for(r=1; r<9; r++){
1086
        if(v&1) return r;
1087
        v>>=1;
1088
    }
1089
    return 0;
1090
}
1091

    
1092
/**
1093
 * idct tranforms the 16 dc values and dequantize them.
1094
 * @param qp quantization parameter
1095
 */
1096
static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1097
    const int qmul= dequant_coeff[qp][0];
1098
#define stride 16
1099
    int i;
1100
    int temp[16]; //FIXME check if this is a good idea
1101
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1102
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1103

    
1104
//memset(block, 64, 2*256);
1105
//return;
1106
    for(i=0; i<4; i++){
1107
        const int offset= y_offset[i];
1108
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1109
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1110
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1111
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1112

    
1113
        temp[4*i+0]= z0+z3;
1114
        temp[4*i+1]= z1+z2;
1115
        temp[4*i+2]= z1-z2;
1116
        temp[4*i+3]= z0-z3;
1117
    }
1118

    
1119
    for(i=0; i<4; i++){
1120
        const int offset= x_offset[i];
1121
        const int z0= temp[4*0+i] + temp[4*2+i];
1122
        const int z1= temp[4*0+i] - temp[4*2+i];
1123
        const int z2= temp[4*1+i] - temp[4*3+i];
1124
        const int z3= temp[4*1+i] + temp[4*3+i];
1125

    
1126
        block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1127
        block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1128
        block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1129
        block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1130
    }
1131
}
1132

    
1133
/**
1134
 * dct tranforms the 16 dc values.
1135
 * @param qp quantization parameter ??? FIXME
1136
 */
1137
static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1138
//    const int qmul= dequant_coeff[qp][0];
1139
    int i;
1140
    int temp[16]; //FIXME check if this is a good idea
1141
    static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};
1142
    static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1143

    
1144
    for(i=0; i<4; i++){
1145
        const int offset= y_offset[i];
1146
        const int z0= block[offset+stride*0] + block[offset+stride*4];
1147
        const int z1= block[offset+stride*0] - block[offset+stride*4];
1148
        const int z2= block[offset+stride*1] - block[offset+stride*5];
1149
        const int z3= block[offset+stride*1] + block[offset+stride*5];
1150

    
1151
        temp[4*i+0]= z0+z3;
1152
        temp[4*i+1]= z1+z2;
1153
        temp[4*i+2]= z1-z2;
1154
        temp[4*i+3]= z0-z3;
1155
    }
1156

    
1157
    for(i=0; i<4; i++){
1158
        const int offset= x_offset[i];
1159
        const int z0= temp[4*0+i] + temp[4*2+i];
1160
        const int z1= temp[4*0+i] - temp[4*2+i];
1161
        const int z2= temp[4*1+i] - temp[4*3+i];
1162
        const int z3= temp[4*1+i] + temp[4*3+i];
1163

    
1164
        block[stride*0 +offset]= (z0 + z3)>>1;
1165
        block[stride*2 +offset]= (z1 + z2)>>1;
1166
        block[stride*8 +offset]= (z1 - z2)>>1;
1167
        block[stride*10+offset]= (z0 - z3)>>1;
1168
    }
1169
}
1170
#undef xStride
1171
#undef stride
1172

    
1173
static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1174
    const int qmul= dequant_coeff[qp][0];
1175
    const int stride= 16*2;
1176
    const int xStride= 16;
1177
    int a,b,c,d,e;
1178

    
1179
    a= block[stride*0 + xStride*0];
1180
    b= block[stride*0 + xStride*1];
1181
    c= block[stride*1 + xStride*0];
1182
    d= block[stride*1 + xStride*1];
1183

    
1184
    e= a-b;
1185
    a= a+b;
1186
    b= c-d;
1187
    c= c+d;
1188

    
1189
    block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1190
    block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1191
    block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1192
    block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1193
}
1194

    
1195
static void chroma_dc_dct_c(DCTELEM *block){
1196
    const int stride= 16*2;
1197
    const int xStride= 16;
1198
    int a,b,c,d,e;
1199

    
1200
    a= block[stride*0 + xStride*0];
1201
    b= block[stride*0 + xStride*1];
1202
    c= block[stride*1 + xStride*0];
1203
    d= block[stride*1 + xStride*1];
1204

    
1205
    e= a-b;
1206
    a= a+b;
1207
    b= c-d;
1208
    c= c+d;
1209

    
1210
    block[stride*0 + xStride*0]= (a+c);
1211
    block[stride*0 + xStride*1]= (e+b);
1212
    block[stride*1 + xStride*0]= (a-c);
1213
    block[stride*1 + xStride*1]= (e-b);
1214
}
1215

    
1216
/**
1217
 * gets the chroma qp.
1218
 */
1219
static inline int get_chroma_qp(H264Context *h, int qscale){
1220
    
1221
    return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1222
}
1223

    
1224

    
1225
/**
1226
 *
1227
 */
1228
static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1229
    int i;
1230
    uint8_t *cm = cropTbl + MAX_NEG_CROP;
1231

    
1232
    block[0] += 32;
1233
#if 1
1234
    for(i=0; i<4; i++){
1235
        const int z0=  block[i + 4*0]     +  block[i + 4*2];
1236
        const int z1=  block[i + 4*0]     -  block[i + 4*2];
1237
        const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
1238
        const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
1239

    
1240
        block[i + 4*0]= z0 + z3;
1241
        block[i + 4*1]= z1 + z2;
1242
        block[i + 4*2]= z1 - z2;
1243
        block[i + 4*3]= z0 - z3;
1244
    }
1245

    
1246
    for(i=0; i<4; i++){
1247
        const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
1248
        const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
1249
        const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
1250
        const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
1251

    
1252
        dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ];
1253
        dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ];
1254
        dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ];
1255
        dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ];
1256
    }
1257
#else
1258
    for(i=0; i<4; i++){
1259
        const int z0=  block[0 + 4*i]     +  block[2 + 4*i];
1260
        const int z1=  block[0 + 4*i]     -  block[2 + 4*i];
1261
        const int z2= (block[1 + 4*i]>>1) -  block[3 + 4*i];
1262
        const int z3=  block[1 + 4*i]     + (block[3 + 4*i]>>1);
1263

    
1264
        block[0 + 4*i]= z0 + z3;
1265
        block[1 + 4*i]= z1 + z2;
1266
        block[2 + 4*i]= z1 - z2;
1267
        block[3 + 4*i]= z0 - z3;
1268
    }
1269

    
1270
    for(i=0; i<4; i++){
1271
        const int z0=  block[i + 4*0]     +  block[i + 4*2];
1272
        const int z1=  block[i + 4*0]     -  block[i + 4*2];
1273
        const int z2= (block[i + 4*1]>>1) -  block[i + 4*3];
1274
        const int z3=  block[i + 4*1]     + (block[i + 4*3]>>1);
1275

    
1276
        dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1277
        dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1278
        dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1279
        dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1280
    }
1281
#endif
1282
}
1283

    
1284
static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1285
    int i;
1286
    //FIXME try int temp instead of block
1287
    
1288
    for(i=0; i<4; i++){
1289
        const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1290
        const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1291
        const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1292
        const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1293
        const int z0= d0 + d3;
1294
        const int z3= d0 - d3;
1295
        const int z1= d1 + d2;
1296
        const int z2= d1 - d2;
1297
        
1298
        block[0 + 4*i]=   z0 +   z1;
1299
        block[1 + 4*i]= 2*z3 +   z2;
1300
        block[2 + 4*i]=   z0 -   z1;
1301
        block[3 + 4*i]=   z3 - 2*z2;
1302
    }    
1303

    
1304
    for(i=0; i<4; i++){
1305
        const int z0= block[0*4 + i] + block[3*4 + i];
1306
        const int z3= block[0*4 + i] - block[3*4 + i];
1307
        const int z1= block[1*4 + i] + block[2*4 + i];
1308
        const int z2= block[1*4 + i] - block[2*4 + i];
1309
        
1310
        block[0*4 + i]=   z0 +   z1;
1311
        block[1*4 + i]= 2*z3 +   z2;
1312
        block[2*4 + i]=   z0 -   z1;
1313
        block[3*4 + i]=   z3 - 2*z2;
1314
    }
1315
}
1316

    
1317
//FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1318
//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1319
static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1320
    int i;
1321
    const int * const quant_table= quant_coeff[qscale];
1322
    const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1323
    const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1324
    const unsigned int threshold2= (threshold1<<1);
1325
    int last_non_zero;
1326

    
1327
    if(seperate_dc){
1328
        if(qscale<=18){
1329
            //avoid overflows
1330
            const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1331
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1332
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1333

    
1334
            int level= block[0]*quant_coeff[qscale+18][0];
1335
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1336
                if(level>0){
1337
                    level= (dc_bias + level)>>(QUANT_SHIFT-2);
1338
                    block[0]= level;
1339
                }else{
1340
                    level= (dc_bias - level)>>(QUANT_SHIFT-2);
1341
                    block[0]= -level;
1342
                }
1343
//                last_non_zero = i;
1344
            }else{
1345
                block[0]=0;
1346
            }
1347
        }else{
1348
            const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1349
            const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1350
            const unsigned int dc_threshold2= (dc_threshold1<<1);
1351

    
1352
            int level= block[0]*quant_table[0];
1353
            if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1354
                if(level>0){
1355
                    level= (dc_bias + level)>>(QUANT_SHIFT+1);
1356
                    block[0]= level;
1357
                }else{
1358
                    level= (dc_bias - level)>>(QUANT_SHIFT+1);
1359
                    block[0]= -level;
1360
                }
1361
//                last_non_zero = i;
1362
            }else{
1363
                block[0]=0;
1364
            }
1365
        }
1366
        last_non_zero= 0;
1367
        i=1;
1368
    }else{
1369
        last_non_zero= -1;
1370
        i=0;
1371
    }
1372

    
1373
    for(; i<16; i++){
1374
        const int j= scantable[i];
1375
        int level= block[j]*quant_table[j];
1376

    
1377
//        if(   bias+level >= (1<<(QMAT_SHIFT - 3))
1378
//           || bias-level >= (1<<(QMAT_SHIFT - 3))){
1379
        if(((unsigned)(level+threshold1))>threshold2){
1380
            if(level>0){
1381
                level= (bias + level)>>QUANT_SHIFT;
1382
                block[j]= level;
1383
            }else{
1384
                level= (bias - level)>>QUANT_SHIFT;
1385
                block[j]= -level;
1386
            }
1387
            last_non_zero = i;
1388
        }else{
1389
            block[j]=0;
1390
        }
1391
    }
1392

    
1393
    return last_non_zero;
1394
}
1395

    
1396
static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1397
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1398
    ((uint32_t*)(src+0*stride))[0]= a;
1399
    ((uint32_t*)(src+1*stride))[0]= a;
1400
    ((uint32_t*)(src+2*stride))[0]= a;
1401
    ((uint32_t*)(src+3*stride))[0]= a;
1402
}
1403

    
1404
static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1405
    ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1406
    ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1407
    ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1408
    ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1409
}
1410

    
1411
static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1412
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1413
                   + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1414
    
1415
    ((uint32_t*)(src+0*stride))[0]= 
1416
    ((uint32_t*)(src+1*stride))[0]= 
1417
    ((uint32_t*)(src+2*stride))[0]= 
1418
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1419
}
1420

    
1421
static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1422
    const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1423
    
1424
    ((uint32_t*)(src+0*stride))[0]= 
1425
    ((uint32_t*)(src+1*stride))[0]= 
1426
    ((uint32_t*)(src+2*stride))[0]= 
1427
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1428
}
1429

    
1430
static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1431
    const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1432
    
1433
    ((uint32_t*)(src+0*stride))[0]= 
1434
    ((uint32_t*)(src+1*stride))[0]= 
1435
    ((uint32_t*)(src+2*stride))[0]= 
1436
    ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101; 
1437
}
1438

    
1439
static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1440
    ((uint32_t*)(src+0*stride))[0]= 
1441
    ((uint32_t*)(src+1*stride))[0]= 
1442
    ((uint32_t*)(src+2*stride))[0]= 
1443
    ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1444
}
1445

    
1446

    
1447
#define LOAD_TOP_RIGHT_EDGE\
1448
    const int t4= topright[0];\
1449
    const int t5= topright[1];\
1450
    const int t6= topright[2];\
1451
    const int t7= topright[3];\
1452

    
1453
#define LOAD_LEFT_EDGE\
1454
    const int l0= src[-1+0*stride];\
1455
    const int l1= src[-1+1*stride];\
1456
    const int l2= src[-1+2*stride];\
1457
    const int l3= src[-1+3*stride];\
1458

    
1459
#define LOAD_TOP_EDGE\
1460
    const int t0= src[ 0-1*stride];\
1461
    const int t1= src[ 1-1*stride];\
1462
    const int t2= src[ 2-1*stride];\
1463
    const int t3= src[ 3-1*stride];\
1464

    
1465
static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1466
    const int lt= src[-1-1*stride];
1467
    LOAD_TOP_EDGE
1468
    LOAD_LEFT_EDGE
1469

    
1470
    src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2; 
1471
    src[0+2*stride]=
1472
    src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2; 
1473
    src[0+1*stride]=
1474
    src[1+2*stride]=
1475
    src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2; 
1476
    src[0+0*stride]=
1477
    src[1+1*stride]=
1478
    src[2+2*stride]=
1479
    src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2; 
1480
    src[1+0*stride]=
1481
    src[2+1*stride]=
1482
    src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1483
    src[2+0*stride]=
1484
    src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1485
    src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1486
}
1487

    
1488
static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1489
    LOAD_TOP_EDGE    
1490
    LOAD_TOP_RIGHT_EDGE    
1491
//    LOAD_LEFT_EDGE    
1492

    
1493
    src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1494
    src[1+0*stride]=
1495
    src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1496
    src[2+0*stride]=
1497
    src[1+1*stride]=
1498
    src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1499
    src[3+0*stride]=
1500
    src[2+1*stride]=
1501
    src[1+2*stride]=
1502
    src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1503
    src[3+1*stride]=
1504
    src[2+2*stride]=
1505
    src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1506
    src[3+2*stride]=
1507
    src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1508
    src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1509
}
1510

    
1511
static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1512
    const int lt= src[-1-1*stride];
1513
    LOAD_TOP_EDGE    
1514
    LOAD_LEFT_EDGE    
1515
    const __attribute__((unused)) int unu= l3;
1516

