Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.h @ 69cc3183

History | View | Annotate | Download (56.9 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 file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21

    
22
/**
23
 * @file libavcodec/h264.h
24
 * H.264 / AVC / MPEG4 part10 codec.
25
 * @author Michael Niedermayer <michaelni@gmx.at>
26
 */
27

    
28
#ifndef AVCODEC_H264_H
29
#define AVCODEC_H264_H
30

    
31
#include "dsputil.h"
32
#include "cabac.h"
33
#include "mpegvideo.h"
34
#include "h264pred.h"
35
#include "rectangle.h"
36

    
37
#define interlaced_dct interlaced_dct_is_a_bad_name
38
#define mb_intra mb_intra_is_not_initialized_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
#define MAX_DELAYED_PIC_COUNT 16
56

    
57
/* Compiling in interlaced support reduces the speed
58
 * of progressive decoding by about 2%. */
59
#define ALLOW_INTERLACE
60

    
61
#define ALLOW_NOCHROMA
62

    
63
/**
64
 * The maximum number of slices supported by the decoder.
65
 * must be a power of 2
66
 */
67
#define MAX_SLICES 16
68

    
69
#ifdef ALLOW_INTERLACE
70
#define MB_MBAFF h->mb_mbaff
71
#define MB_FIELD h->mb_field_decoding_flag
72
#define FRAME_MBAFF h->mb_aff_frame
73
#define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
74
#else
75
#define MB_MBAFF 0
76
#define MB_FIELD 0
77
#define FRAME_MBAFF 0
78
#define FIELD_PICTURE 0
79
#undef  IS_INTERLACED
80
#define IS_INTERLACED(mb_type) 0
81
#endif
82
#define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
83

    
84
#ifdef ALLOW_NOCHROMA
85
#define CHROMA h->sps.chroma_format_idc
86
#else
87
#define CHROMA 1
88
#endif
89

    
90
#ifndef CABAC
91
#define CABAC h->pps.cabac
92
#endif
93

    
94
#define EXTENDED_SAR          255
95

    
96
#define MB_TYPE_REF0       MB_TYPE_ACPRED //dirty but it fits in 16 bit
97
#define MB_TYPE_8x8DCT     0x01000000
98
#define IS_REF0(a)         ((a) & MB_TYPE_REF0)
99
#define IS_8x8DCT(a)       ((a) & MB_TYPE_8x8DCT)
100

    
101
/**
102
 * Value of Picture.reference when Picture is not a reference picture, but
103
 * is held for delayed output.
104
 */
105
#define DELAYED_PIC_REF 4
106

    
107

    
108
/* NAL unit types */
109
enum {
110
    NAL_SLICE=1,
111
    NAL_DPA,
112
    NAL_DPB,
113
    NAL_DPC,
114
    NAL_IDR_SLICE,
115
    NAL_SEI,
116
    NAL_SPS,
117
    NAL_PPS,
118
    NAL_AUD,
119
    NAL_END_SEQUENCE,
120
    NAL_END_STREAM,
121
    NAL_FILLER_DATA,
122
    NAL_SPS_EXT,
123
    NAL_AUXILIARY_SLICE=19
124
};
125

    
126
/**
127
 * SEI message types
128
 */
129
typedef enum {
130
    SEI_BUFFERING_PERIOD             =  0, ///< buffering period (H.264, D.1.1)
131
    SEI_TYPE_PIC_TIMING              =  1, ///< picture timing
132
    SEI_TYPE_USER_DATA_UNREGISTERED  =  5, ///< unregistered user data
133
    SEI_TYPE_RECOVERY_POINT          =  6  ///< recovery point (frame # to decoder sync)
134
} SEI_Type;
135

    
136
/**
137
 * pic_struct in picture timing SEI message
138
 */
139
typedef enum {
140
    SEI_PIC_STRUCT_FRAME             = 0, ///<  0: %frame
141
    SEI_PIC_STRUCT_TOP_FIELD         = 1, ///<  1: top field
142
    SEI_PIC_STRUCT_BOTTOM_FIELD      = 2, ///<  2: bottom field
143
    SEI_PIC_STRUCT_TOP_BOTTOM        = 3, ///<  3: top field, bottom field, in that order
144
    SEI_PIC_STRUCT_BOTTOM_TOP        = 4, ///<  4: bottom field, top field, in that order
145
    SEI_PIC_STRUCT_TOP_BOTTOM_TOP    = 5, ///<  5: top field, bottom field, top field repeated, in that order
146
    SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///<  6: bottom field, top field, bottom field repeated, in that order
147
    SEI_PIC_STRUCT_FRAME_DOUBLING    = 7, ///<  7: %frame doubling
148
    SEI_PIC_STRUCT_FRAME_TRIPLING    = 8  ///<  8: %frame tripling
149
} SEI_PicStructType;
150

    
151
/**
152
 * Sequence parameter set
153
 */
154
typedef struct SPS{
155

    
156
    int profile_idc;
157
    int level_idc;
158
    int chroma_format_idc;
159
    int transform_bypass;              ///< qpprime_y_zero_transform_bypass_flag
160
    int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
161
    int poc_type;                      ///< pic_order_cnt_type
162
    int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
163
    int delta_pic_order_always_zero_flag;
164
    int offset_for_non_ref_pic;
165
    int offset_for_top_to_bottom_field;
166
    int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
167
    int ref_frame_count;               ///< num_ref_frames
168
    int gaps_in_frame_num_allowed_flag;
169
    int mb_width;                      ///< pic_width_in_mbs_minus1 + 1
170
    int mb_height;                     ///< pic_height_in_map_units_minus1 + 1
171
    int frame_mbs_only_flag;
172
    int mb_aff;                        ///<mb_adaptive_frame_field_flag
173
    int direct_8x8_inference_flag;
174
    int crop;                   ///< frame_cropping_flag
175
    unsigned int crop_left;            ///< frame_cropping_rect_left_offset
176
    unsigned int crop_right;           ///< frame_cropping_rect_right_offset
177
    unsigned int crop_top;             ///< frame_cropping_rect_top_offset
178
    unsigned int crop_bottom;          ///< frame_cropping_rect_bottom_offset
179
    int vui_parameters_present_flag;
180
    AVRational sar;
181
    int video_signal_type_present_flag;
182
    int full_range;
183
    int colour_description_present_flag;
184
    enum AVColorPrimaries color_primaries;
185
    enum AVColorTransferCharacteristic color_trc;
186
    enum AVColorSpace colorspace;
187
    int timing_info_present_flag;
188
    uint32_t num_units_in_tick;
189
    uint32_t time_scale;
190
    int fixed_frame_rate_flag;
191
    short offset_for_ref_frame[256]; //FIXME dyn aloc?
192
    int bitstream_restriction_flag;
193
    int num_reorder_frames;
194
    int scaling_matrix_present;
195
    uint8_t scaling_matrix4[6][16];
196
    uint8_t scaling_matrix8[2][64];
197
    int nal_hrd_parameters_present_flag;
198
    int vcl_hrd_parameters_present_flag;
199
    int pic_struct_present_flag;
200
    int time_offset_length;
201
    int cpb_cnt;                       ///< See H.264 E.1.2
202
    int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1
203
    int cpb_removal_delay_length;      ///< cpb_removal_delay_length_minus1 + 1
204
    int dpb_output_delay_length;       ///< dpb_output_delay_length_minus1 + 1
205
    int bit_depth_luma;                ///< bit_depth_luma_minus8 + 8
206
    int bit_depth_chroma;              ///< bit_depth_chroma_minus8 + 8
207
    int residual_color_transform_flag; ///< residual_colour_transform_flag
208
}SPS;
209

