Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.h @ 19fb234e

History | View | Annotate | Download (45.8 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
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 "libavutil/intreadwrite.h"
32
#include "dsputil.h"
33
#include "cabac.h"
34
#include "mpegvideo.h"
35
#include "h264dsp.h"
36
#include "h264pred.h"
37
#include "rectangle.h"
38

    
39
#define interlaced_dct interlaced_dct_is_a_bad_name
40
#define mb_intra mb_intra_is_not_initialized_see_mb_type
41

    
42
#define LUMA_DC_BLOCK_INDEX   25
43
#define CHROMA_DC_BLOCK_INDEX 26
44

    
45
#define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
46
#define COEFF_TOKEN_VLC_BITS           8
47
#define TOTAL_ZEROS_VLC_BITS           9
48
#define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
49
#define RUN_VLC_BITS                   3
50
#define RUN7_VLC_BITS                  6
51

    
52
#define MAX_SPS_COUNT 32
53
#define MAX_PPS_COUNT 256
54

    
55
#define MAX_MMCO_COUNT 66
56

    
57
#define MAX_DELAYED_PIC_COUNT 16
58

    
59
/* Compiling in interlaced support reduces the speed
60
 * of progressive decoding by about 2%. */
61
#define ALLOW_INTERLACE
62

    
63
#define ALLOW_NOCHROMA
64

    
65
#define FMO 0
66

    
67
/**
68
 * The maximum number of slices supported by the decoder.
69
 * must be a power of 2
70
 */
71
#define MAX_SLICES 16
72

    
73
#ifdef ALLOW_INTERLACE
74
#define MB_MBAFF h->mb_mbaff
75
#define MB_FIELD h->mb_field_decoding_flag
76
#define FRAME_MBAFF h->mb_aff_frame
77
#define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
78
#else
79
#define MB_MBAFF 0
80
#define MB_FIELD 0
81
#define FRAME_MBAFF 0
82
#define FIELD_PICTURE 0
83
#undef  IS_INTERLACED
84
#define IS_INTERLACED(mb_type) 0
85
#endif
86
#define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
87

    
88
#ifdef ALLOW_NOCHROMA
89
#define CHROMA h->sps.chroma_format_idc
90
#else
91
#define CHROMA 1
92
#endif
93

    
94
#ifndef CABAC
95
#define CABAC h->pps.cabac
96
#endif
97

    
98
#define EXTENDED_SAR          255
99

    
100
#define MB_TYPE_REF0       MB_TYPE_ACPRED //dirty but it fits in 16 bit
101
#define MB_TYPE_8x8DCT     0x01000000
102
#define IS_REF0(a)         ((a) & MB_TYPE_REF0)
103
#define IS_8x8DCT(a)       ((a) & MB_TYPE_8x8DCT)
104

    
105
/**
106
 * Value of Picture.reference when Picture is not a reference picture, but
107
 * is held for delayed output.
108
 */
109
#define DELAYED_PIC_REF 4
110

    
111

    
112
/* NAL unit types */
113
enum {
114
    NAL_SLICE=1,
115
    NAL_DPA,
116
    NAL_DPB,
117
    NAL_DPC,
118
    NAL_IDR_SLICE,
119
    NAL_SEI,
120
    NAL_SPS,
121
    NAL_PPS,
122
    NAL_AUD,
123
    NAL_END_SEQUENCE,
124
    NAL_END_STREAM,
125
    NAL_FILLER_DATA,
126
    NAL_SPS_EXT,
127
    NAL_AUXILIARY_SLICE=19
128
};
129

    
130
/**
131
 * SEI message types
132
 */
133
typedef enum {
134
    SEI_BUFFERING_PERIOD             =  0, ///< buffering period (H.264, D.1.1)
135
    SEI_TYPE_PIC_TIMING              =  1, ///< picture timing
136
    SEI_TYPE_USER_DATA_UNREGISTERED  =  5, ///< unregistered user data
137
    SEI_TYPE_RECOVERY_POINT          =  6  ///< recovery point (frame # to decoder sync)
138
} SEI_Type;
139