    
1517
    src[0+0*stride]=
1518
    src[1+2*stride]=(lt + t0 + 1)>>1;
1519
    src[1+0*stride]=
1520
    src[2+2*stride]=(t0 + t1 + 1)>>1;
1521
    src[2+0*stride]=
1522
    src[3+2*stride]=(t1 + t2 + 1)>>1;
1523
    src[3+0*stride]=(t2 + t3 + 1)>>1;
1524
    src[0+1*stride]=
1525
    src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1526
    src[1+1*stride]=
1527
    src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1528
    src[2+1*stride]=
1529
    src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1530
    src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1531
    src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1532
    src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1533
}
1534

    
1535
static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1536
    LOAD_TOP_EDGE    
1537
    LOAD_TOP_RIGHT_EDGE    
1538
    const __attribute__((unused)) int unu= t7;
1539

    
1540
    src[0+0*stride]=(t0 + t1 + 1)>>1;
1541
    src[1+0*stride]=
1542
    src[0+2*stride]=(t1 + t2 + 1)>>1;
1543
    src[2+0*stride]=
1544
    src[1+2*stride]=(t2 + t3 + 1)>>1;
1545
    src[3+0*stride]=
1546
    src[2+2*stride]=(t3 + t4+ 1)>>1;
1547
    src[3+2*stride]=(t4 + t5+ 1)>>1;
1548
    src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1549
    src[1+1*stride]=
1550
    src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1551
    src[2+1*stride]=
1552
    src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1553
    src[3+1*stride]=
1554
    src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1555
    src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1556
}
1557

    
1558
static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1559
    LOAD_LEFT_EDGE    
1560

    
1561
    src[0+0*stride]=(l0 + l1 + 1)>>1;
1562
    src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1563
    src[2+0*stride]=
1564
    src[0+1*stride]=(l1 + l2 + 1)>>1;
1565
    src[3+0*stride]=
1566
    src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1567
    src[2+1*stride]=
1568
    src[0+2*stride]=(l2 + l3 + 1)>>1;
1569
    src[3+1*stride]=
1570
    src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1571
    src[3+2*stride]=
1572
    src[1+3*stride]=
1573
    src[0+3*stride]=
1574
    src[2+2*stride]=
1575
    src[2+3*stride]=
1576
    src[3+3*stride]=l3;
1577
}
1578
    
1579
static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1580
    const int lt= src[-1-1*stride];
1581
    LOAD_TOP_EDGE    
1582
    LOAD_LEFT_EDGE    
1583
    const __attribute__((unused)) int unu= t3;
1584

    
1585
    src[0+0*stride]=
1586
    src[2+1*stride]=(lt + l0 + 1)>>1;
1587
    src[1+0*stride]=
1588
    src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1589
    src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1590
    src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1591
    src[0+1*stride]=
1592
    src[2+2*stride]=(l0 + l1 + 1)>>1;
1593
    src[1+1*stride]=
1594
    src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1595
    src[0+2*stride]=
1596
    src[2+3*stride]=(l1 + l2+ 1)>>1;
1597
    src[1+2*stride]=
1598
    src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1599
    src[0+3*stride]=(l2 + l3 + 1)>>1;
1600
    src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1601
}
1602

    
1603
static void pred16x16_vertical_c(uint8_t *src, int stride){
1604
    int i;
1605
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1606
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1607
    const uint32_t c= ((uint32_t*)(src-stride))[2];
1608
    const uint32_t d= ((uint32_t*)(src-stride))[3];
1609
    
1610
    for(i=0; i<16; i++){
1611
        ((uint32_t*)(src+i*stride))[0]= a;
1612
        ((uint32_t*)(src+i*stride))[1]= b;
1613
        ((uint32_t*)(src+i*stride))[2]= c;
1614
        ((uint32_t*)(src+i*stride))[3]= d;
1615
    }
1616
}
1617

    
1618
static void pred16x16_horizontal_c(uint8_t *src, int stride){
1619
    int i;
1620

    
1621
    for(i=0; i<16; i++){
1622
        ((uint32_t*)(src+i*stride))[0]=
1623
        ((uint32_t*)(src+i*stride))[1]=
1624
        ((uint32_t*)(src+i*stride))[2]=
1625
        ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1626
    }
1627
}
1628

    
1629
static void pred16x16_dc_c(uint8_t *src, int stride){
1630
    int i, dc=0;
1631

    
1632
    for(i=0;i<16; i++){
1633
        dc+= src[-1+i*stride];
1634
    }
1635
    
1636
    for(i=0;i<16; i++){
1637
        dc+= src[i-stride];
1638
    }
1639

    
1640
    dc= 0x01010101*((dc + 16)>>5);
1641

    
1642
    for(i=0; i<16; i++){
1643
        ((uint32_t*)(src+i*stride))[0]=
1644
        ((uint32_t*)(src+i*stride))[1]=
1645
        ((uint32_t*)(src+i*stride))[2]=
1646
        ((uint32_t*)(src+i*stride))[3]= dc;
1647
    }
1648
}
1649

    
1650
static void pred16x16_left_dc_c(uint8_t *src, int stride){
1651
    int i, dc=0;
1652

    
1653
    for(i=0;i<16; i++){
1654
        dc+= src[-1+i*stride];
1655
    }
1656
    
1657
    dc= 0x01010101*((dc + 8)>>4);
1658

    
1659
    for(i=0; i<16; i++){
1660
        ((uint32_t*)(src+i*stride))[0]=
1661
        ((uint32_t*)(src+i*stride))[1]=
1662
        ((uint32_t*)(src+i*stride))[2]=
1663
        ((uint32_t*)(src+i*stride))[3]= dc;
1664
    }
1665
}
1666

    
1667
static void pred16x16_top_dc_c(uint8_t *src, int stride){
1668
    int i, dc=0;
1669

    
1670
    for(i=0;i<16; i++){
1671
        dc+= src[i-stride];
1672
    }
1673
    dc= 0x01010101*((dc + 8)>>4);
1674

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

    
1683
static void pred16x16_128_dc_c(uint8_t *src, int stride){
1684
    int i;
1685

    
1686
    for(i=0; i<16; i++){
1687
        ((uint32_t*)(src+i*stride))[0]=
1688
        ((uint32_t*)(src+i*stride))[1]=
1689
        ((uint32_t*)(src+i*stride))[2]=
1690
        ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1691
    }
1692
}
1693

    
1694
static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1695
  int i, j, k;
1696
  int a;
1697
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
1698
  const uint8_t * const src0 = src+7-stride;
1699
  const uint8_t *src1 = src+8*stride-1;
1700
  const uint8_t *src2 = src1-2*stride;      // == src+6*stride-1;
1701
  int H = src0[1] - src0[-1];
1702
  int V = src1[0] - src2[ 0];
1703
  for(k=2; k<=8; ++k) {
1704
    src1 += stride; src2 -= stride;
1705
    H += k*(src0[k] - src0[-k]);
1706
    V += k*(src1[0] - src2[ 0]);
1707
  }
1708
  if(svq3){
1709
    H = ( 5*(H/4) ) / 16;
1710
    V = ( 5*(V/4) ) / 16;
1711

    
1712
    /* required for 100% accuracy */
1713
    i = H; H = V; V = i;
1714
  }else{
1715
    H = ( 5*H+32 ) >> 6;
1716
    V = ( 5*V+32 ) >> 6;
1717
  }
1718

    
1719
  a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1720
  for(j=16; j>0; --j) {
1721
    int b = a;
1722
    a += V;
1723
    for(i=-16; i<0; i+=4) {
1724
      src[16+i] = cm[ (b    ) >> 5 ];
1725
      src[17+i] = cm[ (b+  H) >> 5 ];
1726
      src[18+i] = cm[ (b+2*H) >> 5 ];
1727
      src[19+i] = cm[ (b+3*H) >> 5 ];
1728
      b += 4*H;
1729
    }
1730
    src += stride;
1731
  }
1732
}
1733

    
1734
static void pred16x16_plane_c(uint8_t *src, int stride){
1735
    pred16x16_plane_compat_c(src, stride, 0);
1736
}
1737

    
1738
static void pred8x8_vertical_c(uint8_t *src, int stride){
1739
    int i;
1740
    const uint32_t a= ((uint32_t*)(src-stride))[0];
1741
    const uint32_t b= ((uint32_t*)(src-stride))[1];
1742
    
1743
    for(i=0; i<8; i++){
1744
        ((uint32_t*)(src+i*stride))[0]= a;
1745
        ((uint32_t*)(src+i*stride))[1]= b;
1746
    }
1747
}
1748

    
1749
static void pred8x8_horizontal_c(uint8_t *src, int stride){
1750
    int i;
1751

    
1752
    for(i=0; i<8; i++){
1753
        ((uint32_t*)(src+i*stride))[0]=
1754
        ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1755
    }
1756
}
1757

    
1758
static void pred8x8_128_dc_c(uint8_t *src, int stride){
1759
    int i;
1760

    
1761
    for(i=0; i<4; i++){
1762
        ((uint32_t*)(src+i*stride))[0]= 
1763
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1764
    }
1765
    for(i=4; i<8; i++){
1766
        ((uint32_t*)(src+i*stride))[0]= 
1767
        ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1768
    }
1769
}
1770

    
1771
static void pred8x8_left_dc_c(uint8_t *src, int stride){
1772
    int i;
1773
    int dc0, dc2;
1774

    
1775
    dc0=dc2=0;
1776
    for(i=0;i<4; i++){
1777
        dc0+= src[-1+i*stride];
1778
        dc2+= src[-1+(i+4)*stride];
1779
    }
1780
    dc0= 0x01010101*((dc0 + 2)>>2);
1781
    dc2= 0x01010101*((dc2 + 2)>>2);
1782

    
1783
    for(i=0; i<4; i++){
1784
        ((uint32_t*)(src+i*stride))[0]=
1785
        ((uint32_t*)(src+i*stride))[1]= dc0;
1786
    }
1787
    for(i=4; i<8; i++){
1788
        ((uint32_t*)(src+i*stride))[0]=
1789
        ((uint32_t*)(src+i*stride))[1]= dc2;
1790
    }
1791
}
1792

    
1793
static void pred8x8_top_dc_c(uint8_t *src, int stride){
1794
    int i;
1795
    int dc0, dc1;
1796

    
1797
    dc0=dc1=0;
1798
    for(i=0;i<4; i++){
1799
        dc0+= src[i-stride];
1800
        dc1+= src[4+i-stride];
1801
    }
1802
    dc0= 0x01010101*((dc0 + 2)>>2);
1803
    dc1= 0x01010101*((dc1 + 2)>>2);
1804

    
1805
    for(i=0; i<4; i++){
1806
        ((uint32_t*)(src+i*stride))[0]= dc0;
1807
        ((uint32_t*)(src+i*stride))[1]= dc1;
1808
    }
1809
    for(i=4; i<8; i++){
1810
        ((uint32_t*)(src+i*stride))[0]= dc0;
1811
        ((uint32_t*)(src+i*stride))[1]= dc1;
1812
    }
1813
}
1814

    
1815

    
1816
static void pred8x8_dc_c(uint8_t *src, int stride){
1817
    int i;
1818
    int dc0, dc1, dc2, dc3;
1819

    
1820
    dc0=dc1=dc2=0;
1821
    for(i=0;i<4; i++){
1822
        dc0+= src[-1+i*stride] + src[i-stride];
1823
        dc1+= src[4+i-stride];
1824
        dc2+= src[-1+(i+4)*stride];
1825
    }
1826
    dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1827
    dc0= 0x01010101*((dc0 + 4)>>3);
1828
    dc1= 0x01010101*((dc1 + 2)>>2);
1829
    dc2= 0x01010101*((dc2 + 2)>>2);
1830

    
1831
    for(i=0; i<4; i++){
1832
        ((uint32_t*)(src+i*stride))[0]= dc0;
1833
        ((uint32_t*)(src+i*stride))[1]= dc1;
1834
    }
1835
    for(i=4; i<8; i++){
1836
        ((uint32_t*)(src+i*stride))[0]= dc2;
1837
        ((uint32_t*)(src+i*stride))[1]= dc3;
1838
    }
1839
}
1840

    
1841
static void pred8x8_plane_c(uint8_t *src, int stride){
1842
  int j, k;
1843
  int a;
1844
  uint8_t *cm = cropTbl + MAX_NEG_CROP;
1845
  const uint8_t * const src0 = src+3-stride;
1846
  const uint8_t *src1 = src+4*stride-1;
1847
  const uint8_t *src2 = src1-2*stride;      // == src+2*stride-1;
1848
  int H = src0[1] - src0[-1];
1849
  int V = src1[0] - src2[ 0];
1850
  for(k=2; k<=4; ++k) {
1851
    src1 += stride; src2 -= stride;
1852
    H += k*(src0[k] - src0[-k]);
1853
    V += k*(src1[0] - src2[ 0]);
1854
  }
1855
  H = ( 17*H+16 ) >> 5;
1856
  V = ( 17*V+16 ) >> 5;
1857

    
1858
  a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1859
  for(j=8; j>0; --j) {
1860
    int b = a;
1861
    a += V;
1862
    src[0] = cm[ (b    ) >> 5 ];
1863
    src[1] = cm[ (b+  H) >> 5 ];
1864
    src[2] = cm[ (b+2*H) >> 5 ];
1865
    src[3] = cm[ (b+3*H) >> 5 ];
1866
    src[4] = cm[ (b+4*H) >> 5 ];
1867
    src[5] = cm[ (b+5*H) >> 5 ];
1868
    src[6] = cm[ (b+6*H) >> 5 ];
1869
    src[7] = cm[ (b+7*H) >> 5 ];
1870
    src += stride;
1871
  }
1872
}
1873

    
1874
static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1875
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1876
                           int src_x_offset, int src_y_offset,
1877
                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1878
    MpegEncContext * const s = &h->s;
1879
    const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1880
    const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1881
    const int luma_xy= (mx&3) + ((my&3)<<2);
1882
    uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1883
    uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1884
    uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1885
    int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1886
    int extra_height= extra_width;
1887
    int emu=0;
1888
    const int full_mx= mx>>2;
1889
    const int full_my= my>>2;
1890
    