    
210
/**
211
 * Picture parameter set
212
 */
213
typedef struct PPS{
214
    unsigned int sps_id;
215
    int cabac;                  ///< entropy_coding_mode_flag
216
    int pic_order_present;      ///< pic_order_present_flag
217
    int slice_group_count;      ///< num_slice_groups_minus1 + 1
218
    int mb_slice_group_map_type;
219
    unsigned int ref_count[2];  ///< num_ref_idx_l0/1_active_minus1 + 1
220
    int weighted_pred;          ///< weighted_pred_flag
221
    int weighted_bipred_idc;
222
    int init_qp;                ///< pic_init_qp_minus26 + 26
223
    int init_qs;                ///< pic_init_qs_minus26 + 26
224
    int chroma_qp_index_offset[2];
225
    int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
226
    int constrained_intra_pred; ///< constrained_intra_pred_flag
227
    int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
228
    int transform_8x8_mode;     ///< transform_8x8_mode_flag
229
    uint8_t scaling_matrix4[6][16];
230
    uint8_t scaling_matrix8[2][64];
231
    uint8_t chroma_qp_table[2][64];  ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
232
    int chroma_qp_diff;
233
}PPS;
234

    
235
/**
236
 * Memory management control operation opcode.
237
 */
238
typedef enum MMCOOpcode{
239
    MMCO_END=0,
240
    MMCO_SHORT2UNUSED,
241
    MMCO_LONG2UNUSED,
242
    MMCO_SHORT2LONG,
243
    MMCO_SET_MAX_LONG,
244
    MMCO_RESET,
245
    MMCO_LONG,
246
} MMCOOpcode;
247

    
248
/**
249
 * Memory management control operation.
250
 */
251
typedef struct MMCO{
252
    MMCOOpcode opcode;
253
    int short_pic_num;  ///< pic_num without wrapping (pic_num & max_pic_num)
254
    int long_arg;       ///< index, pic_num, or num long refs depending on opcode
255
} MMCO;
256

    
257
/**
258
 * H264Context
259
 */
260
typedef struct H264Context{
261
    MpegEncContext s;
262
    int nal_ref_idc;
263
    int nal_unit_type;
264
    uint8_t *rbsp_buffer[2];
265
    unsigned int rbsp_buffer_size[2];
266

    
267
    /**
268
      * Used to parse AVC variant of h264
269
      */
270
    int is_avc; ///< this flag is != 0 if codec is avc1
271
    int got_avcC; ///< flag used to parse avcC data only once
272
    int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
273

    
274
    int chroma_qp[2]; //QPc
275

    
276
    int qp_thresh;      ///< QP threshold to skip loopfilter
277

    
278
    int prev_mb_skipped;
279
    int next_mb_skipped;
280

    
281
    //prediction stuff
282
    int chroma_pred_mode;
283
    int intra16x16_pred_mode;
284

    
285
    int topleft_mb_xy;
286
    int top_mb_xy;
287
    int topright_mb_xy;
288
    int left_mb_xy[2];
289

    
290
    int topleft_type;
291
    int top_type;
292
    int topright_type;
293
    int left_type[2];
294

    
295
    const uint8_t * left_block;
296
    int topleft_partition;
297

    
298
    int8_t intra4x4_pred_mode_cache[5*8];
299
    int8_t (*intra4x4_pred_mode)[8];
300
    H264PredContext hpc;
301
    unsigned int topleft_samples_available;
302
    unsigned int top_samples_available;
303
    unsigned int topright_samples_available;
304
    unsigned int left_samples_available;
305
    uint8_t (*top_borders[2])[16+2*8];
306
    uint8_t left_border[2*(17+2*9)];
307

    
308
    /**
309
     * non zero coeff count cache.
310
     * is 64 if not available.
311
     */
312
    DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache)[6*8];
313

    
314
    /*
315
    .UU.YYYY
316
    .UU.YYYY
317
    .vv.YYYY
318
    .VV.YYYY
319
    */
320
    uint8_t (*non_zero_count)[32];
321

    
322
    /**
323
     * Motion vector cache.
324
     */
325
    DECLARE_ALIGNED_16(int16_t, mv_cache)[2][5*8][2];
326
    DECLARE_ALIGNED_8(int8_t, ref_cache)[2][5*8];
327
#define LIST_NOT_USED -1 //FIXME rename?
328
#define PART_NOT_AVAILABLE -2
329

    
330
    /**
331
     * is 1 if the specific list MV&references are set to 0,0,-2.
332
     */
333
    int mv_cache_clean[2];
334

    
335
    /**
336
     * number of neighbors (top and/or left) that used 8x8 dct
337
     */
338
    int neighbor_transform_size;
339

    
340
    /**
341
     * block_offset[ 0..23] for frame macroblocks
342
     * block_offset[24..47] for field macroblocks
343
     */
344
    int block_offset[2*(16+8)];
345

    
346
    uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
347
    uint32_t *mb2b8_xy;
348
    int b_stride; //FIXME use s->b4_stride
349
    int b8_stride;
350