    
140
/**
141
 * pic_struct in picture timing SEI message
142
 */
143
typedef enum {
144
    SEI_PIC_STRUCT_FRAME             = 0, ///<  0: %frame
145
    SEI_PIC_STRUCT_TOP_FIELD         = 1, ///<  1: top field
146
    SEI_PIC_STRUCT_BOTTOM_FIELD      = 2, ///<  2: bottom field
147
    SEI_PIC_STRUCT_TOP_BOTTOM        = 3, ///<  3: top field, bottom field, in that order
148
    SEI_PIC_STRUCT_BOTTOM_TOP        = 4, ///<  4: bottom field, top field, in that order
149
    SEI_PIC_STRUCT_TOP_BOTTOM_TOP    = 5, ///<  5: top field, bottom field, top field repeated, in that order
150
    SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///<  6: bottom field, top field, bottom field repeated, in that order
151
    SEI_PIC_STRUCT_FRAME_DOUBLING    = 7, ///<  7: %frame doubling
152
    SEI_PIC_STRUCT_FRAME_TRIPLING    = 8  ///<  8: %frame tripling
153
} SEI_PicStructType;
154

    
155
/**
156
 * Sequence parameter set
157
 */
158
typedef struct SPS{
159

    
160
    int profile_idc;
161
    int level_idc;
162
    int chroma_format_idc;
163
    int transform_bypass;              ///< qpprime_y_zero_transform_bypass_flag
164
    int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
165
    int poc_type;                      ///< pic_order_cnt_type
166
    int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
167
    int delta_pic_order_always_zero_flag;
168
    int offset_for_non_ref_pic;
169
    int offset_for_top_to_bottom_field;
170
    int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
171
    int ref_frame_count;               ///< num_ref_frames
172
    int gaps_in_frame_num_allowed_flag;
173
    int mb_width;                      ///< pic_width_in_mbs_minus1 + 1
174
    int mb_height;                     ///< pic_height_in_map_units_minus1 + 1
175
    int frame_mbs_only_flag;
176
    int mb_aff;                        ///<mb_adaptive_frame_field_flag
177
    int direct_8x8_inference_flag;
178
    int crop;                   ///< frame_cropping_flag
179
    unsigned int crop_left;            ///< frame_cropping_rect_left_offset
180
    unsigned int crop_right;           ///< frame_cropping_rect_right_offset
181
    unsigned int crop_top;             ///< frame_cropping_rect_top_offset
182
    unsigned int crop_bottom;          ///< frame_cropping_rect_bottom_offset
183
    int vui_parameters_present_flag;
184
    AVRational sar;
185
    int video_signal_type_present_flag;
186
    int full_range;
187
    int colour_description_present_flag;
188
    enum AVColorPrimaries color_primaries;
189
    enum AVColorTransferCharacteristic color_trc;
190
    enum AVColorSpace colorspace;
191
    int timing_info_present_flag;
192
    uint32_t num_units_in_tick;
193
    uint32_t time_scale;
194
    int fixed_frame_rate_flag;
195
    short offset_for_ref_frame[256]; //FIXME dyn aloc?
196
    int bitstream_restriction_flag;
197
    int num_reorder_frames;
198
    int scaling_matrix_present;
199
    uint8_t scaling_matrix4[6][16];
200
    uint8_t scaling_matrix8[2][64];
201
    int nal_hrd_parameters_present_flag;
202
    int vcl_hrd_parameters_present_flag;
203
    int pic_struct_present_flag;
204
    int time_offset_length;
205
    int cpb_cnt;                       ///< See H.264 E.1.2
206
    int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1
207
    int cpb_removal_delay_length;      ///< cpb_removal_delay_length_minus1 + 1
208
    int dpb_output_delay_length;       ///< dpb_output_delay_length_minus1 + 1
209
    int bit_depth_luma;                ///< bit_depth_luma_minus8 + 8
210
    int bit_depth_chroma;              ///< bit_depth_chroma_minus8 + 8
211
    int residual_color_transform_flag; ///< residual_colour_transform_flag
212
}SPS;
213