1891
    assert(pic->data[0]);
1892
    
1893
    if(mx&7) extra_width -= 3;
1894
    if(my&7) extra_height -= 3;
1895
    
1896
    if(   full_mx < 0-extra_width 
1897
       || full_my < 0-extra_height 
1898
       || full_mx + 16/*FIXME*/ > s->width + extra_width 
1899
       || full_my + 16/*FIXME*/ > s->height + extra_height){
1900
        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);
1901
            src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1902
        emu=1;
1903
    }
1904
    
1905
    qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1906
    if(!square){
1907
        qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1908
    }
1909
    
1910
    if(s->flags&CODEC_FLAG_GRAY) return;
1911
    
1912
    if(emu){
1913
        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);
1914
            src_cb= s->edge_emu_buffer;
1915
    }
1916
    chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1917

    
1918
    if(emu){
1919
        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);
1920
            src_cr= s->edge_emu_buffer;
1921
    }
1922
    chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1923
}
1924

    
1925
static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1926
                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1927
                           int x_offset, int y_offset,
1928
                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1929
                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1930
                           int list0, int list1){
1931
    MpegEncContext * const s = &h->s;
1932
    qpel_mc_func *qpix_op=  qpix_put;
1933
    h264_chroma_mc_func chroma_op= chroma_put;
1934
    
1935
    dest_y  += 2*x_offset + 2*y_offset*s->  linesize;
1936
    dest_cb +=   x_offset +   y_offset*s->uvlinesize;
1937
    dest_cr +=   x_offset +   y_offset*s->uvlinesize;
1938
    x_offset += 8*s->mb_x;
1939
    y_offset += 8*s->mb_y;
1940
    
1941
    if(list0){
1942
        Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1943
        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1944
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1945
                           qpix_op, chroma_op);
1946

    
1947
        qpix_op=  qpix_avg;
1948
        chroma_op= chroma_avg;
1949
    }
1950

    
1951
    if(list1){
1952
        Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1953
        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1954
                           dest_y, dest_cb, dest_cr, x_offset, y_offset,
1955
                           qpix_op, chroma_op);
1956
    }
1957
}
1958

    
1959
static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1960
                      qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1961
                      qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
1962
    MpegEncContext * const s = &h->s;
1963
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1964
    const int mb_type= s->current_picture.mb_type[mb_xy];
1965
    
1966
    assert(IS_INTER(mb_type));
1967
    
1968
    if(IS_16X16(mb_type)){
1969
        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1970
                qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1971
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1972
    }else if(IS_16X8(mb_type)){
1973
        mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1974
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1975
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1976
        mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1977
                qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1978
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1979
    }else if(IS_8X16(mb_type)){
1980
        mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
1981
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1982
                IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1983
        mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
1984
                qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1985
                IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1986
    }else{
1987
        int i;
1988
        
1989
        assert(IS_8X8(mb_type));
1990

    
1991
        for(i=0; i<4; i++){
1992
            const int sub_mb_type= h->sub_mb_type[i];
1993
            const int n= 4*i;
1994
            int x_offset= (i&1)<<2;
1995
            int y_offset= (i&2)<<1;
1996

    
1997
            if(IS_SUB_8X8(sub_mb_type)){
1998
                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1999
                    qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2000
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2001
            }else if(IS_SUB_8X4(sub_mb_type)){
2002
                mc_part(h, n  , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2003
                    qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2004
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2005
                mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
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
            }else if(IS_SUB_4X8(sub_mb_type)){
2009
                mc_part(h, n  , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2010
                    qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2011
                    IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2012
                mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, 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
            }else{
2016
                int j;
2017
                assert(IS_SUB_4X4(sub_mb_type));
2018
                for(j=0; j<4; j++){
2019
                    int sub_x_offset= x_offset + 2*(j&1);
2020
                    int sub_y_offset= y_offset +   (j&2);
2021
                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2022
                        qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2023
                        IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2024
                }
2025
            }
2026
        }
2027
    }
2028
}
2029

    
2030
static void decode_init_vlc(H264Context *h){
2031
    static int done = 0;
2032

    
2033
    if (!done) {
2034
        int i;
2035
        done = 1;
2036

    
2037
        init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5, 
2038
                 &chroma_dc_coeff_token_len [0], 1, 1,
2039
                 &chroma_dc_coeff_token_bits[0], 1, 1);
2040

    
2041
        for(i=0; i<4; i++){
2042
            init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17, 
2043
                     &coeff_token_len [i][0], 1, 1,
2044
                     &coeff_token_bits[i][0], 1, 1);
2045
        }
2046

    
2047
        for(i=0; i<3; i++){
2048
            init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2049
                     &chroma_dc_total_zeros_len [i][0], 1, 1,
2050
                     &chroma_dc_total_zeros_bits[i][0], 1, 1);
2051
        }
2052
        for(i=0; i<15; i++){
2053
            init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16, 
2054
                     &total_zeros_len [i][0], 1, 1,
2055
                     &total_zeros_bits[i][0], 1, 1);
2056
        }
2057

    
2058
        for(i=0; i<6; i++){
2059
            init_vlc(&run_vlc[i], RUN_VLC_BITS, 7, 
2060
                     &run_len [i][0], 1, 1,
2061
                     &run_bits[i][0], 1, 1);
2062
        }
2063
        init_vlc(&run7_vlc, RUN7_VLC_BITS, 16, 
2064
                 &run_len [6][0], 1, 1,
2065
                 &run_bits[6][0], 1, 1);
2066
    }
2067
}
2068

    
2069
/**
2070
 * Sets the intra prediction function pointers.
2071
 */
2072
static void init_pred_ptrs(H264Context *h){
2073
//    MpegEncContext * const s = &h->s;
2074

    
2075
    h->pred4x4[VERT_PRED           ]= pred4x4_vertical_c;
2076
    h->pred4x4[HOR_PRED            ]= pred4x4_horizontal_c;
2077
    h->pred4x4[DC_PRED             ]= pred4x4_dc_c;
2078
    h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2079
    h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2080
    h->pred4x4[VERT_RIGHT_PRED     ]= pred4x4_vertical_right_c;
2081
    h->pred4x4[HOR_DOWN_PRED       ]= pred4x4_horizontal_down_c;
2082
    h->pred4x4[VERT_LEFT_PRED      ]= pred4x4_vertical_left_c;
2083
    h->pred4x4[HOR_UP_PRED         ]= pred4x4_horizontal_up_c;
2084
    h->pred4x4[LEFT_DC_PRED        ]= pred4x4_left_dc_c;
2085
    h->pred4x4[TOP_DC_PRED         ]= pred4x4_top_dc_c;
2086
    h->pred4x4[DC_128_PRED         ]= pred4x4_128_dc_c;
2087

    
2088
    h->pred8x8[DC_PRED8x8     ]= pred8x8_dc_c;
2089
    h->pred8x8[VERT_PRED8x8   ]= pred8x8_vertical_c;
2090
    h->pred8x8[HOR_PRED8x8    ]= pred8x8_horizontal_c;
2091
    h->pred8x8[PLANE_PRED8x8  ]= pred8x8_plane_c;
2092
    h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2093
    h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2094
    h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2095

    
2096
    h->pred16x16[DC_PRED8x8     ]= pred16x16_dc_c;
2097
    h->pred16x16[VERT_PRED8x8   ]= pred16x16_vertical_c;
2098
    h->pred16x16[HOR_PRED8x8    ]= pred16x16_horizontal_c;
2099
    h->pred16x16[PLANE_PRED8x8  ]= pred16x16_plane_c;
2100
    h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2101
    h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2102
    h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2103
}
2104

    
2105
static void free_tables(H264Context *h){
2106
    av_freep(&h->intra4x4_pred_mode);
2107
    av_freep(&h->non_zero_count);
2108
    av_freep(&h->slice_table_base);
2109
    h->slice_table= NULL;
2110
    
2111
    av_freep(&h->mb2b_xy);
2112
    av_freep(&h->mb2b8_xy);
2113
}
2114

    
2115
/**
2116
 * allocates tables.
2117
 * needs widzh/height
2118
 */
2119
static int alloc_tables(H264Context *h){
2120
    MpegEncContext * const s = &h->s;
2121
    const int big_mb_num= s->mb_stride * (s->mb_height+1);
2122
    int x,y;
2123

    
2124
    CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8  * sizeof(uint8_t))
2125
    CHECKED_ALLOCZ(h->non_zero_count    , big_mb_num * 16 * sizeof(uint8_t))
2126
    CHECKED_ALLOCZ(h->slice_table_base  , big_mb_num * sizeof(uint8_t))
2127

    
2128
    memset(h->slice_table_base, -1, big_mb_num  * sizeof(uint8_t));
2129
    h->slice_table= h->slice_table_base + s->mb_stride + 1;
2130

    
2131
    CHECKED_ALLOCZ(h->mb2b_xy  , big_mb_num * sizeof(uint16_t));
2132
    CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2133
    for(y=0; y<s->mb_height; y++){
2134
        for(x=0; x<s->mb_width; x++){
2135
            const int mb_xy= x + y*s->mb_stride;
2136
            const int b_xy = 4*x + 4*y*h->b_stride;
2137
            const int b8_xy= 2*x + 2*y*h->b8_stride;
2138
        
2139
            h->mb2b_xy [mb_xy]= b_xy;
2140
            h->mb2b8_xy[mb_xy]= b8_xy;
2141
        }
2142
    }
2143
    
2144
    return 0;
2145
fail:
2146
    free_tables(h);
2147
    return -1;
2148
}
2149

    
2150
static void common_init(H264Context *h){
2151
    MpegEncContext * const s = &h->s;
2152

    
2153
    s->width = s->avctx->width;
2154
    s->height = s->avctx->height;
2155
    s->codec_id= s->avctx->codec->id;
2156
    
2157
    init_pred_ptrs(h);
2158

    
2159
    s->decode=1; //FIXME
2160
}
2161

    
2162
static int decode_init(AVCodecContext *avctx){
2163
    H264Context *h= avctx->priv_data;
2164
    MpegEncContext * const s = &h->s;
2165

    
2166
    s->avctx = avctx;
2167
    common_init(h);
2168

    
2169
    s->out_format = FMT_H264;
2170
    s->workaround_bugs= avctx->workaround_bugs;
2171

    
2172
    // set defaults
2173
    s->progressive_sequence=1;
2174
//    s->decode_mb= ff_h263_decode_mb;
2175
    s->low_delay= 1;
2176
    avctx->pix_fmt= PIX_FMT_YUV420P;
2177

    
2178
    decode_init_vlc(h);
2179
    
2180
    return 0;
2181
}
2182

    
2183
static void frame_start(H264Context *h){
2184
    MpegEncContext * const s = &h->s;
2185
    int i;
2186

    
2187
    MPV_frame_start(s, s->avctx);
2188
    ff_er_frame_start(s);
2189
    h->mmco_index=0;
2190

    
2191
    assert(s->linesize && s->uvlinesize);
2192

    
2193
    for(i=0; i<16; i++){
2194
        h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2195
        h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2196
    }
2197
    for(i=0; i<4; i++){
2198
        h->block_offset[16+i]=
2199
        h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2200
    }
2201

    
2202
//    s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2203
}
2204

    
2205
static void hl_decode_mb(H264Context *h){
2206
    MpegEncContext * const s = &h->s;
2207
    const int mb_x= s->mb_x;
2208
    const int mb_y= s->mb_y;
2209
    const int mb_xy= mb_x + mb_y*s->mb_stride;
2210
    const int mb_type= s->current_picture.mb_type[mb_xy];
2211
    uint8_t  *dest_y, *dest_cb, *dest_cr;
2212
    int linesize, uvlinesize /*dct_offset*/;
2213
    int i;
2214

    
2215
    if(!s->decode)
2216
        return;
2217

    
2218
    if(s->mb_skiped){
2219
    }
2220

    
2221
    dest_y  = s->current_picture.data[0] + (mb_y * 16* s->linesize  ) + mb_x * 16;
2222
    dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2223
    dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2224

    
2225
    if (h->mb_field_decoding_flag) {
2226
        linesize = s->linesize * 2;
2227
        uvlinesize = s->uvlinesize * 2;
2228
        if(mb_y&1){ //FIXME move out of this func?
2229
            dest_y -= s->linesize*15;
2230
            dest_cb-= s->linesize*7;
2231
            dest_cr-= s->linesize*7;
2232
        }
2233
    } else {
2234
        linesize = s->linesize;
2235
        uvlinesize = s->uvlinesize;
2236
//        dct_offset = s->linesize * 16;
2237
    }
2238

    
2239
    if(IS_INTRA(mb_type)){
2240
        if(!(s->flags&CODEC_FLAG_GRAY)){
2241
            h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2242
            h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2243
        }
2244

    
2245
        if(IS_INTRA4x4(mb_type)){
2246
            if(!s->encoding){
2247
                for(i=0; i<16; i++){
2248
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2249
                    uint8_t *topright= ptr + 4 - linesize;
2250
                    const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2251
                    const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2252
                    int tr;
2253

    
2254
                    if(!topright_avail){
2255
                        tr= ptr[3 - linesize]*0x01010101;
2256
                        topright= (uint8_t*) &tr;
2257
                    }
2258

    
2259
                    h->pred4x4[ dir ](ptr, topright, linesize);
2260
                    if(h->non_zero_count_cache[ scan8[i] ]){
2261
                        if(s->codec_id == CODEC_ID_H264)
2262
                            h264_add_idct_c(ptr, h->mb + i*16, linesize);
2263
                        else
2264
                            svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2265
                    }
2266
                }
2267
            }
2268
        }else{
2269
            h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2270
            if(s->codec_id == CODEC_ID_H264)
2271
                h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2272
            else
2273
                svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2274
        }
2275
    }else if(s->codec_id == CODEC_ID_H264){
2276
        hl_motion(h, dest_y, dest_cb, dest_cr,
2277
                  s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab, 
2278
                  s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2279
    }
2280

    
2281

    
2282
    if(!IS_INTRA4x4(mb_type)){
2283
        if(s->codec_id == CODEC_ID_H264){
2284
            for(i=0; i<16; i++){
2285
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2286
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2287
                    h264_add_idct_c(ptr, h->mb + i*16, linesize);
2288
                }
2289
            }
2290
        }else{
2291
            for(i=0; i<16; i++){
2292
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2293
                    uint8_t * const ptr= dest_y + h->block_offset[i];
2294
                    svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2295
                }
2296
            }
2297
        }
2298
    }
2299