    
351
    int mb_linesize;   ///< may be equal to s->linesize or s->linesize*2, for mbaff
352
    int mb_uvlinesize;
353

    
354
    int emu_edge_width;
355
    int emu_edge_height;
356

    
357
    int halfpel_flag;
358
    int thirdpel_flag;
359

    
360
    int unknown_svq3_flag;
361
    int next_slice_index;
362

    
363
    SPS *sps_buffers[MAX_SPS_COUNT];
364
    SPS sps; ///< current sps
365

    
366
    PPS *pps_buffers[MAX_PPS_COUNT];
367
    /**
368
     * current pps
369
     */
370
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
371

    
372
    uint32_t dequant4_buffer[6][52][16];
373
    uint32_t dequant8_buffer[2][52][64];
374
    uint32_t (*dequant4_coeff[6])[16];
375
    uint32_t (*dequant8_coeff[2])[64];
376
    int dequant_coeff_pps;     ///< reinit tables when pps changes
377

    
378
    int slice_num;
379
    uint16_t *slice_table_base;
380
    uint16_t *slice_table;     ///< slice_table_base + 2*mb_stride + 1
381
    int slice_type;
382
    int slice_type_nos;        ///< S free slice type (SI/SP are remapped to I/P)
383
    int slice_type_fixed;
384

    
385
    //interlacing specific flags
386
    int mb_aff_frame;
387
    int mb_field_decoding_flag;
388
    int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag
389

    
390
    DECLARE_ALIGNED_8(uint16_t, sub_mb_type)[4];
391

    
392
    //POC stuff
393
    int poc_lsb;
394
    int poc_msb;
395
    int delta_poc_bottom;
396
    int delta_poc[2];
397
    int frame_num;
398
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
399
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
400
    int frame_num_offset;         ///< for POC type 2
401
    int prev_frame_num_offset;    ///< for POC type 2
402
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
403

    
404
    /**
405
     * frame_num for frames or 2*frame_num+1 for field pics.
406
     */
407
    int curr_pic_num;
408

    
409
    /**
410
     * max_frame_num or 2*max_frame_num for field pics.
411
     */
412
    int max_pic_num;
413

    
414
    //Weighted pred stuff
415
    int use_weight;
416
    int use_weight_chroma;
417
    int luma_log2_weight_denom;
418
    int chroma_log2_weight_denom;
419
    int luma_weight[2][48];
420
    int luma_offset[2][48];
421
    int chroma_weight[2][48][2];
422
    int chroma_offset[2][48][2];
423
    int implicit_weight[48][48];
424

    
425
    //deblock
426
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
427
    int slice_alpha_c0_offset;
428
    int slice_beta_offset;
429

    
430
    int redundant_pic_count;
431

    
432
    int direct_spatial_mv_pred;
433
    int col_parity;
434
    int col_fieldoff;
435
    int dist_scale_factor[16];
436
    int dist_scale_factor_field[2][32];
437
    int map_col_to_list0[2][16+32];
438
    int map_col_to_list0_field[2][2][16+32];
439

    
440
    /**
441
     * num_ref_idx_l0/1_active_minus1 + 1
442
     */
443
    unsigned int ref_count[2];   ///< counts frames or fields, depending on current mb mode
444
    unsigned int list_count;
445
    uint8_t *list_counts;            ///< Array of list_count per MB specifying the slice type
446
    Picture *short_ref[32];
447
    Picture *long_ref[32];
448
    Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
449
    Picture ref_list[2][48];         /**< 0..15: frame refs, 16..47: mbaff field refs.
450
                                          Reordered version of default_ref_list
451
                                          according to picture reordering in slice header */
452
    int ref2frm[MAX_SLICES][2][64];  ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
453
    Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
454
    int outputed_poc;
455

    
456
    /**
457
     * memory management control operations buffer.
458
     */
459
    MMCO mmco[MAX_MMCO_COUNT];
460
    int mmco_index;
461

    
462
    int long_ref_count;  ///< number of actual long term references
463
    int short_ref_count; ///< number of actual short term references
464

    
465
    //data partitioning
466
    GetBitContext intra_gb;
467
    GetBitContext inter_gb;
468
    GetBitContext *intra_gb_ptr;
469
    GetBitContext *inter_gb_ptr;
470

    
471
    DECLARE_ALIGNED_16(DCTELEM, mb)[16*24];
472
    DCTELEM mb_padding[256];        ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either check that i is not too large or ensure that there is some unused stuff after mb
473

    
474
    /**
475
     * Cabac
476
     */
477
    CABACContext cabac;
478
    uint8_t      cabac_state[460];
479
    int          cabac_init_idc;
480

    
481
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
482
    uint16_t     *cbp_table;
483
    int cbp;
484
    int top_cbp;
485
    int left_cbp;
486
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
487
    uint8_t     *chroma_pred_mode_table;
488
    int         last_qscale_diff;
489
    int16_t     (*mvd_table[2])[2];
490
    DECLARE_ALIGNED_16(int16_t, mvd_cache)[2][5*8][2];
491
    uint8_t     *direct_table;
492
    uint8_t     direct_cache[5*8];
493

    
494
    uint8_t zigzag_scan[16];
495
    uint8_t zigzag_scan8x8[64];
496
    uint8_t zigzag_scan8x8_cavlc[64];
497
    uint8_t field_scan[16];
498
    uint8_t field_scan8x8[64];
499
    uint8_t field_scan8x8_cavlc[64];
500
    const uint8_t *zigzag_scan_q0;
501
    const uint8_t *zigzag_scan8x8_q0;
502
    const uint8_t *zigzag_scan8x8_cavlc_q0;
503
    const uint8_t *field_scan_q0;
504
    const uint8_t *field_scan8x8_q0;
505
    const uint8_t *field_scan8x8_cavlc_q0;
506

    
507
    int x264_build;
508

    
509
    /**
510
     * @defgroup multithreading Members for slice based multithreading
511
     * @{
512
     */
513
    struct H264Context *thread_context[MAX_THREADS];
514

    
515
    /**
516
     * current slice number, used to initalize slice_num of each thread/context
517
     */
518
    int current_slice;
519

    
520
    /**
521
     * Max number of threads / contexts.
522
     * This is equal to AVCodecContext.thread_count unless
523
     * multithreaded decoding is impossible, in which case it is
524
     * reduced to 1.
525
     */
526
    int max_contexts;
527