    
2300
    if(!(s->flags&CODEC_FLAG_GRAY)){
2301
        chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2302
        chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2303
        if(s->codec_id == CODEC_ID_H264){
2304
            for(i=16; i<16+4; i++){
2305
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2306
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2307
                    h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2308
                }
2309
            }
2310
            for(i=20; i<20+4; i++){
2311
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2312
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2313
                    h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2314
                }
2315
            }
2316
        }else{
2317
            for(i=16; i<16+4; i++){
2318
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2319
                    uint8_t * const ptr= dest_cb + h->block_offset[i];
2320
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2321
                }
2322
            }
2323
            for(i=20; i<20+4; i++){
2324
                if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2325
                    uint8_t * const ptr= dest_cr + h->block_offset[i];
2326
                    svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2327
                }
2328
            }
2329
        }
2330
    }
2331
}
2332

    
2333
static void decode_mb_cabac(H264Context *h){
2334
//    MpegEncContext * const s = &h->s;
2335
}
2336

    
2337
/**
2338
 * fills the default_ref_list.
2339
 */
2340
static int fill_default_ref_list(H264Context *h){
2341
    MpegEncContext * const s = &h->s;
2342
    int i;
2343
    Picture sorted_short_ref[16];
2344
    
2345
    if(h->slice_type==B_TYPE){
2346
        int out_i;
2347
        int limit= -1;
2348

    
2349
        for(out_i=0; out_i<h->short_ref_count; out_i++){
2350
            int best_i=-1;
2351
            int best_poc=-1;
2352

    
2353
            for(i=0; i<h->short_ref_count; i++){
2354
                const int poc= h->short_ref[i]->poc;
2355
                if(poc > limit && poc < best_poc){
2356
                    best_poc= poc;
2357
                    best_i= i;
2358
                }
2359
            }
2360
            
2361
            assert(best_i != -1);
2362
            
2363
            limit= best_poc;
2364
            sorted_short_ref[out_i]= *h->short_ref[best_i];
2365
        }
2366
    }
2367

    
2368
    if(s->picture_structure == PICT_FRAME){
2369
        if(h->slice_type==B_TYPE){
2370
            const int current_poc= s->current_picture_ptr->poc;
2371
            int list;
2372

    
2373
            for(list=0; list<2; list++){
2374
                int index=0;
2375

    
2376
                for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2377
                    const int i2= list ? h->short_ref_count - i - 1 : i;
2378
                    const int poc= sorted_short_ref[i2].poc;
2379
                    
2380
                    if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2381

    
2382
                    if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2383
                        h->default_ref_list[list][index  ]= sorted_short_ref[i2];
2384
                        h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2385
                    }
2386
                }
2387

    
2388
                for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2389
                    if(h->long_ref[i]->reference != 3) continue;
2390

    
2391
                    h->default_ref_list[ list ][index  ]= *h->long_ref[i];
2392
                    h->default_ref_list[ list ][index++].pic_id= i;;
2393
                }
2394
                
2395
                if(h->long_ref_count > 1 && h->short_ref_count==0){
2396
                    Picture temp= h->default_ref_list[1][0];
2397
                    h->default_ref_list[1][0] = h->default_ref_list[1][1];
2398
                    h->default_ref_list[1][0] = temp;
2399
                }
2400

    
2401
                if(index < h->ref_count[ list ])
2402
                    memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2403
            }
2404
        }else{
2405
            int index=0;
2406
            for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2407
                if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2408
                h->default_ref_list[0][index  ]= *h->short_ref[i];
2409
                h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2410
            }
2411
            for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2412
                if(h->long_ref[i]->reference != 3) continue;
2413
                h->default_ref_list[0][index  ]= *h->long_ref[i];
2414
                h->default_ref_list[0][index++].pic_id= i;;
2415
            }
2416
            if(index < h->ref_count[0])
2417
                memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2418
        }
2419
    }else{ //FIELD
2420
        if(h->slice_type==B_TYPE){
2421
        }else{
2422
            //FIXME second field balh
2423
        }
2424
    }
2425
    return 0;
2426
}
2427

    
2428
static int decode_ref_pic_list_reordering(H264Context *h){
2429
    MpegEncContext * const s = &h->s;
2430
    int list;
2431
    
2432
    if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2433
    
2434
    for(list=0; list<2; list++){
2435
        memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2436

    
2437
        if(get_bits1(&s->gb)){
2438
            int pred= h->curr_pic_num;
2439
            int index;
2440

    
2441
            for(index=0; ; index++){
2442
                int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2443
                int pic_id;
2444
                int i;
2445
                
2446
                
2447
                if(index >= h->ref_count[list]){
2448
                    fprintf(stderr, "reference count overflow\n");
2449
                    return -1;
2450
                }
2451
                
2452
                if(reordering_of_pic_nums_idc<3){
2453
                    if(reordering_of_pic_nums_idc<2){
2454
                        const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2455

    
2456
                        if(abs_diff_pic_num >= h->max_pic_num){
2457
                            fprintf(stderr, "abs_diff_pic_num overflow\n");
2458
                            return -1;
2459
                        }
2460

    
2461
                        if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2462
                        else                                pred+= abs_diff_pic_num;
2463
                        pred &= h->max_pic_num - 1;
2464
                    
2465
                        for(i= h->ref_count[list]-1; i>=index; i--){
2466
                            if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2467
                                break;
2468
                        }
2469
                    }else{
2470
                        pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2471

    
2472
                        for(i= h->ref_count[list]-1; i>=index; i--){
2473
                            if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2474
                                break;
2475
                        }
2476
                    }
2477

    
2478
                    if(i < index){
2479
                        fprintf(stderr, "reference picture missing during reorder\n");
2480
                        memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2481
                    }else if(i > index){
2482
                        Picture tmp= h->ref_list[list][i];
2483
                        for(; i>index; i--){
2484
                            h->ref_list[list][i]= h->ref_list[list][i-1];
2485
                        }
2486
                        h->ref_list[list][index]= tmp;
2487
                    }
2488
                }else if(reordering_of_pic_nums_idc==3) 
2489
                    break;
2490
                else{
2491
                    fprintf(stderr, "illegal reordering_of_pic_nums_idc\n");
2492
                    return -1;
2493
                }
2494
            }
2495
        }
2496

    
2497
        if(h->slice_type!=B_TYPE) break;
2498
    }
2499
    return 0;    
2500
}
2501

    
2502
static int pred_weight_table(H264Context *h){
2503
    MpegEncContext * const s = &h->s;
2504
    int list, i;
2505
    
2506
    h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2507
    h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2508

    
2509
    for(list=0; list<2; list++){
2510
        for(i=0; i<h->ref_count[list]; i++){
2511
            int luma_weight_flag, chroma_weight_flag;
2512
            
2513
            luma_weight_flag= get_bits1(&s->gb);
2514
            if(luma_weight_flag){
2515
                h->luma_weight[list][i]= get_se_golomb(&s->gb);
2516
                h->luma_offset[list][i]= get_se_golomb(&s->gb);
2517
            }
2518

    
2519
            chroma_weight_flag= get_bits1(&s->gb);
2520
            if(chroma_weight_flag){
2521
                int j;
2522
                for(j=0; j<2; j++){
2523
                    h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2524
                    h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2525
                }
2526
            }
2527
        }
2528
        if(h->slice_type != B_TYPE) break;
2529
    }
2530
    return 0;
2531
}
2532

    
2533
/**
2534
 * instantaneos decoder refresh.
2535
 */
2536
static void idr(H264Context *h){
2537
    int i;
2538

    
2539
    for(i=0; i<h->long_ref_count; i++){
2540
        h->long_ref[i]->reference=0;
2541
        h->long_ref[i]= NULL;
2542
    }
2543
    h->long_ref_count=0;
2544

    
2545
    for(i=0; i<h->short_ref_count; i++){
2546
        h->short_ref[i]->reference=0;
2547
        h->short_ref[i]= NULL;
2548
    }
2549
    h->short_ref_count=0;
2550
}
2551

    
2552
/**
2553
 *
2554
 * @return the removed picture or NULL if an error occures
2555
 */
2556
static Picture * remove_short(H264Context *h, int frame_num){
2557
    MpegEncContext * const s = &h->s;
2558
    int i;
2559
    
2560
    if(s->avctx->debug&FF_DEBUG_MMCO)
2561
        printf("remove short %d count %d\n", frame_num, h->short_ref_count);
2562
    
2563
    for(i=0; i<h->short_ref_count; i++){
2564
        Picture *pic= h->short_ref[i];
2565
        if(s->avctx->debug&FF_DEBUG_MMCO)
2566
            printf("%d %d %p\n", i, pic->frame_num, pic);
2567
        if(pic->frame_num == frame_num){
2568
            h->short_ref[i]= NULL;
2569
            memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2570
            h->short_ref_count--;
2571
            return pic;
2572
        }
2573
    }
2574
    return NULL;
2575
}
2576

    
2577
/**
2578
 *
2579
 * @return the removed picture or NULL if an error occures
2580
 */
2581
static Picture * remove_long(H264Context *h, int i){
2582
    Picture *pic;
2583

    
2584
    if(i >= h->long_ref_count) return NULL;
2585
    pic= h->long_ref[i];
2586
    if(pic==NULL) return NULL;
2587
    
2588
    h->long_ref[i]= NULL;
2589
    memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2590
    h->long_ref_count--;
2591

    
2592
    return pic;
2593
}
2594

    
2595
/**
2596
 * Executes the reference picture marking (memory management control operations).
2597
 */
2598
static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2599
    MpegEncContext * const s = &h->s;
2600
    int i;
2601
    int current_is_long=0;
2602
    Picture *pic;
2603
    
2604
    if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2605
        printf("no mmco here\n");
2606
        
2607
    for(i=0; i<mmco_count; i++){
2608
        if(s->avctx->debug&FF_DEBUG_MMCO)
2609
            printf("mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
2610

    
2611
        switch(mmco[i].opcode){
2612
        case MMCO_SHORT2UNUSED:
2613
            pic= remove_short(h, mmco[i].short_frame_num);
2614
            if(pic==NULL) return -1;
2615
            pic->reference= 0;
2616
            break;
2617
        case MMCO_SHORT2LONG:
2618
            pic= remove_long(h, mmco[i].long_index);
2619
            if(pic) pic->reference=0;
2620
            
2621
            h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2622
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
2623
            break;
2624
        case MMCO_LONG2UNUSED:
2625
            pic= remove_long(h, mmco[i].long_index);
2626
            if(pic==NULL) return -1;
2627
            pic->reference= 0;
2628
            break;
2629
        case MMCO_LONG:
2630
            pic= remove_long(h, mmco[i].long_index);
2631
            if(pic) pic->reference=0;
2632
            
2633
            h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2634
            h->long_ref[ mmco[i].long_index ]->long_ref=1;
2635
            h->long_ref_count++;
2636
            
2637
            current_is_long=1;
2638
            break;
2639
        case MMCO_SET_MAX_LONG:
2640
            assert(mmco[i].long_index <= 16);
2641
            while(mmco[i].long_index < h->long_ref_count){
2642
                pic= remove_long(h, mmco[i].long_index);
2643
                pic->reference=0;
2644
            }
2645
            while(mmco[i].long_index > h->long_ref_count){
2646
                h->long_ref[ h->long_ref_count++ ]= NULL;
2647
            }
2648
            break;
2649
        case MMCO_RESET:
2650
            while(h->short_ref_count){
2651
                pic= remove_short(h, h->short_ref[0]->frame_num);
2652
                pic->reference=0;
2653
            }
2654
            while(h->long_ref_count){
2655
                pic= remove_long(h, h->long_ref_count-1);
2656
                pic->reference=0;
2657
            }
2658
            break;
2659
        default: assert(0);
2660
        }
2661
    }
2662
    
2663
    if(!current_is_long){
2664
        pic= remove_short(h, s->current_picture_ptr->frame_num);
2665
        if(pic){
2666
            pic->reference=0;
2667
            fprintf(stderr, "illegal short term buffer state detected\n");
2668
        }
2669
        
2670
        if(h->short_ref_count)
2671
            memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2672

    
2673
        h->short_ref[0]= s->current_picture_ptr;
2674
        h->short_ref[0]->long_ref=0;
2675
        h->short_ref_count++;
2676
    }
2677
    
2678
    return 0; 
2679
}
2680

    
2681
static int decode_ref_pic_marking(H264Context *h){
2682
    MpegEncContext * const s = &h->s;
2683
    int i;
2684
    
2685
    if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2686
        s->broken_link= get_bits1(&s->gb) -1;
2687
        h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2688
        if(h->mmco[0].long_index == -1)
2689
            h->mmco_index= 0;
2690
        else{
2691
            h->mmco[0].opcode= MMCO_LONG;
2692
            h->mmco_index= 1;
2693
        } 
2694
    }else{
2695
        if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2696
            for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) { 
2697
                MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2698

    
2699
                h->mmco[i].opcode= opcode;
2700
                if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2701
                    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
2702
/*                    if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2703
                        fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2704
                        return -1;
2705
                    }*/
2706
                }
2707
                if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2708
                    h->mmco[i].long_index= get_ue_golomb(&s->gb);
2709
                    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){
2710
                        fprintf(stderr, "illegal long ref in memory management control operation %d\n", opcode);
2711
                        return -1;
2712
                    }
2713
                }
2714
                    
2715
                if(opcode > MMCO_LONG){
2716
                    fprintf(stderr, "illegal memory management control operation %d\n", opcode);
2717
                    return -1;
2718
                }
2719
            }
2720
            h->mmco_index= i;
2721
        }else{
2722
            assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2723

    
2724
            if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2725
                h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2726
                h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2727
                h->mmco_index= 1;
2728
            }else
2729
                h->mmco_index= 0;
2730
        }
2731
    }
2732
    
2733
    return 0; 
2734
}
2735

    
2736
static int init_poc(H264Context *h){
2737
    MpegEncContext * const s = &h->s;
2738
    const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2739
    int field_poc[2];
2740

    
2741
    if(h->nal_unit_type == NAL_IDR_SLICE){
2742
        h->frame_num_offset= 0;
2743
    }else{
2744
        if(h->frame_num < h->prev_frame_num)
2745
            h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2746
        else
2747
            h->frame_num_offset= h->prev_frame_num_offset;
2748
    }
2749

    
2750
    if(h->sps.poc_type==0){
2751
        const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2752

    
2753
        if     (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2754
            h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2755
        else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2756
            h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2757
        else
2758
            h->poc_msb = h->prev_poc_msb;
2759
//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2760
        field_poc[0] = 
2761
        field_poc[1] = h->poc_msb + h->poc_lsb;
2762
        if(s->picture_structure == PICT_FRAME) 
2763
            field_poc[1] += h->delta_poc_bottom;
2764
    }else if(h->sps.poc_type==1){
2765
        int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2766
        int i;
2767

    
2768
        if(h->sps.poc_cycle_length != 0)
2769
            abs_frame_num = h->frame_num_offset + h->frame_num;
2770
        else
2771
            abs_frame_num = 0;
2772

    
2773
        if(h->nal_ref_idc==0 && abs_frame_num > 0)
2774
            abs_frame_num--;
2775
            
2776
        expected_delta_per_poc_cycle = 0;
2777
        for(i=0; i < h->sps.poc_cycle_length; i++)
2778
            expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2779

    
2780
        if(abs_frame_num > 0){
2781
            int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2782
            int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2783

    
2784
            expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2785
            for(i = 0; i <= frame_num_in_poc_cycle; i++)
2786
                expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2787
        } else
2788
            expectedpoc = 0;
2789

    
2790
        if(h->nal_ref_idc == 0) 
2791
            expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2792
        
2793
        field_poc[0] = expectedpoc + h->delta_poc[0];
2794
        field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2795

    
2796
        if(s->picture_structure == PICT_FRAME)
2797
            field_poc[1] += h->delta_poc[1];
2798
    }else{
2799
        int poc;
2800
        if(h->nal_unit_type == NAL_IDR_SLICE){
2801
            poc= 0;
2802
        }else{
2803
            if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2804
            else               poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2805
        }
2806
        field_poc[0]= poc;
2807
        field_poc[1]= poc;
2808
    }
2809
    
2810
    if(s->picture_structure != PICT_BOTTOM_FIELD)
2811
        s->current_picture_ptr->field_poc[0]= field_poc[0];
2812
    if(s->picture_structure != PICT_TOP_FIELD)
2813
        s->current_picture_ptr->field_poc[1]= field_poc[1];
2814
    if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2815
        s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2816

    
2817
    return 0;
2818
}
2819

    
2820
/**
2821
 * decodes a slice header.
2822
 * this will allso call MPV_common_init() and frame_start() as needed
2823
 */
2824
static int decode_slice_header(H264Context *h){
2825
    MpegEncContext * const s = &h->s;
2826
    int first_mb_in_slice, pps_id;
2827
    int num_ref_idx_active_override_flag;
2828
    static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2829
    float new_aspect;
2830

    
2831
    s->current_picture.reference= h->nal_ref_idc != 0;
2832

    
2833
    first_mb_in_slice= get_ue_golomb(&s->gb);
2834

    
2835
    h->slice_type= get_ue_golomb(&s->gb);
2836
    if(h->slice_type > 9){
2837
        fprintf(stderr, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
2838
    }
2839
    if(h->slice_type > 4){
2840
        h->slice_type -= 5;
2841
        h->slice_type_fixed=1;
2842
    }else
2843
        h->slice_type_fixed=0;
2844
    
2845
    h->slice_type= slice_type_map[ h->slice_type ];
2846
    
2847
    s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
2848
        
2849
    pps_id= get_ue_golomb(&s->gb);
2850
    if(pps_id>255){
2851
        fprintf(stderr, "pps_id out of range\n");
2852
        return -1;
2853
    }
2854
    h->pps= h->pps_buffer[pps_id];
2855
    if(h->pps.slice_group_count == 0){
2856
        fprintf(stderr, "non existing PPS referenced\n");
2857
        return -1;
2858
    }
2859

    
2860
    h->sps= h->sps_buffer[ h->pps.sps_id ];
2861
    if(h->sps.log2_max_frame_num == 0){
2862
        fprintf(stderr, "non existing SPS referenced\n");
2863
        return -1;
2864
    }
2865
    
2866
    s->mb_width= h->sps.mb_width;
2867
    s->mb_height= h->sps.mb_height;
2868
    
2869
    h->b_stride=  s->mb_width*4;
2870
    h->b8_stride= s->mb_width*2;
2871

    
2872
    s->mb_x = first_mb_in_slice % s->mb_width;
2873
    s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
2874
    
2875
    s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
2876
    if(h->sps.frame_mbs_only_flag)
2877
        s->height= 16*s->mb_height - 2*(h->sps.crop_top  + h->sps.crop_bottom);
2878
    else
2879
        s->height= 16*s->mb_height - 4*(h->sps.crop_top  + h->sps.crop_bottom); //FIXME recheck
2880
    
2881
    if(s->aspected_height) //FIXME emms at end of slice ?
2882
        new_aspect= h->sps.sar_width*s->width / (float)(s->height*h->sps.sar_height);
2883
    else
2884
        new_aspect=0;
2885

    
2886
    if (s->context_initialized 
2887
        && (   s->width != s->avctx->width || s->height != s->avctx->height 
2888
            || ABS(new_aspect - s->avctx->aspect_ratio) > 0.001)) {
2889
        free_tables(h);
2890
        MPV_common_end(s);
2891
    }
2892
    if (!s->context_initialized) {
2893
        if (MPV_common_init(s) < 0)
2894
            return -1;
2895

    
2896
        alloc_tables(h);
2897

    
2898
        s->avctx->width = s->width;
2899
        s->avctx->height = s->height;
2900
        s->avctx->aspect_ratio= new_aspect;
2901
    }
2902

    
2903
    if(first_mb_in_slice == 0){
2904
        frame_start(h);
2905
    }
2906

    
2907
    s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
2908
    h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2909

    
2910
    if(h->sps.frame_mbs_only_flag){
2911
        s->picture_structure= PICT_FRAME;
2912
    }else{
2913
        if(get_bits1(&s->gb)) //field_pic_flag
2914
            s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2915
        else
2916
            s->picture_structure= PICT_FRAME;
2917
    }
2918

    
2919
    if(s->picture_structure==PICT_FRAME){
2920
        h->curr_pic_num=   h->frame_num;
2921
        h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2922
    }else{
2923
        h->curr_pic_num= 2*h->frame_num;
2924
        h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2925
    }
2926
        
2927
    if(h->nal_unit_type == NAL_IDR_SLICE){
2928
        int idr_pic_id= get_ue_golomb(&s->gb);
2929
    }
2930
   
2931
    if(h->sps.poc_type==0){
2932
        h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2933
        
2934
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2935
            h->delta_poc_bottom= get_se_golomb(&s->gb);
2936
        }
2937
    }
2938
    
2939
    if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2940
        h->delta_poc[0]= get_se_golomb(&s->gb);
2941
        
2942
        if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2943
            h->delta_poc[1]= get_se_golomb(&s->gb);
2944
    }
2945
    
2946
    init_poc(h);
2947
    
2948
    if(h->pps.redundant_pic_cnt_present){
2949
        h->redundant_pic_count= get_ue_golomb(&s->gb);
2950
    }
2951

    
2952
    //set defaults, might be overriden a few line later
2953
    h->ref_count[0]= h->pps.ref_count[0];
2954
    h->ref_count[1]= h->pps.ref_count[1];
2955

    
2956
    if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
2957
        if(h->slice_type == B_TYPE){
2958
            h->direct_spatial_mv_pred= get_bits1(&s->gb);
2959
        }
2960
        num_ref_idx_active_override_flag= get_bits1(&s->gb);
2961
    
2962
        if(num_ref_idx_active_override_flag){
2963
            h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2964
            if(h->slice_type==B_TYPE)
2965
                h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2966

    
2967
            if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
2968
                fprintf(stderr, "reference overflow\n");
2969
                return -1;
2970
            }
2971
        }
2972
    }
2973

    
2974
    if(first_mb_in_slice == 0){
2975
        fill_default_ref_list(h);
2976
    }
2977

    
2978
    decode_ref_pic_list_reordering(h);
2979

    
2980
    if(   (h->pps.weighted_pred          && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE )) 
2981
       || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
2982
        pred_weight_table(h);
2983
    
2984
    if(s->current_picture.reference)
2985
        decode_ref_pic_marking(h);
2986
    //FIXME CABAC stuff
2987

    
2988
    s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta
2989
    //FIXME qscale / qp ... stuff
2990
    if(h->slice_type == SP_TYPE){
2991
        int sp_for_switch_flag= get_bits1(&s->gb);
2992
    }
2993
    if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
2994
        int slice_qs_delta= get_se_golomb(&s->gb);
2995
    }
2996

    
2997
    if( h->pps.deblocking_filter_parameters_present ) {
2998
        h->disable_deblocking_filter_idc= get_ue_golomb(&s->gb);
2999
        if( h->disable_deblocking_filter_idc  !=  1 ) {
3000
            h->slice_alpha_c0_offset_div2= get_se_golomb(&s->gb);
3001
            h->slice_beta_offset_div2= get_se_golomb(&s->gb);
3002
        }
3003
    }else
3004
        h->disable_deblocking_filter_idc= 0;
3005

    
3006
#if 0 //FMO
3007
    if( h->pps.num_slice_groups > 1  && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3008
        slice_group_change_cycle= get_bits(&s->gb, ?);
3009
#endif
3010

    
3011
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3012
        printf("mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n", 
3013
               first_mb_in_slice, 
3014
               av_get_pict_type_char(h->slice_type),
3015
               pps_id, h->frame_num,
3016
               s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3017
               h->ref_count[0], h->ref_count[1],
3018
               s->qscale,
3019
               h->disable_deblocking_filter_idc
3020
               );
3021
    }
3022

    
3023
    return 0;
3024
}
3025

    
3026
/**
3027
 *
3028
 */
3029
static inline int get_level_prefix(GetBitContext *gb){
3030
    unsigned int buf;
3031
    int log;
3032
    
3033
    OPEN_READER(re, gb);
3034
    UPDATE_CACHE(re, gb);
3035
    buf=GET_CACHE(re, gb);
3036
    
3037
    log= 32 - av_log2(buf);
3038
#ifdef TRACE
3039
    print_bin(buf>>(32-log), log);
3040
    printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3041
#endif
3042

    
3043
    LAST_SKIP_BITS(re, gb, log);
3044
    CLOSE_READER(re, gb);
3045

    
3046
    return log-1;
3047
}
3048

    
3049
/**
3050
 * decodes a residual block.
3051
 * @param n block index
3052
 * @param scantable scantable
3053
 * @param max_coeff number of coefficients in the block
3054
 * @return <0 if an error occured
3055
 */
3056
static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3057
    MpegEncContext * const s = &h->s;
3058
    const uint16_t *qmul= dequant_coeff[qp];
3059
    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};
3060
    int level[16], run[16];
3061
    int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3062

    
3063
    //FIXME put trailing_onex into the context
3064

    
3065
    if(n == CHROMA_DC_BLOCK_INDEX){
3066
        coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3067
        total_coeff= coeff_token>>2;
3068
    }else{    
3069
        if(n == LUMA_DC_BLOCK_INDEX){
3070
            total_coeff= pred_non_zero_count(h, 0);
3071
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3072
            total_coeff= coeff_token>>2;
3073
        }else{
3074
            total_coeff= pred_non_zero_count(h, n);
3075
            coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3076
            total_coeff= coeff_token>>2;
3077
            h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3078
        }
3079
    }
3080

    
3081
    //FIXME set last_non_zero?
3082

    
3083
    if(total_coeff==0)
3084
        return 0;
3085
        
3086
    trailing_ones= coeff_token&3;
3087
    tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3088
    assert(total_coeff<=16);
3089
    
3090
    for(i=0; i<trailing_ones; i++){
3091
        level[i]= 1 - 2*get_bits1(gb);
3092
    }
3093

    
3094
    suffix_length= total_coeff > 10 && trailing_ones < 3;
3095

    
3096
    for(; i<total_coeff; i++){
3097
        const int prefix= get_level_prefix(gb);
3098
        int level_code, mask;
3099

    
3100
        if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3101
            if(suffix_length)
3102
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3103
            else
3104
                level_code= (prefix<<suffix_length); //part
3105
        }else if(prefix==14){
3106
            if(suffix_length)
3107
                level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3108
            else
3109
                level_code= prefix + get_bits(gb, 4); //part
3110
        }else if(prefix==15){
3111
            level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3112
            if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3113
        }else{
3114
            fprintf(stderr, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3115
            return -1;
3116
        }
3117

    
3118
        if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3119

    
3120
        mask= -(level_code&1);
3121
        level[i]= (((2+level_code)>>1) ^ mask) - mask;
3122

    
3123
        if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3124

    
3125
#if 1
3126
        if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3127
#else        
3128
        if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3129
        ? == prefix > 2 or sth
3130
#endif
3131
        tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3132
    }
3133

    
3134
    if(total_coeff == max_coeff)
3135
        zeros_left=0;
3136
    else{
3137
        if(n == CHROMA_DC_BLOCK_INDEX)
3138
            zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3139
        else
3140
            zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3141
    }
3142
    
3143
    for(i=0; i<total_coeff-1; i++){
3144
        if(zeros_left <=0)
3145
            break;
3146
        else if(zeros_left < 7){
3147
            run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3148
        }else{
3149
            run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3150
        }
3151
        zeros_left -= run[i];
3152
    }
3153

    
3154
    if(zeros_left<0){
3155
        fprintf(stderr, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3156
        return -1;
3157
    }
3158
    
3159
    for(; i<total_coeff-1; i++){
3160
        run[i]= 0;
3161
    }
3162

    
3163
    run[i]= zeros_left;
3164

    
3165
    coeff_num=-1;
3166
    if(n > 24){
3167
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3168
            int j;
3169

    
3170
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3171
            j= scantable[ coeff_num ];
3172

    
3173
            block[j]= level[i];
3174
        }
3175
    }else{
3176
        for(i=total_coeff-1; i>=0; i--){ //FIXME merge into  rundecode?
3177
            int j;
3178