    
528
    /**
529
     *  1 if the single thread fallback warning has already been
530
     *  displayed, 0 otherwise.
531
     */
532
    int single_decode_warning;
533

    
534
    int last_slice_type;
535
    /** @} */
536

    
537
    int mb_xy;
538

    
539
    uint32_t svq3_watermark_key;
540

    
541
    /**
542
     * pic_struct in picture timing SEI message
543
     */
544
    SEI_PicStructType sei_pic_struct;
545

    
546
    /**
547
     * Complement sei_pic_struct
548
     * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
549
     * However, soft telecined frames may have these values.
550
     * This is used in an attempt to flag soft telecine progressive.
551
     */
552
    int prev_interlaced_frame;
553

    
554
    /**
555
     * Bit set of clock types for fields/frames in picture timing SEI message.
556
     * For each found ct_type, appropriate bit is set (e.g., bit 1 for
557
     * interlaced).
558
     */
559
    int sei_ct_type;
560

    
561
    /**
562
     * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
563
     */
564
    int sei_dpb_output_delay;
565

    
566
    /**
567
     * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
568
     */
569
    int sei_cpb_removal_delay;
570

    
571
    /**
572
     * recovery_frame_cnt from SEI message
573
     *
574
     * Set to -1 if no recovery point SEI message found or to number of frames
575
     * before playback synchronizes. Frames having recovery point are key
576
     * frames.
577
     */
578
    int sei_recovery_frame_cnt;
579

    
580
    int is_complex;
581

    
582
    int luma_weight_flag[2];   ///< 7.4.3.2 luma_weight_lX_flag
583
    int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
584

    
585
    // Timestamp stuff
586
    int sei_buffering_period_present;  ///< Buffering period SEI flag
587
    int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
588
}H264Context;
589

    
590

    
591
extern const uint8_t ff_h264_chroma_qp[52];
592

    
593
void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
594

    
595
void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
596

    
597
/**
598
 * Decode SEI
599
 */
600
int ff_h264_decode_sei(H264Context *h);
601

    
602
/**
603
 * Decode SPS
604
 */
605
int ff_h264_decode_seq_parameter_set(H264Context *h);
606

    
607
/**
608
 * Decode PPS
609
 */
610
int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
611

    
612
/**
613
 * Decodes a network abstraction layer unit.
614
 * @param consumed is the number of bytes used as input
615
 * @param length is the length of the array
616
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
617
 * @returns decoded bytes, might be src+1 if no escapes
618
 */
619
const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
620

    
621
/**
622
 * identifies the exact end of the bitstream
623
 * @return the length of the trailing, or 0 if damaged
624
 */
625
int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
626

    
627
/**
628
 * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
629
 */
630
av_cold void ff_h264_free_context(H264Context *h);
631

    
632
/**
633
 * reconstructs bitstream slice_type.
634
 */
635
int ff_h264_get_slice_type(const H264Context *h);
636

    
637
/**
638
 * allocates tables.
639
 * needs width/height
640
 */
641
int ff_h264_alloc_tables(H264Context *h);
642

    
643
/**
644
 * fills the default_ref_list.
645
 */
646
int ff_h264_fill_default_ref_list(H264Context *h);
647

    
648
int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
649
void ff_h264_fill_mbaff_ref_list(H264Context *h);
650
void ff_h264_remove_all_refs(H264Context *h);
651

    
652
/**
653
 * Executes the reference picture marking (memory management control operations).
654
 */
655
int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
656

    
657
int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
658

    
659

    
660
/**
661
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
662
 */
663
int ff_h264_check_intra4x4_pred_mode(H264Context *h);
664

    
665
/**
666
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
667
 */
668
int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
669

    
670
void ff_h264_write_back_intra_pred_mode(H264Context *h);
671
void ff_h264_hl_decode_mb(H264Context *h);
672
int ff_h264_frame_start(H264Context *h);
673
av_cold int ff_h264_decode_init(AVCodecContext *avctx);
674
av_cold int ff_h264_decode_end(AVCodecContext *avctx);
675
av_cold void ff_h264_decode_init_vlc(void);
676

    
677
/**
678
 * decodes a macroblock
679
 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
680
 */
681
int ff_h264_decode_mb_cavlc(H264Context *h);
682

    
683
/**
684
 * decodes a CABAC coded macroblock
685
 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
686
 */
687
int ff_h264_decode_mb_cabac(H264Context *h);
688

    
689
void ff_h264_init_cabac_states(H264Context *h);
690

    
691
void ff_h264_direct_dist_scale_factor(H264Context * const h);
692
void ff_h264_direct_ref_list_init(H264Context * const h);
693
void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
694

    
695
void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
696
void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
697

    
698
/**
699
 * Reset SEI values at the beginning of the frame.
700
 *
701
 * @param h H.264 context.
702
 */
703
void ff_h264_reset_sei(H264Context *h);
704

    
705

    
706
/*
707
o-o o-o
708
 / / /
709
o-o o-o
710
 ,---'
711
o-o o-o
712
 / / /
713
o-o o-o
714
*/
715
//This table must be here because scan8[constant] must be known at compiletime
716
static const uint8_t scan8[16 + 2*4]={
717
 4+1*8, 5+1*8, 4+2*8, 5+2*8,
718
 6+1*8, 7+1*8, 6+2*8, 7+2*8,
719
 4+3*8, 5+3*8, 4+4*8, 5+4*8,
720
 6+3*8, 7+3*8, 6+4*8, 7+4*8,
721
 1+1*8, 2+1*8,
722
 1+2*8, 2+2*8,
723
 1+4*8, 2+4*8,
724
 1+5*8, 2+5*8,
725
};
726

    
727
static av_always_inline uint32_t pack16to32(int a, int b){
728
#if HAVE_BIGENDIAN
729
   return (b&0xFFFF) + (a<<16);
730
#else
731
   return (a&0xFFFF) + (b<<16);
732
#endif
733
}
734

    
735
/**
736
 * gets the chroma qp.
737
 */
738
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
739
    return h->pps.chroma_qp_table[t][qscale];
740
}
741