    
3179
            coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3180
            j= scantable[ coeff_num ];
3181

    
3182
            block[j]= level[i] * qmul[j];
3183
//            printf("%d %d  ", block[j], qmul[j]);
3184
        }
3185
    }
3186
    return 0;
3187
}
3188

    
3189
/**
3190
 * decodes a macroblock
3191
 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3192
 */
3193
static int decode_mb(H264Context *h){
3194
    MpegEncContext * const s = &h->s;
3195
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3196
    int mb_type, partition_count, cbp;
3197

    
3198
    s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?    
3199

    
3200
    tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3201

    
3202
    if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3203
        if(s->mb_skip_run==-1)
3204
            s->mb_skip_run= get_ue_golomb(&s->gb);
3205
        
3206
        if (s->mb_skip_run--) {
3207
            int mx, my;
3208
            /* skip mb */
3209
//FIXME b frame
3210
            mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3211

    
3212
            memset(h->non_zero_count[mb_xy], 0, 16);
3213
            memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3214

    
3215
            if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3216
                h->mb_field_decoding_flag= get_bits1(&s->gb);
3217
            }
3218

    
3219
            if(h->mb_field_decoding_flag)
3220
                mb_type|= MB_TYPE_INTERLACED;
3221
            
3222
            fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3223
            pred_pskip_motion(h, &mx, &my);
3224
            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3225
            fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3226
            write_back_motion(h, mb_type);
3227

    
3228
            s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3229
            h->slice_table[ mb_xy ]= h->slice_num;
3230

    
3231
            h->prev_mb_skiped= 1;
3232
            return 0;
3233
        }
3234
    }
3235
    if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3236
        if((s->mb_y&1)==0)
3237
            h->mb_field_decoding_flag = get_bits1(&s->gb);
3238
    }else
3239
        h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3240
    
3241
    h->prev_mb_skiped= 0;
3242
    
3243
    mb_type= get_ue_golomb(&s->gb);
3244
    if(h->slice_type == B_TYPE){
3245
        if(mb_type < 23){
3246
            partition_count= b_mb_type_info[mb_type].partition_count;
3247
            mb_type=         b_mb_type_info[mb_type].type;
3248
        }else{
3249
            mb_type -= 23;
3250
            goto decode_intra_mb;
3251
        }
3252
    }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3253
        if(mb_type < 5){
3254
            partition_count= p_mb_type_info[mb_type].partition_count;
3255
            mb_type=         p_mb_type_info[mb_type].type;
3256
        }else{
3257
            mb_type -= 5;
3258
            goto decode_intra_mb;
3259
        }
3260
    }else{
3261
       assert(h->slice_type == I_TYPE);
3262
decode_intra_mb:
3263
        if(mb_type > 25){
3264
            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);
3265
            return -1;
3266
        }
3267
        partition_count=0;
3268
        cbp= i_mb_type_info[mb_type].cbp;
3269
        h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3270
        mb_type= i_mb_type_info[mb_type].type;
3271
    }
3272

    
3273
    if(h->mb_field_decoding_flag)
3274
        mb_type |= MB_TYPE_INTERLACED;
3275

    
3276
    s->current_picture.mb_type[mb_xy]= mb_type;
3277
    h->slice_table[ mb_xy ]= h->slice_num;
3278
    
3279
    if(IS_INTRA_PCM(mb_type)){
3280
        const uint8_t *ptr;
3281
        int x, y;
3282
        
3283
        // we assume these blocks are very rare so we dont optimize it
3284
        align_get_bits(&s->gb);
3285
        
3286
        ptr= s->gb.buffer + get_bits_count(&s->gb);
3287
    
3288
        for(y=0; y<16; y++){
3289
            const int index= 4*(y&3) + 64*(y>>2);
3290
            for(x=0; x<16; x++){
3291
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3292
            }
3293
        }
3294
        for(y=0; y<8; y++){
3295
            const int index= 256 + 4*(y&3) + 32*(y>>2);
3296
            for(x=0; x<8; x++){
3297
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3298
            }
3299
        }
3300
        for(y=0; y<8; y++){
3301
            const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3302
            for(x=0; x<8; x++){
3303
                h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3304
            }
3305
        }
3306
    
3307
        skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3308
        
3309
        memset(h->non_zero_count[mb_xy], 16, 16);
3310
        
3311
        return 0;
3312
    }
3313
        
3314
    fill_caches(h, mb_type);
3315

    
3316
    //mb_pred
3317
    if(IS_INTRA(mb_type)){
3318
//            init_top_left_availability(h);
3319
            if(IS_INTRA4x4(mb_type)){
3320
                int i;
3321

    
3322
//                fill_intra4x4_pred_table(h);
3323
                for(i=0; i<16; i++){
3324
                    const int mode_coded= !get_bits1(&s->gb);
3325
                    const int predicted_mode=  pred_intra_mode(h, i);
3326
                    int mode;
3327

    
3328
                    if(mode_coded){
3329
                        const int rem_mode= get_bits(&s->gb, 3);
3330
                        if(rem_mode<predicted_mode)
3331
                            mode= rem_mode;
3332
                        else
3333
                            mode= rem_mode + 1;
3334
                    }else{
3335
                        mode= predicted_mode;
3336
                    }
3337
                    
3338
                    h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3339
                }
3340
                write_back_intra_pred_mode(h);
3341
                if( check_intra4x4_pred_mode(h) < 0)
3342
                    return -1;
3343
            }else{
3344
                h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3345
                if(h->intra16x16_pred_mode < 0)
3346
                    return -1;
3347
            }
3348
            h->chroma_pred_mode= get_ue_golomb(&s->gb);
3349

    
3350
            h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3351
            if(h->chroma_pred_mode < 0)
3352
                return -1;
3353
    }else if(partition_count==4){
3354
        int i, j, sub_partition_count[4], list, ref[2][4];
3355
        
3356
        if(h->slice_type == B_TYPE){
3357
            for(i=0; i<4; i++){
3358
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3359
                if(h->sub_mb_type[i] >=13){
3360
                    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);
3361
                    return -1;
3362
                }
3363
                sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3364
                h->sub_mb_type[i]=      b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3365
            }
3366
        }else{
3367
            assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3368
            for(i=0; i<4; i++){
3369
                h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3370
                if(h->sub_mb_type[i] >=4){
3371
                    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);
3372
                    return -1;
3373
                }
3374
                sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3375
                h->sub_mb_type[i]=      p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3376
            }
3377
        }
3378
        
3379
        for(list=0; list<2; list++){
3380
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3381
            if(ref_count == 0) continue;
3382
            for(i=0; i<4; i++){
3383
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3384
                    ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3385
                }else{
3386
                 //FIXME
3387
                    ref[list][i] = -1;
3388
                }
3389
            }
3390
        }
3391
        
3392
        for(list=0; list<2; list++){
3393
            const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3394
            if(ref_count == 0) continue;
3395

    
3396
            for(i=0; i<4; i++){
3397
                h->ref_cache[list][ scan8[4*i]   ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3398
                h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3399

    
3400
                if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3401
                    const int sub_mb_type= h->sub_mb_type[i];
3402
                    const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3403
                    for(j=0; j<sub_partition_count[i]; j++){
3404
                        int mx, my;
3405
                        const int index= 4*i + block_width*j;
3406
                        int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3407
                        pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3408
                        mx += get_se_golomb(&s->gb);
3409
                        my += get_se_golomb(&s->gb);
3410
                        tprintf("final mv:%d %d\n", mx, my);
3411

    
3412
                        if(IS_SUB_8X8(sub_mb_type)){
3413
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= 
3414
                            mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3415
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= 
3416
                            mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3417
                        }else if(IS_SUB_8X4(sub_mb_type)){
3418
                            mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3419
                            mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3420
                        }else if(IS_SUB_4X8(sub_mb_type)){
3421
                            mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3422
                            mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3423
                        }else{
3424
                            assert(IS_SUB_4X4(sub_mb_type));
3425
                            mv_cache[ 0 ][0]= mx;
3426
                            mv_cache[ 0 ][1]= my;
3427
                        }
3428
                    }
3429
                }else{
3430
                    uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3431
                    p[0] = p[1]=
3432
                    p[8] = p[9]= 0;
3433
                }
3434
            }
3435
        }
3436
    }else if(!IS_DIRECT(mb_type)){
3437
        int list, mx, my, i;
3438
         //FIXME we should set ref_idx_l? to 0 if we use that later ...
3439
        if(IS_16X16(mb_type)){
3440
            for(list=0; list<2; list++){
3441
                if(h->ref_count[0]>0){
3442
                    if(IS_DIR(mb_type, 0, list)){
3443
                        const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3444
                        fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3445
                    }
3446
                }
3447
            }
3448
            for(list=0; list<2; list++){
3449
                if(IS_DIR(mb_type, 0, list)){
3450
                    pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3451
                    mx += get_se_golomb(&s->gb);
3452
                    my += get_se_golomb(&s->gb);
3453
                    tprintf("final mv:%d %d\n", mx, my);
3454

    
3455
                    fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
3456
                }
3457
            }
3458
        }
3459
        else if(IS_16X8(mb_type)){
3460
            for(list=0; list<2; list++){
3461
                if(h->ref_count[list]>0){
3462
                    for(i=0; i<2; i++){
3463
                        if(IS_DIR(mb_type, i, list)){
3464
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3465
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3466
                        }
3467
                    }
3468
                }
3469
            }
3470
            for(list=0; list<2; list++){
3471
                for(i=0; i<2; i++){
3472
                    if(IS_DIR(mb_type, i, list)){
3473
                        pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3474
                        mx += get_se_golomb(&s->gb);
3475
                        my += get_se_golomb(&s->gb);
3476
                        tprintf("final mv:%d %d\n", mx, my);
3477

    
3478
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
3479
                    }
3480
                }
3481
            }
3482
        }else{
3483
            assert(IS_8X16(mb_type));
3484
            for(list=0; list<2; list++){
3485
                if(h->ref_count[list]>0){
3486
                    for(i=0; i<2; i++){
3487
                        if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3488
                            const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3489
                            fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3490
                        }
3491
                    }
3492
                }
3493
            }
3494
            for(list=0; list<2; list++){
3495
                for(i=0; i<2; i++){
3496
                    if(IS_DIR(mb_type, i, list)){
3497
                        pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3498
                        mx += get_se_golomb(&s->gb);
3499
                        my += get_se_golomb(&s->gb);
3500
                        tprintf("final mv:%d %d\n", mx, my);
3501

    
3502
                        fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
3503
                    }
3504
                }
3505
            }
3506
        }
3507
    }
3508
    
3509
    if(IS_INTER(mb_type))
3510
        write_back_motion(h, mb_type);
3511
    
3512
    if(!IS_INTRA16x16(mb_type)){
3513
        cbp= get_ue_golomb(&s->gb);
3514
        if(cbp > 47){
3515
            fprintf(stderr, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3516
            return -1;
3517
        }
3518
        
3519
        if(IS_INTRA4x4(mb_type))
3520
            cbp= golomb_to_intra4x4_cbp[cbp];
3521
        else
3522
            cbp= golomb_to_inter_cbp[cbp];
3523
    }
3524

    
3525
    if(cbp || IS_INTRA16x16(mb_type)){
3526
        int i8x8, i4x4, chroma_idx;
3527
        int chroma_qp, dquant;
3528
        GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3529
        const uint8_t *scan, *dc_scan;
3530
        
3531
//        fill_non_zero_count_cache(h);
3532

    
3533
        if(IS_INTERLACED(mb_type)){
3534
            scan= field_scan;
3535
            dc_scan= luma_dc_field_scan;
3536
        }else{
3537
            scan= zigzag_scan;
3538
            dc_scan= luma_dc_zigzag_scan;
3539
        }
3540

    
3541
        dquant= get_se_golomb(&s->gb);
3542

    
3543
        if( dquant > 25 || dquant < -26 ){
3544
            fprintf(stderr, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3545
            return -1;
3546
        }
3547
        
3548
        s->qscale += dquant;
3549
        if(((unsigned)s->qscale) > 51){
3550
            if(s->qscale<0) s->qscale+= 52;
3551
            else            s->qscale-= 52;
3552
        }
3553
        
3554
        h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3555
        if(IS_INTRA16x16(mb_type)){
3556
            if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3557
                return -1; //FIXME continue if partotioned and other retirn -1 too
3558
            }
3559

    
3560
            assert((cbp&15) == 0 || (cbp&15) == 15);
3561

    
3562
            if(cbp&15){
3563
                for(i8x8=0; i8x8<4; i8x8++){
3564
                    for(i4x4=0; i4x4<4; i4x4++){
3565
                        const int index= i4x4 + 4*i8x8;
3566
                        if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3567
                            return -1;
3568
                        }
3569
                    }
3570
                }
3571
            }else{
3572
                memset(&h->non_zero_count_cache[8], 0, 8*4); //FIXME stupid & slow
3573
            }
3574
        }else{
3575
            for(i8x8=0; i8x8<4; i8x8++){
3576
                if(cbp & (1<<i8x8)){
3577
                    for(i4x4=0; i4x4<4; i4x4++){
3578
                        const int index= i4x4 + 4*i8x8;
3579
                        
3580
                        if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3581
                            return -1;
3582
                        }
3583
                    }
3584
                }else{
3585
                    uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3586
                    nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3587
                }
3588
            }
3589
        }
3590
        
3591
        if(cbp&0x30){
3592
            for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3593
                if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3594
                    return -1;
3595
                }
3596
        }
3597

    
3598
        if(cbp&0x20){
3599
            for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3600
                for(i4x4=0; i4x4<4; i4x4++){
3601
                    const int index= 16 + 4*chroma_idx + i4x4;
3602
                    if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3603
                        return -1;
3604
                    }
3605
                }
3606
            }
3607
        }else{
3608
            uint8_t * const nnz= &h->non_zero_count_cache[0];
3609
            nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3610
            nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3611
        }
3612
    }else{
3613
        memset(&h->non_zero_count_cache[8], 0, 8*5);
3614
    }
3615
    write_back_non_zero_count(h);
3616

    
3617
    return 0;
3618
}
3619

    
3620
static int decode_slice(H264Context *h){
3621
    MpegEncContext * const s = &h->s;
3622
    const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3623

    
3624
    s->mb_skip_run= -1;
3625
    
3626
#if 1
3627
    for(;;){
3628
        int ret= decode_mb(h);
3629
            
3630
        hl_decode_mb(h);
3631
        
3632
        if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
3633
            s->mb_y++;
3634
            ret= decode_mb(h);
3635
            
3636
            hl_decode_mb(h);
3637
            s->mb_y--;
3638
        }
3639

    
3640
        if(ret<0){
3641
            fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3642
            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);
3643

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

    
3653
                if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3654
                    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);
3655

    
3656
                    return 0;
3657
                }else{
3658
                    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);
3659

    
3660
                    return -1;
3661
                }
3662
            }
3663
        }
3664
        
3665
        if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3666
            if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3667
                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);
3668

    
3669
                return 0;
3670
            }else{
3671
                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3672

    
3673
                return -1;
3674
            }
3675
        }
3676
    }
3677
#endif
3678
#if 0
3679
    for(;s->mb_y < s->mb_height; s->mb_y++){
3680
        for(;s->mb_x < s->mb_width; s->mb_x++){
3681
            int ret= decode_mb(h);
3682
            
3683
            hl_decode_mb(h);
3684

3685
            if(ret<0){
3686
                fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3687
                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);
3688

3689
                return -1;
3690
            }
3691
        
3692
            if(++s->mb_x >= s->mb_width){
3693
                s->mb_x=0;
3694
                if(++s->mb_y >= s->mb_height){
3695
                    if(get_bits_count(s->gb) == s->gb.size_in_bits){
3696
                        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);
3697

3698
                        return 0;
3699
                    }else{
3700
                        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);
3701

3702
                        return -1;
3703
                    }
3704
                }
3705
            }
3706
        
3707
            if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
3708
                if(get_bits_count(s->gb) == s->gb.size_in_bits){
3709
                    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);
3710

3711
                    return 0;
3712
                }else{
3713
                    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);
3714

3715
                    return -1;
3716
                }
3717
            }
3718
        }
3719
        s->mb_x=0;
3720
        ff_draw_horiz_band(s, 16*s->mb_y, 16);
3721
    }
3722
#endif
3723
    return -1; //not reached
3724
}
3725

    
3726
static inline int decode_vui_parameters(H264Context *h, SPS *sps){
3727
    MpegEncContext * const s = &h->s;
3728
    int aspect_ratio_info_present_flag, aspect_ratio_idc;
3729

    
3730
    aspect_ratio_info_present_flag= get_bits1(&s->gb);
3731
    
3732
    if( aspect_ratio_info_present_flag ) {
3733
        aspect_ratio_idc= get_bits(&s->gb, 8);
3734
        if( aspect_ratio_idc == EXTENDED_SAR ) {
3735
            sps->sar_width= get_bits(&s->gb, 16);
3736
            sps->sar_height= get_bits(&s->gb, 16);
3737
        }else if(aspect_ratio_idc < 16){
3738
            sps->sar_width=  pixel_aspect[aspect_ratio_idc][0];
3739
            sps->sar_height= pixel_aspect[aspect_ratio_idc][1];
3740
        }else{
3741
            fprintf(stderr, "illegal aspect ratio\n");
3742
            return -1;
3743
        }
3744
    }else{
3745
        sps->sar_width= 
3746
        sps->sar_height= 0;
3747
    }
3748
//            s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
3749
#if 0
3750
| overscan_info_present_flag                        |0  |u(1)    |
3751
| if( overscan_info_present_flag )                  |   |        |
3752
|  overscan_appropriate_flag                        |0  |u(1)    |
3753
| video_signal_type_present_flag                    |0  |u(1)    |
3754
| if( video_signal_type_present_flag ) {            |   |        |
3755
|  video_format                                     |0  |u(3)    |
3756
|  video_full_range_flag                            |0  |u(1)    |
3757
|  colour_description_present_flag                  |0  |u(1)    |
3758
|  if( colour_description_present_flag ) {          |   |        |
3759
|   colour_primaries                                |0  |u(8)    |
3760
|   transfer_characteristics                        |0  |u(8)    |
3761
|   matrix_coefficients                             |0  |u(8)    |
3762
|  }                                                |   |        |
3763
| }                                                 |   |        |
3764
| chroma_location_info_present_flag                 |0  |u(1)    |
3765
| if ( chroma_location_info_present_flag ) {        |   |        |
3766
|  chroma_sample_location_type_top_field            |0  |ue(v)   |
3767
|  chroma_sample_location_type_bottom_field         |0  |ue(v)   |
3768
| }                                                 |   |        |
3769
| timing_info_present_flag                          |0  |u(1)    |
3770
| if( timing_info_present_flag ) {                  |   |        |
3771
|  num_units_in_tick                                |0  |u(32)   |
3772
|  time_scale                                       |0  |u(32)   |
3773
|  fixed_frame_rate_flag                            |0  |u(1)    |
3774
| }                                                 |   |        |
3775
| nal_hrd_parameters_present_flag                   |0  |u(1)    |
3776
| if( nal_hrd_parameters_present_flag  = =  1)      |   |        |
3777
|  hrd_parameters( )                                |   |        |
3778
| vcl_hrd_parameters_present_flag                   |0  |u(1)    |
3779
| if( vcl_hrd_parameters_present_flag  = =  1)      |   |        |
3780
|  hrd_parameters( )                                |   |        |
3781
| if( ( nal_hrd_parameters_present_flag  = =  1  | ||   |        |
3782
|                                                   |   |        |
3783
|( vcl_hrd_parameters_present_flag  = =  1 ) )      |   |        |
3784
|  low_delay_hrd_flag                               |0  |u(1)    |
3785
| bitstream_restriction_flag                        |0  |u(1)    |
3786
| if( bitstream_restriction_flag ) {                |0  |u(1)    |
3787
|  motion_vectors_over_pic_boundaries_flag          |0  |u(1)    |
3788
|  max_bytes_per_pic_denom                          |0  |ue(v)   |
3789
|  max_bits_per_mb_denom                            |0  |ue(v)   |
3790
|  log2_max_mv_length_horizontal                    |0  |ue(v)   |
3791
|  log2_max_mv_length_vertical                      |0  |ue(v)   |
3792
|  num_reorder_frames                               |0  |ue(v)   |
3793
|  max_dec_frame_buffering                          |0  |ue(v)   |
3794
| }                                                 |   |        |
3795
|}                                                  |   |        |
3796
#endif
3797
    return 0;
3798
}
3799

    
3800
static inline int decode_seq_parameter_set(H264Context *h){
3801
    MpegEncContext * const s = &h->s;
3802
    int profile_idc, level_idc;
3803
    int sps_id, i;
3804
    SPS *sps;
3805
    
3806
    profile_idc= get_bits(&s->gb, 8);
3807
    get_bits1(&s->gb);   //constraint_set0_flag
3808
    get_bits1(&s->gb);   //constraint_set1_flag
3809
    get_bits1(&s->gb);   //constraint_set2_flag
3810
    get_bits(&s->gb, 5); // reserved
3811
    level_idc= get_bits(&s->gb, 8);
3812
    sps_id= get_ue_golomb(&s->gb);
3813
    
3814
    sps= &h->sps_buffer[ sps_id ];
3815
    sps->profile_idc= profile_idc;
3816
    sps->level_idc= level_idc;
3817
    
3818
    sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
3819
    sps->poc_type= get_ue_golomb(&s->gb);
3820
    
3821
    if(sps->poc_type == 0){ //FIXME #define
3822
        sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
3823
    } else if(sps->poc_type == 1){//FIXME #define
3824
        sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
3825
        sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
3826
        sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
3827
        sps->poc_cycle_length= get_ue_golomb(&s->gb);
3828
        
3829
        for(i=0; i<sps->poc_cycle_length; i++)
3830
            sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
3831
    }
3832
    if(sps->poc_type > 2){
3833
        fprintf(stderr, "illegal POC type %d\n", sps->poc_type);
3834
        return -1;
3835
    }
3836

    
3837
    sps->ref_frame_count= get_ue_golomb(&s->gb);
3838
    sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
3839
    sps->mb_width= get_ue_golomb(&s->gb) + 1;
3840
    sps->mb_height= get_ue_golomb(&s->gb) + 1;
3841
    sps->frame_mbs_only_flag= get_bits1(&s->gb);
3842
    if(!sps->frame_mbs_only_flag)
3843
        sps->mb_aff= get_bits1(&s->gb);
3844
    else
3845
        sps->mb_aff= 0;
3846

    
3847
    sps->direct_8x8_inference_flag= get_bits1(&s->gb);
3848

    
3849
    sps->crop= get_bits1(&s->gb);
3850
    if(sps->crop){
3851
        sps->crop_left  = get_ue_golomb(&s->gb);
3852
        sps->crop_right = get_ue_golomb(&s->gb);
3853
        sps->crop_top   = get_ue_golomb(&s->gb);
3854
        sps->crop_bottom= get_ue_golomb(&s->gb);
3855
        if(sps->crop_left || sps->crop_top){
3856
            fprintf(stderr, "insane croping not completly supported, this could look slightly wrong ...\n");
3857
        }
3858
    }else{
3859
        sps->crop_left  = 
3860
        sps->crop_right = 
3861
        sps->crop_top   = 
3862
        sps->crop_bottom= 0;
3863
    }
3864

    
3865
    sps->vui_parameters_present_flag= get_bits1(&s->gb);
3866
    if( sps->vui_parameters_present_flag )
3867
        decode_vui_parameters(h, sps);
3868
    
3869
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3870
        printf("sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s crop:%d/%d/%d/%d %s\n", 
3871
               sps_id, sps->profile_idc, sps->level_idc,
3872
               sps->poc_type,
3873
               sps->ref_frame_count,
3874
               sps->mb_width, sps->mb_height,
3875
               sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
3876
               sps->direct_8x8_inference_flag ? "8B8" : "",
3877
               sps->crop_left, sps->crop_right, 
3878
               sps->crop_top, sps->crop_bottom, 
3879
               sps->vui_parameters_present_flag ? "VUI" : ""
3880
               );
3881
    }
3882
    return 0;
3883
}
3884

    
3885
static inline int decode_picture_parameter_set(H264Context *h){
3886
    MpegEncContext * const s = &h->s;
3887
    int pps_id= get_ue_golomb(&s->gb);
3888
    PPS *pps= &h->pps_buffer[pps_id];
3889
    
3890
    pps->sps_id= get_ue_golomb(&s->gb);
3891
    pps->cabac= get_bits1(&s->gb);
3892
    pps->pic_order_present= get_bits1(&s->gb);
3893
    pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
3894
    if(pps->slice_group_count > 1 ){
3895
        pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
3896
fprintf(stderr, "FMO not supported\n");
3897
        switch(pps->mb_slice_group_map_type){
3898
        case 0:
3899
#if 0
3900
|   for( i = 0; i <= num_slice_groups_minus1; i++ ) |   |        |
3901
|    run_length[ i ]                                |1  |ue(v)   |
3902
#endif
3903
            break;
3904
        case 2:
3905
#if 0
3906
|   for( i = 0; i < num_slice_groups_minus1; i++ )  |   |        |
3907
|{                                                  |   |        |
3908
|    top_left_mb[ i ]                               |1  |ue(v)   |
3909
|    bottom_right_mb[ i ]                           |1  |ue(v)   |
3910
|   }                                               |   |        |
3911
#endif
3912
            break;
3913
        case 3:
3914
        case 4:
3915
        case 5:
3916
#if 0
3917
|   slice_group_change_direction_flag               |1  |u(1)    |
3918
|   slice_group_change_rate_minus1                  |1  |ue(v)   |
3919
#endif
3920
            break;
3921
        case 6:
3922
#if 0
3923
|   slice_group_id_cnt_minus1                       |1  |ue(v)   |
3924
|   for( i = 0; i <= slice_group_id_cnt_minus1; i++ |   |        |
3925
|)                                                  |   |        |
3926
|    slice_group_id[ i ]                            |1  |u(v)    |
3927
#endif
3928
            break;
3929
        }
3930
    }
3931
    pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3932
    pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3933
    if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
3934
        fprintf(stderr, "reference overflow (pps)\n");
3935
        return -1;
3936
    }
3937
    
3938
    pps->weighted_pred= get_bits1(&s->gb);
3939
    pps->weighted_bipred_idc= get_bits(&s->gb, 2);
3940
    pps->init_qp= get_se_golomb(&s->gb) + 26;
3941
    pps->init_qs= get_se_golomb(&s->gb) + 26;
3942
    pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
3943
    pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
3944
    pps->constrained_intra_pred= get_bits1(&s->gb);
3945
    pps->redundant_pic_cnt_present = get_bits1(&s->gb);
3946
    
3947
    if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3948
        printf("pps:%d sps:%d %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s\n", 
3949
               pps_id, pps->sps_id,
3950
               pps->cabac ? "CABAC" : "CAVLC",
3951
               pps->slice_group_count,
3952
               pps->ref_count[0], pps->ref_count[1],
3953
               pps->weighted_pred ? "weighted" : "",
3954
               pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
3955
               pps->deblocking_filter_parameters_present ? "LPAR" : "",
3956
               pps->constrained_intra_pred ? "CONSTR" : "",
3957
               pps->redundant_pic_cnt_present ? "REDU" : ""
3958
               );
3959
    }
3960
    
3961
    return 0;
3962
}
3963

    
3964
/**
3965
 * finds the end of the current frame in the bitstream.
3966
 * @return the position of the first byte of the next frame, or -1
3967
 */
3968
static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){
3969
    ParseContext *pc= &s->parse_context;
3970
    int i;
3971
    uint32_t state;
3972
//printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
3973
//    mb_addr= pc->mb_addr - 1;
3974
    state= pc->state;
3975
    //FIXME this will fail with slices
3976
    for(i=0; i<buf_size; i++){
3977
        state= (state<<8) | buf[i];
3978
        if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
3979
            if(pc->frame_start_found){
3980
                pc->state=-1; 
3981
                pc->frame_start_found= 0;
3982
                return i-3;
3983
            }
3984
            pc->frame_start_found= 1;
3985
        }
3986
    }
3987
    