    
742
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
743

    
744
static void fill_decode_neighbors(H264Context *h, int mb_type){
745
    MpegEncContext * const s = &h->s;
746
    const int mb_xy= h->mb_xy;
747
    int topleft_xy, top_xy, topright_xy, left_xy[2];
748
    static const uint8_t left_block_options[4][16]={
749
        {0,1,2,3,7,10,8,11,7+0*8, 7+1*8, 7+2*8, 7+3*8, 2+0*8, 2+3*8, 2+1*8, 2+2*8},
750
        {2,2,3,3,8,11,8,11,7+2*8, 7+2*8, 7+3*8, 7+3*8, 2+1*8, 2+2*8, 2+1*8, 2+2*8},
751
        {0,0,1,1,7,10,7,10,7+0*8, 7+0*8, 7+1*8, 7+1*8, 2+0*8, 2+3*8, 2+0*8, 2+3*8},
752
        {0,2,0,2,7,10,7,10,7+0*8, 7+2*8, 7+0*8, 7+2*8, 2+0*8, 2+3*8, 2+0*8, 2+3*8}
753
    };
754

    
755
    h->topleft_partition= -1;
756

    
757
    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
758

    
759
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
760
     * stuff, I can't imagine that these complex rules are worth it. */
761

    
762
    topleft_xy = top_xy - 1;
763
    topright_xy= top_xy + 1;
764
    left_xy[1] = left_xy[0] = mb_xy-1;
765
    h->left_block = left_block_options[0];
766
    if(FRAME_MBAFF){
767
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
768
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
769
        if(s->mb_y&1){
770
            if (left_mb_field_flag != curr_mb_field_flag) {
771
                left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
772
                if (curr_mb_field_flag) {
773
                    left_xy[1] += s->mb_stride;
774
                    h->left_block = left_block_options[3];
775
                } else {
776
                    topleft_xy += s->mb_stride;
777
                    // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
778
                    h->topleft_partition = 0;
779
                    h->left_block = left_block_options[1];
780
                }
781
            }
782
        }else{
783
            if(curr_mb_field_flag){
784
                topleft_xy  += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
785
                topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
786
                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
787
            }
788
            if (left_mb_field_flag != curr_mb_field_flag) {
789
                left_xy[1] = left_xy[0] = mb_xy - 1;
790
                if (curr_mb_field_flag) {
791
                    left_xy[1] += s->mb_stride;
792
                    h->left_block = left_block_options[3];
793
                } else {
794
                    h->left_block = left_block_options[2];
795
                }
796
            }
797
        }
798
    }
799

    
800
    h->topleft_mb_xy = topleft_xy;
801
    h->top_mb_xy     = top_xy;
802
    h->topright_mb_xy= topright_xy;
803
    h->left_mb_xy[0] = left_xy[0];
804
    h->left_mb_xy[1] = left_xy[1];
805
    //FIXME do we need all in the context?
806
    h->topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
807
    h->top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
808
    h->topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
809
    h->left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
810
    h->left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
811
}
812

    
813
static void fill_decode_caches(H264Context *h, int mb_type){
814
    MpegEncContext * const s = &h->s;
815
    const int mb_xy= h->mb_xy;
816
    int topleft_xy, top_xy, topright_xy, left_xy[2];
817
    int topleft_type, top_type, topright_type, left_type[2];
818
    const uint8_t * left_block= h->left_block;
819
    int i;
820

    
821
    topleft_xy   = h->topleft_mb_xy ;
822
    top_xy       = h->top_mb_xy     ;
823
    topright_xy  = h->topright_mb_xy;
824
    left_xy[0]   = h->left_mb_xy[0] ;
825
    left_xy[1]   = h->left_mb_xy[1] ;
826
    topleft_type = h->topleft_type  ;
827
    top_type     = h->top_type      ;
828
    topright_type= h->topright_type ;
829
    left_type[0] = h->left_type[0]  ;
830
    left_type[1] = h->left_type[1]  ;
831

    
832
    if(!IS_SKIP(mb_type)){
833
        if(IS_INTRA(mb_type)){
834
            int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
835
            h->topleft_samples_available=
836
            h->top_samples_available=
837
            h->left_samples_available= 0xFFFF;
838
            h->topright_samples_available= 0xEEEA;
839

    
840
            if(!(top_type & type_mask)){
841
                h->topleft_samples_available= 0xB3FF;
842
                h->top_samples_available= 0x33FF;
843
                h->topright_samples_available= 0x26EA;
844
            }
845
            if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
846
                if(IS_INTERLACED(mb_type)){
847
                    if(!(left_type[0] & type_mask)){
848
                        h->topleft_samples_available&= 0xDFFF;
849
                        h->left_samples_available&= 0x5FFF;
850
                    }
851
                    if(!(left_type[1] & type_mask)){
852
                        h->topleft_samples_available&= 0xFF5F;
853
                        h->left_samples_available&= 0xFF5F;
854
                    }
855
                }else{
856
                    int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
857
                                    ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
858
                    assert(left_xy[0] == left_xy[1]);
859
                    if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
860
                        h->topleft_samples_available&= 0xDF5F;
861
                        h->left_samples_available&= 0x5F5F;
862
                    }
863
                }
864
            }else{
865
                if(!(left_type[0] & type_mask)){
866
                    h->topleft_samples_available&= 0xDF5F;
867
                    h->left_samples_available&= 0x5F5F;
868
                }
869
            }
870

    
871
            if(!(topleft_type & type_mask))
872
                h->topleft_samples_available&= 0x7FFF;
873

    
874
            if(!(topright_type & type_mask))
875
                h->topright_samples_available&= 0xFBFF;
876

    
877
            if(IS_INTRA4x4(mb_type)){
878
                if(IS_INTRA4x4(top_type)){
879
                    h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
880
                    h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
881
                    h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
882
                    h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
883
                }else{
884
                    int pred;
885
                    if(!(top_type & type_mask))
886
                        pred= -1;
887
                    else{
888
                        pred= 2;
889
                    }
890
                    h->intra4x4_pred_mode_cache[4+8*0]=
891
                    h->intra4x4_pred_mode_cache[5+8*0]=
892
                    h->intra4x4_pred_mode_cache[6+8*0]=
893
                    h->intra4x4_pred_mode_cache[7+8*0]= pred;
894
                }
895
                for(i=0; i<2; i++){
896
                    if(IS_INTRA4x4(left_type[i])){
897
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
898
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
899
                    }else{
900
                        int pred;
901
                        if(!(left_type[i] & type_mask))
902
                            pred= -1;
903
                        else{
904
                            pred= 2;
905
                        }
906
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
907
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
908
                    }
909
                }
910
            }
911
        }
912