3988
    pc->state= state;
3989
    return END_NOT_FOUND;
3990
}
3991

    
3992
static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
3993
    MpegEncContext * const s = &h->s;
3994
    AVCodecContext * const avctx= s->avctx;
3995
    int buf_index=0;
3996
#if 0
3997
    int i;
3998
    for(i=0; i<32; i++){
3999
        printf("%X ", buf[i]);
4000
    }
4001
#endif
4002
    for(;;){
4003
        int consumed;
4004
        int dst_length;
4005
        int bit_length;
4006
        uint8_t *ptr;
4007
        
4008
        // start code prefix search
4009
        for(; buf_index + 3 < buf_size; buf_index++){
4010
            // this should allways succeed in the first iteration
4011
            if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
4012
                break;
4013
        }
4014
        
4015
        if(buf_index+3 >= buf_size) break;
4016
        
4017
        buf_index+=3;
4018
        
4019
        ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index);
4020
        if(ptr[dst_length - 1] == 0) dst_length--;
4021
        bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
4022

    
4023
        if(s->avctx->debug&FF_DEBUG_STARTCODE){
4024
            printf("NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
4025
        }
4026
        
4027
        buf_index += consumed;
4028

    
4029
        if(h->nal_ref_idc < s->hurry_up)
4030
            continue;
4031
        
4032
        switch(h->nal_unit_type){
4033
        case NAL_IDR_SLICE:
4034
            idr(h); //FIXME ensure we dont loose some frames if there is reordering
4035
        case NAL_SLICE:
4036
            init_get_bits(&s->gb, ptr, bit_length);
4037
            h->intra_gb_ptr=
4038
            h->inter_gb_ptr= &s->gb;
4039
            s->data_partitioning = 0;
4040
            
4041
            if(decode_slice_header(h) < 0) return -1;
4042
            if(h->redundant_pic_count==0)
4043
                decode_slice(h);
4044
            break;
4045
        case NAL_DPA:
4046
            init_get_bits(&s->gb, ptr, bit_length);
4047
            h->intra_gb_ptr=
4048
            h->inter_gb_ptr= NULL;
4049
            s->data_partitioning = 1;
4050
            
4051
            if(decode_slice_header(h) < 0) return -1;
4052
            break;
4053
        case NAL_DPB:
4054
            init_get_bits(&h->intra_gb, ptr, bit_length);
4055
            h->intra_gb_ptr= &h->intra_gb;
4056
            break;
4057
        case NAL_DPC:
4058
            init_get_bits(&h->inter_gb, ptr, bit_length);
4059
            h->inter_gb_ptr= &h->inter_gb;
4060

    
4061
            if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning)
4062
                decode_slice(h);
4063
            break;
4064
        case NAL_SEI:
4065
            break;
4066
        case NAL_SPS:
4067
            init_get_bits(&s->gb, ptr, bit_length);
4068
            decode_seq_parameter_set(h);
4069
            
4070
            if(s->flags& CODEC_FLAG_LOW_DELAY)
4071
                s->low_delay=1;
4072
      
4073
            avctx->has_b_frames= !s->low_delay;
4074
            break;
4075
        case NAL_PPS:
4076
            init_get_bits(&s->gb, ptr, bit_length);
4077
            
4078
            decode_picture_parameter_set(h);
4079

    
4080
            break;
4081
        case NAL_PICTURE_DELIMITER:
4082
            break;
4083
        case NAL_FILTER_DATA:
4084
            break;
4085
        }        
4086

    
4087
        //FIXME move after where irt is set
4088
        s->current_picture.pict_type= s->pict_type;
4089
        s->current_picture.key_frame= s->pict_type == I_TYPE;
4090
    }
4091
    
4092
    if(!s->current_picture_ptr) return buf_index; //no frame
4093
    
4094
    h->prev_frame_num_offset= h->frame_num_offset;
4095
    h->prev_frame_num= h->frame_num;
4096
    if(s->current_picture_ptr->reference){
4097
        h->prev_poc_msb= h->poc_msb;
4098
        h->prev_poc_lsb= h->poc_lsb;
4099
    }
4100
    if(s->current_picture_ptr->reference)
4101
        execute_ref_pic_marking(h, h->mmco, h->mmco_index);
4102
    else
4103
        assert(h->mmco_index==0);
4104

    
4105
    ff_er_frame_end(s);
4106
    MPV_frame_end(s);
4107

    
4108
    return buf_index;
4109
}
4110

    
4111
/**
4112
 * retunrs the number of bytes consumed for building the current frame
4113
 */
4114
static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
4115
    if(s->flags&CODEC_FLAG_TRUNCATED){
4116
        pos -= s->parse_context.last_index;
4117
        if(pos<0) pos=0; // FIXME remove (uneeded?)
4118
        
4119
        return pos;
4120
    }else{
4121
        if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
4122
        if(pos+10>buf_size) pos=buf_size; // oops ;)
4123

    
4124
        return pos;
4125
    }
4126
}
4127

    
4128
static int decode_frame(AVCodecContext *avctx, 
4129
                             void *data, int *data_size,
4130
                             uint8_t *buf, int buf_size)
4131
{
4132
    H264Context *h = avctx->priv_data;
4133
    MpegEncContext *s = &h->s;
4134
    AVFrame *pict = data; 
4135
    int buf_index;
4136
    
4137
    s->flags= avctx->flags;
4138

    
4139
    *data_size = 0;
4140
   
4141
   /* no supplementary picture */
4142
    if (buf_size == 0) {
4143
        return 0;
4144
    }
4145
    
4146
    if(s->flags&CODEC_FLAG_TRUNCATED){
4147
        int next= find_frame_end(s, buf, buf_size);
4148
        
4149
        if( ff_combine_frame(s, next, &buf, &buf_size) < 0 )
4150
            return buf_size;
4151
//printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
4152
    }
4153

    
4154
    if(s->avctx->extradata_size && s->picture_number==0){
4155
        if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) ) 
4156
            return -1;
4157
    }
4158

    
4159
    buf_index=decode_nal_units(h, buf, buf_size);
4160
    if(buf_index < 0) 
4161
        return -1;
4162

    
4163
    //FIXME do something with unavailable reference frames    
4164
 
4165
//    if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
4166
#if 0
4167
    if(s->pict_type==B_TYPE || s->low_delay){
4168
        *pict= *(AVFrame*)&s->current_picture;
4169
    } else {
4170
        *pict= *(AVFrame*)&s->last_picture;
4171
    }
4172
#endif
4173
    if(!s->current_picture_ptr){
4174
        fprintf(stderr, "error, NO frame\n");
4175
        return -1;
4176
    }
4177

    
4178
    *pict= *(AVFrame*)&s->current_picture; //FIXME 
4179
    ff_print_debug_info(s, s->current_picture_ptr);
4180
    assert(pict->data[0]);
4181
//printf("out %d\n", (int)pict->data[0]);
4182
#if 0 //?
4183

4184
    /* Return the Picture timestamp as the frame number */
4185
    /* we substract 1 because it is added on utils.c    */
4186
    avctx->frame_number = s->picture_number - 1;
4187
#endif
4188
#if 0
4189
    /* dont output the last pic after seeking */
4190
    if(s->last_picture_ptr || s->low_delay)
4191
    //Note this isnt a issue as a IDR pic should flush teh buffers
4192
#endif
4193
        *data_size = sizeof(AVFrame);
4194
    return get_consumed_bytes(s, buf_index, buf_size);
4195
}
4196
#if 0
4197
static inline void fill_mb_avail(H264Context *h){
4198
    MpegEncContext * const s = &h->s;
4199
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4200

4201
    if(s->mb_y){
4202
        h->mb_avail[0]= s->mb_x                 && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
4203
        h->mb_avail[1]=                            h->slice_table[mb_xy - s->mb_stride    ] == h->slice_num;
4204
        h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
4205
    }else{
4206
        h->mb_avail[0]=
4207
        h->mb_avail[1]=
4208
        h->mb_avail[2]= 0;
4209
    }
4210
    h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
4211
    h->mb_avail[4]= 1; //FIXME move out
4212
    h->mb_avail[5]= 0; //FIXME move out
4213
}
4214
#endif
4215

    
4216
#if 0 //selftest
4217
#define COUNT 8000
4218
#define SIZE (COUNT*40)
4219
int main(){
4220
    int i;
4221
    uint8_t temp[SIZE];
4222
    PutBitContext pb;
4223
    GetBitContext gb;
4224
//    int int_temp[10000];
4225
    DSPContext dsp;
4226
    AVCodecContext avctx;
4227
    
4228
    dsputil_init(&dsp, &avctx);
4229

4230
    init_put_bits(&pb, temp, SIZE, NULL, NULL);
4231
    printf("testing unsigned exp golomb\n");
4232
    for(i=0; i<COUNT; i++){
4233
        START_TIMER
4234
        set_ue_golomb(&pb, i);
4235
        STOP_TIMER("set_ue_golomb");
4236
    }
4237
    flush_put_bits(&pb);
4238
    
4239
    init_get_bits(&gb, temp, 8*SIZE);
4240
    for(i=0; i<COUNT; i++){
4241
        int j, s;
4242
        
4243
        s= show_bits(&gb, 24);
4244
        
4245
        START_TIMER
4246
        j= get_ue_golomb(&gb);
4247
        if(j != i){
4248
            printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4249
//            return -1;
4250
        }
4251
        STOP_TIMER("get_ue_golomb");
4252
    }
4253
    
4254
    
4255
    init_put_bits(&pb, temp, SIZE, NULL, NULL);
4256
    printf("testing signed exp golomb\n");
4257
    for(i=0; i<COUNT; i++){
4258
        START_TIMER
4259
        set_se_golomb(&pb, i - COUNT/2);
4260
        STOP_TIMER("set_se_golomb");
4261
    }
4262
    flush_put_bits(&pb);
4263
    
4264
    init_get_bits(&gb, temp, 8*SIZE);
4265
    for(i=0; i<COUNT; i++){
4266
        int j, s;
4267
        
4268
        s= show_bits(&gb, 24);
4269
        
4270
        START_TIMER
4271
        j= get_se_golomb(&gb);
4272
        if(j != i - COUNT/2){
4273
            printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4274
//            return -1;
4275
        }
4276
        STOP_TIMER("get_se_golomb");
4277
    }
4278

4279
    printf("testing 4x4 (I)DCT\n");
4280
    
4281
    DCTELEM block[16];
4282
    uint8_t src[16], ref[16];
4283
    uint64_t error= 0, max_error=0;
4284

4285
    for(i=0; i<COUNT; i++){
4286
        int j;
4287
//        printf("%d %d %d\n", r1, r2, (r2-r1)*16);
4288
        for(j=0; j<16; j++){
4289
            ref[j]= random()%255;
4290
            src[j]= random()%255;
4291
        }
4292

4293
        h264_diff_dct_c(block, src, ref, 4);
4294
        
4295
        //normalize
4296
        for(j=0; j<16; j++){
4297
//            printf("%d ", block[j]);
4298
            block[j]= block[j]*4;
4299
            if(j&1) block[j]= (block[j]*4 + 2)/5;
4300
            if(j&4) block[j]= (block[j]*4 + 2)/5;
4301
        }
4302
//        printf("\n");
4303
        
4304
        h264_add_idct_c(ref, block, 4);
4305
/*        for(j=0; j<16; j++){
4306
            printf("%d ", ref[j]);
4307
        }
4308
        printf("\n");*/
4309
            
4310
        for(j=0; j<16; j++){
4311
            int diff= ABS(src[j] - ref[j]);
4312
            
4313
            error+= diff*diff;
4314
            max_error= FFMAX(max_error, diff);
4315
        }
4316
    }
4317
    printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
4318
#if 0
4319
    printf("testing quantizer\n");
4320
    for(qp=0; qp<52; qp++){
4321
        for(i=0; i<16; i++)
4322
            src1_block[i]= src2_block[i]= random()%255;
4323
        
4324
    }
4325
#endif
4326
    printf("Testing NAL layer\n");
4327
    
4328
    uint8_t bitstream[COUNT];
4329
    uint8_t nal[COUNT*2];
4330
    H264Context h;
4331
    memset(&h, 0, sizeof(H264Context));
4332
    
4333
    for(i=0; i<COUNT; i++){
4334
        int zeros= i;
4335
        int nal_length;
4336
        int consumed;
4337
        int out_length;
4338
        uint8_t *out;
4339
        int j;
4340
        
4341
        for(j=0; j<COUNT; j++){
4342
            bitstream[j]= (random() % 255) + 1;
4343
        }
4344
        
4345
        for(j=0; j<zeros; j++){
4346
            int pos= random() % COUNT;
4347
            while(bitstream[pos] == 0){
4348
                pos++;
4349
                pos %= COUNT;
4350
            }
4351
            bitstream[pos]=0;
4352
        }
4353
        
4354
        START_TIMER
4355
        
4356
        nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
4357
        if(nal_length<0){
4358
            printf("encoding failed\n");
4359
            return -1;
4360
        }
4361
        
4362
        out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
4363

    
4364
        STOP_TIMER("NAL")
4365
        
4366
        if(out_length != COUNT){
4367
            printf("incorrect length %d %d\n", out_length, COUNT);
4368
            return -1;
4369
        }
4370
        
4371
        if(consumed != nal_length){
4372
            printf("incorrect consumed length %d %d\n", nal_length, consumed);
4373
            return -1;
4374
        }
4375
        
4376
        if(memcmp(bitstream, out, COUNT)){
4377
            printf("missmatch\n");
4378
            return -1;
4379
        }
4380
    }
4381
    
4382
    printf("Testing RBSP\n");
4383
    
4384
    
4385
    return 0;
4386
}
4387
#endif
4388

    
4389

    
4390
static int decode_end(AVCodecContext *avctx)
4391
{
4392
    H264Context *h = avctx->priv_data;
4393
    MpegEncContext *s = &h->s;
4394
    
4395
    free_tables(h); //FIXME cleanup init stuff perhaps
4396
    MPV_common_end(s);
4397

    
4398
//    memset(h, 0, sizeof(H264Context));
4399
        
4400
    return 0;
4401
}
4402

    
4403

    
4404
AVCodec h264_decoder = {
4405
    "h264",
4406
    CODEC_TYPE_VIDEO,
4407
    CODEC_ID_H264,
4408
    sizeof(H264Context),
4409
    decode_init,
4410
    NULL,
4411
    decode_end,
4412
    decode_frame,
4413
    /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,
4414
};
4415

    
4416
#include "svq3.c"