    
913

    
914
/*
915
0 . T T. T T T T
916
1 L . .L . . . .
917
2 L . .L . . . .
918
3 . T TL . . . .
919
4 L . .L . . . .
920
5 L . .. . . . .
921
*/
922
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
923
    if(top_type){
924
        *(uint32_t*)&h->non_zero_count_cache[4+8*0]= *(uint32_t*)&h->non_zero_count[top_xy][4+3*8];
925
            h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
926
            h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
927

    
928
            h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
929
            h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
930
    }else {
931
            h->non_zero_count_cache[1+8*0]=
932
            h->non_zero_count_cache[2+8*0]=
933

    
934
            h->non_zero_count_cache[1+8*3]=
935
            h->non_zero_count_cache[2+8*3]=
936
            *(uint32_t*)&h->non_zero_count_cache[4+8*0]= CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040;
937
    }
938

    
939
    for (i=0; i<2; i++) {
940
        if(left_type[i]){
941
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
942
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
943
                h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
944
                h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
945
        }else{
946
                h->non_zero_count_cache[3+8*1 + 2*8*i]=
947
                h->non_zero_count_cache[3+8*2 + 2*8*i]=
948
                h->non_zero_count_cache[0+8*1 +   8*i]=
949
                h->non_zero_count_cache[0+8*4 +   8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
950
        }
951
    }
952

    
953
    if( CABAC ) {
954
        // top_cbp
955
        if(top_type) {
956
            h->top_cbp = h->cbp_table[top_xy];
957
        } else if(IS_INTRA(mb_type)) {
958
            h->top_cbp = 0x1CF;
959
        } else {
960
            h->top_cbp = 0x00F;
961
        }
962
        // left_cbp
963
        if (left_type[0]) {
964
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
965
        } else if(IS_INTRA(mb_type)) {
966
            h->left_cbp = 0x1CF;
967
        } else {
968
            h->left_cbp = 0x00F;
969
        }
970
        if (left_type[0]) {
971
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
972
        }
973
        if (left_type[1]) {
974
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
975
        }
976
    }
977
    }
978

    
979
#if 1
980
    if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
981
        int list;
982
        for(list=0; list<h->list_count; list++){
983
            if(!USES_LIST(mb_type, list)){
984
                /*if(!h->mv_cache_clean[list]){
985
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
986
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
987
                    h->mv_cache_clean[list]= 1;
988
                }*/
989
                continue;
990
            }
991
            assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
992

    
993
            h->mv_cache_clean[list]= 0;
994

    
995
            if(USES_LIST(top_type, list)){
996
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
997
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
998
                AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
999
                    h->ref_cache[list][scan8[0] + 0 - 1*8]=
1000
                    h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
1001
                    h->ref_cache[list][scan8[0] + 2 - 1*8]=
1002
                    h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
1003
            }else{
1004
                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1005
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
1006
            }
1007

    
1008
            for(i=0; i<2; i++){
1009
                int cache_idx = scan8[0] - 1 + i*2*8;
1010
                if(USES_LIST(left_type[i], list)){
1011
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1012
                    const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
1013
                    *(uint32_t*)h->mv_cache[list][cache_idx  ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]];
1014
                    *(uint32_t*)h->mv_cache[list][cache_idx+8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]];
1015
                        h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
1016
                        h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
1017
                }else{
1018
                    *(uint32_t*)h->mv_cache [list][cache_idx  ]=
1019
                    *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
1020
                    h->ref_cache[list][cache_idx  ]=
1021
                    h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1022
                }
1023
            }
1024

    
1025
            if(USES_LIST(topleft_type, list)){
1026
                const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
1027
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (h->topleft_partition & h->b8_stride);
1028
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
1029
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1030
            }else{
1031
                *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
1032
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1033
            }
1034

    
1035
            if(USES_LIST(topright_type, list)){
1036
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1037
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
1038
                *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
1039
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1040
            }else{
1041
                *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
1042
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1043
            }
1044

    
1045
            if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
1046
                continue;
1047

    
1048
            if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
1049
            h->ref_cache[list][scan8[5 ]+1] =
1050
            h->ref_cache[list][scan8[7 ]+1] =
1051
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
1052
            h->ref_cache[list][scan8[4 ]] =
1053
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1054
            *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
1055
            *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
1056
            *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
1057
            *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
1058
            *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
1059

    
1060
            if( CABAC ) {
1061
                /* XXX beurk, Load mvd */
1062
                if(USES_LIST(top_type, list)){
1063
                    const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1064
                    AV_COPY128(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1065
                }else{
1066
                    AV_ZERO128(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1067
                }
1068
                if(USES_LIST(left_type[0], list)){
1069
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1070
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
1071
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
1072
                }else{
1073
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
1074
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
1075
                }
1076
                if(USES_LIST(left_type[1], list)){
1077
                    const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
1078
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
1079
                    *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
1080
                }else{
1081
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
1082
                    *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
1083
                }
1084
                *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
1085
                *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
1086
                *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
1087
                *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
1088
                *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
1089

    
1090
                if(h->slice_type_nos == FF_B_TYPE){
1091
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
1092

    
1093
                    if(IS_DIRECT(top_type)){
1094
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101*(MB_TYPE_DIRECT2>>1);
1095
                    }else if(IS_8X8(top_type)){
1096
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
1097
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
1098
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
1099
                    }else{
1100
                        *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101*(MB_TYPE_16x16>>1);
1101
                    }
1102

    
1103
                    if(IS_DIRECT(left_type[0]))
1104
                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
1105
                    else if(IS_8X8(left_type[0]))
1106
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
1107
                    else
1108
                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
1109

    
1110
                    if(IS_DIRECT(left_type[1]))
1111
                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
1112
                    else if(IS_8X8(left_type[1]))
1113
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
1114
                    else
1115
                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
1116
                }
1117
            }
1118
            }
1119
            if(FRAME_MBAFF){
1120
#define MAP_MVS\
1121
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1122
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1123
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1124
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1125
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1126
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1127
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1128
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1129
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1130
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1131
                if(MB_FIELD){
1132
#define MAP_F2F(idx, mb_type)\
1133
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1134
                        h->ref_cache[list][idx] <<= 1;\
1135
                        h->mv_cache[list][idx][1] /= 2;\
1136
                        h->mvd_cache[list][idx][1] /= 2;\
1137
                    }
1138
                    MAP_MVS
1139
#undef MAP_F2F
1140
                }else{
1141
#define MAP_F2F(idx, mb_type)\
1142
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1143
                        h->ref_cache[list][idx] >>= 1;\
1144
                        h->mv_cache[list][idx][1] <<= 1;\
1145
                        h->mvd_cache[list][idx][1] <<= 1;\
1146
                    }
1147
                    MAP_MVS
1148
#undef MAP_F2F
1149
                }
1150
            }
1151
        }
1152
    }
1153
#endif
1154

    
1155
        h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1156
}
1157

    
1158
/**
1159
 *
1160
 * @returns non zero if the loop filter can be skiped
1161
 */
1162
static int fill_filter_caches(H264Context *h, int mb_type){
1163
    MpegEncContext * const s = &h->s;
1164
    const int mb_xy= h->mb_xy;
1165
    int top_xy, left_xy[2];
1166
    int top_type, left_type[2];
1167

    
1168
    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
1169

    
1170
    //FIXME deblocking could skip the intra and nnz parts.
1171

    
1172
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
1173
     * stuff, I can't imagine that these complex rules are worth it. */
1174

    
1175
    left_xy[1] = left_xy[0] = mb_xy-1;
1176
    if(FRAME_MBAFF){
1177
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
1178
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
1179
        if(s->mb_y&1){
1180
            if (left_mb_field_flag != curr_mb_field_flag) {
1181
                left_xy[0] -= s->mb_stride;
1182
            }
1183
        }else{
1184
            if(curr_mb_field_flag){
1185
                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
1186
            }
1187
            if (left_mb_field_flag != curr_mb_field_flag) {
1188
                left_xy[1] += s->mb_stride;
1189
            }
1190
        }
1191
    }
1192

    
1193
    h->top_mb_xy = top_xy;
1194
    h->left_mb_xy[0] = left_xy[0];
1195
    h->left_mb_xy[1] = left_xy[1];
1196
    {
1197
        //for sufficiently low qp, filtering wouldn't do anything
1198
        //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
1199
        int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
1200
        int qp = s->current_picture.qscale_table[mb_xy];
1201
        if(qp <= qp_thresh
1202
           && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
1203
           && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
1204
            if(!FRAME_MBAFF)
1205
                return 1;
1206
            if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
1207
               && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
1208
                return 1;
1209
        }
1210
    }
1211

    
1212
    if(h->deblocking_filter == 2){
1213
        h->top_type    = top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
1214
        h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
1215
        h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
1216
    }else{
1217
        h->top_type    = top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
1218
        h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
1219
        h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
1220
    }
1221
    if(IS_INTRA(mb_type))
1222
        return 0;
1223

    
1224
    AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
1225
    AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
1226
    *((uint32_t*)&h->non_zero_count_cache[0+8*5])= *((uint32_t*)&h->non_zero_count[mb_xy][16]);
1227
    *((uint32_t*)&h->non_zero_count_cache[4+8*3])= *((uint32_t*)&h->non_zero_count[mb_xy][20]);
1228
    AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
1229

    
1230
    h->cbp= h->cbp_table[mb_xy];
1231

    
1232
    {
1233
        int list;
1234
        for(list=0; list<h->list_count; list++){
1235
            int8_t *ref;
1236
            int y, b_stride;
1237
            int16_t (*mv_dst)[2];
1238
            int16_t (*mv_src)[2];
1239

    
1240
            if(!USES_LIST(mb_type, list)){
1241
                fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
1242
                *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
1243
                *(uint32_t*)&h->ref_cache[list][scan8[ 2]] =
1244
                *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
1245
                *(uint32_t*)&h->ref_cache[list][scan8[10]] = ((LIST_NOT_USED)&0xFF)*0x01010101U;
1246
                continue;
1247
            }
1248

    
1249
            ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
1250
            {
1251
                int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1252
                *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
1253
                *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
1254
                ref += h->b8_stride;
1255
                *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
1256
                *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
1257
            }
1258

    
1259
            b_stride = h->b_stride;
1260
            mv_dst   = &h->mv_cache[list][scan8[0]];
1261
            mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
1262
            for(y=0; y<4; y++){
1263
                AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
1264
            }
1265

    
1266
        }
1267
    }
1268

    
1269

    
1270
/*
1271
0 . T T. T T T T
1272
1 L . .L . . . .
1273
2 L . .L . . . .
1274
3 . T TL . . . .
1275
4 L . .L . . . .
1276
5 L . .. . . . .
1277
*/
1278
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
1279
    if(top_type){
1280
        *(uint32_t*)&h->non_zero_count_cache[4+8*0]= *(uint32_t*)&h->non_zero_count[top_xy][4+3*8];
1281
    }
1282

    
1283
    if(left_type[0]){
1284
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
1285
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
1286
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
1287
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
1288
    }
1289

    
1290
    // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
1291
    if(!CABAC && h->pps.transform_8x8_mode){
1292
        if(IS_8x8DCT(top_type)){
1293
            h->non_zero_count_cache[4+8*0]=
1294
            h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
1295
            h->non_zero_count_cache[6+8*0]=
1296
            h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
1297
        }
1298
        if(IS_8x8DCT(left_type[0])){
1299
            h->non_zero_count_cache[3+8*1]=
1300
            h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
1301
        }
1302
        if(IS_8x8DCT(left_type[1])){
1303
            h->non_zero_count_cache[3+8*3]=
1304
            h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
1305
        }
1306

    
1307
        if(IS_8x8DCT(mb_type)){
1308
            h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
1309
            h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
1310

    
1311
            h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
1312
            h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
1313

    
1314
            h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
1315
            h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
1316

    
1317
            h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
1318
            h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
1319
        }
1320
    }
1321

    
1322
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
1323
        int list;
1324
        for(list=0; list<h->list_count; list++){
1325
            if(USES_LIST(top_type, list)){
1326
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1327
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
1328
                int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1329
                AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1330
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
1331
                h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
1332
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
1333
                h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
1334
            }else{
1335
                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1336
                *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((LIST_NOT_USED)&0xFF)*0x01010101U;
1337
            }
1338

    
1339
            if(!IS_INTERLACED(mb_type^left_type[0])){
1340
                if(USES_LIST(left_type[0], list)){
1341
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1342
                    const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
1343
                    int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1344
                    *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 0 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*0];
1345
                    *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 8 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*1];
1346
                    *(uint32_t*)h->mv_cache[list][scan8[0] - 1 +16 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*2];
1347
                    *(uint32_t*)h->mv_cache[list][scan8[0] - 1 +24 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*3];
1348
                    h->ref_cache[list][scan8[0] - 1 + 0 ]=
1349
                    h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + h->b8_stride*0]];
1350
                    h->ref_cache[list][scan8[0] - 1 +16 ]=
1351
                    h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + h->b8_stride*1]];
1352
                }else{
1353
                    *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0 ]=
1354
                    *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 8 ]=
1355
                    *(uint32_t*)h->mv_cache [list][scan8[0] - 1 +16 ]=
1356
                    *(uint32_t*)h->mv_cache [list][scan8[0] - 1 +24 ]= 0;
1357
                    h->ref_cache[list][scan8[0] - 1 + 0  ]=
1358
                    h->ref_cache[list][scan8[0] - 1 + 8  ]=
1359
                    h->ref_cache[list][scan8[0] - 1 + 16 ]=
1360
                    h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
1361
                }
1362
            }
1363
        }
1364
    }
1365

    
1366
    return 0;
1367
}
1368

    
1369
/**
1370
 * gets the predicted intra4x4 prediction mode.
1371
 */
1372
static inline int pred_intra_mode(H264Context *h, int n){
1373
    const int index8= scan8[n];
1374
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1375
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1376
    const int min= FFMIN(left, top);
1377

    
1378
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1379

    
1380
    if(min<0) return DC_PRED;
1381
    else      return min;
1382
}
1383

    
1384
static inline void write_back_non_zero_count(H264Context *h){
1385
    const int mb_xy= h->mb_xy;
1386

    
1387
    AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1388
    AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1389
    *((uint32_t*)&h->non_zero_count[mb_xy][16]) = *((uint32_t*)&h->non_zero_count_cache[0+8*5]);
1390
    *((uint32_t*)&h->non_zero_count[mb_xy][20]) = *((uint32_t*)&h->non_zero_count_cache[4+8*3]);
1391
    AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1392
}
1393

    
1394
static inline void write_back_motion(H264Context *h, int mb_type){
1395
    MpegEncContext * const s = &h->s;
1396
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1397
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1398
    int list;
1399

    
1400
    if(!USES_LIST(mb_type, 0))
1401
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1402

    
1403
    for(list=0; list<h->list_count; list++){
1404
        int y, b_stride;
1405
        int16_t (*mv_dst)[2];
1406
        int16_t (*mv_src)[2];
1407

    
1408
        if(!USES_LIST(mb_type, list))
1409
            continue;
1410

    
1411
        b_stride = h->b_stride;
1412
        mv_dst   = &s->current_picture.motion_val[list][b_xy];
1413
        mv_src   = &h->mv_cache[list][scan8[0]];
1414
        for(y=0; y<4; y++){
1415
            AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1416
        }
1417
        if( CABAC ) {
1418
            int16_t (*mvd_dst)[2] = &h->mvd_table[list][b_xy];
1419
            int16_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1420
            if(IS_SKIP(mb_type))
1421
                fill_rectangle(mvd_dst, 4, 4, h->b_stride, 0, 4);
1422
            else
1423
            for(y=0; y<4; y++){
1424
                AV_COPY128(mvd_dst + y*b_stride, mvd_src + 8*y);
1425
            }
1426
        }
1427

    
1428
        {
1429
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1430
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1431
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1432
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1433
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1434
        }
1435
    }
1436

    
1437
    if(h->slice_type_nos == FF_B_TYPE && CABAC){
1438
        if(IS_8X8(mb_type)){
1439
            uint8_t *direct_table = &h->direct_table[b8_xy];
1440
            direct_table[1+0*h->b8_stride] = h->sub_mb_type[1]>>1;
1441
            direct_table[0+1*h->b8_stride] = h->sub_mb_type[2]>>1;
1442
            direct_table[1+1*h->b8_stride] = h->sub_mb_type[3]>>1;
1443
        }
1444
    }
1445
}
1446

    
1447
static inline int get_dct8x8_allowed(H264Context *h){
1448
    if(h->sps.direct_8x8_inference_flag)
1449
        return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
1450
    else
1451
        return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1452
}
1453

    
1454
static void predict_field_decoding_flag(H264Context *h){
1455
    MpegEncContext * const s = &h->s;
1456
    const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1457
    int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
1458
                ? s->current_picture.mb_type[mb_xy-1]
1459
                : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
1460
                ? s->current_picture.mb_type[mb_xy-s->mb_stride]
1461
                : 0;
1462
    h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
1463
}
1464

    
1465
/**
1466
 * decodes a P_SKIP or B_SKIP macroblock
1467
 */
1468
static void decode_mb_skip(H264Context *h){
1469
    MpegEncContext * const s = &h->s;
1470
    const int mb_xy= h->mb_xy;
1471
    int mb_type=0;
1472

    
1473
    memset(h->non_zero_count[mb_xy], 0, 32);
1474
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1475

    
1476
    if(MB_FIELD)
1477
        mb_type|= MB_TYPE_INTERLACED;
1478

    
1479
    if( h->slice_type_nos == FF_B_TYPE )
1480
    {
1481
        // just for fill_caches. pred_direct_motion will set the real mb_type
1482
        mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1483
        if(h->direct_spatial_mv_pred){
1484
            fill_decode_neighbors(h, mb_type);
1485
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1486
        }
1487
        ff_h264_pred_direct_motion(h, &mb_type);
1488
        mb_type|= MB_TYPE_SKIP;
1489
    }
1490
    else
1491
    {
1492
        int mx, my;
1493
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1494

    
1495
        fill_decode_neighbors(h, mb_type);
1496
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1497
        pred_pskip_motion(h, &mx, &my);
1498
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1499
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1500
    }
1501

    
1502
    write_back_motion(h, mb_type);
1503
    s->current_picture.mb_type[mb_xy]= mb_type;
1504
    s->current_picture.qscale_table[mb_xy]= s->qscale;
1505
    h->slice_table[ mb_xy ]= h->slice_num;
1506
    h->prev_mb_skipped= 1;
1507
}
1508

    
1509
#include "h264_mvpred.h" //For pred_pskip_motion()
1510

    
1511
#endif /* AVCODEC_H264_H */