    
214
/**
215
 * Picture parameter set
216
 */
217
typedef struct PPS{
218
    unsigned int sps_id;
219
    int cabac;                  ///< entropy_coding_mode_flag
220
    int pic_order_present;      ///< pic_order_present_flag
221
    int slice_group_count;      ///< num_slice_groups_minus1 + 1
222
    int mb_slice_group_map_type;
223
    unsigned int ref_count[2];  ///< num_ref_idx_l0/1_active_minus1 + 1
224
    int weighted_pred;          ///< weighted_pred_flag
225
    int weighted_bipred_idc;
226
    int init_qp;                ///< pic_init_qp_minus26 + 26
227
    int init_qs;                ///< pic_init_qs_minus26 + 26
228
    int chroma_qp_index_offset[2];
229
    int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
230
    int constrained_intra_pred; ///< constrained_intra_pred_flag
231
    int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
232
    int transform_8x8_mode;     ///< transform_8x8_mode_flag
233
    uint8_t scaling_matrix4[6][16];
234
    uint8_t scaling_matrix8[2][64];
235
    uint8_t chroma_qp_table[2][64];  ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
236
    int chroma_qp_diff;
237
}PPS;
238

    
239
/**
240
 * Memory management control operation opcode.
241
 */
242
typedef enum MMCOOpcode{
243
    MMCO_END=0,
244
    MMCO_SHORT2UNUSED,
245
    MMCO_LONG2UNUSED,
246
    MMCO_SHORT2LONG,
247
    MMCO_SET_MAX_LONG,
248
    MMCO_RESET,
249
    MMCO_LONG,
250
} MMCOOpcode;
251

    
252
/**
253
 * Memory management control operation.
254
 */
255
typedef struct MMCO{
256
    MMCOOpcode opcode;
257
    int short_pic_num;  ///< pic_num without wrapping (pic_num & max_pic_num)
258
    int long_arg;       ///< index, pic_num, or num long refs depending on opcode
259
} MMCO;
260

    
261
/**
262
 * H264Context
263
 */
264
typedef struct H264Context{
265
    MpegEncContext s;
266
    H264DSPContext h264dsp;
267
    int chroma_qp[2]; //QPc
268

    
269
    int qp_thresh;      ///< QP threshold to skip loopfilter
270

    
271
    int prev_mb_skipped;
272
    int next_mb_skipped;
273

    
274
    //prediction stuff
275
    int chroma_pred_mode;
276
    int intra16x16_pred_mode;
277

    
278
    int topleft_mb_xy;
279
    int top_mb_xy;
280
    int topright_mb_xy;
281
    int left_mb_xy[2];
282

    
283
    int topleft_type;
284
    int top_type;
285
    int topright_type;
286
    int left_type[2];
287

    
288
    const uint8_t * left_block;
289
    int topleft_partition;
290

    
291
    int8_t intra4x4_pred_mode_cache[5*8];
292
    int8_t (*intra4x4_pred_mode);
293
    H264PredContext hpc;
294
    unsigned int topleft_samples_available;
295
    unsigned int top_samples_available;
296
    unsigned int topright_samples_available;
297
    unsigned int left_samples_available;
298
    uint8_t (*top_borders[2])[16+2*8];
299

    
300
    /**
301
     * non zero coeff count cache.
302
     * is 64 if not available.
303
     */
304
    DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[6*8];
305

    
306
    /*
307
    .UU.YYYY
308
    .UU.YYYY
309
    .vv.YYYY
310
    .VV.YYYY
311
    */
312
    uint8_t (*non_zero_count)[32];
313

    
314
    /**
315
     * Motion vector cache.
316
     */
317
    DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5*8][2];
318
    DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5*8];
319
#define LIST_NOT_USED -1 //FIXME rename?
320
#define PART_NOT_AVAILABLE -2
321

    
322
    /**
323
     * is 1 if the specific list MV&references are set to 0,0,-2.
324
     */
325
    int mv_cache_clean[2];
326

    
327
    /**
328
     * number of neighbors (top and/or left) that used 8x8 dct
329
     */
330
    int neighbor_transform_size;
331

    
332
    /**
333
     * block_offset[ 0..23] for frame macroblocks
334
     * block_offset[24..47] for field macroblocks
335
     */
336
    int block_offset[2*(16+8)];
337

    
338
    uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
339
    uint32_t *mb2br_xy;
340
    int b_stride; //FIXME use s->b4_stride
341

    
342
    int mb_linesize;   ///< may be equal to s->linesize or s->linesize*2, for mbaff
343
    int mb_uvlinesize;
344

    
345
    int emu_edge_width;
346
    int emu_edge_height;
347

    
348
    SPS sps; ///< current sps
349

    
350
    /**
351
     * current pps
352
     */
353
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
354

    
355
    uint32_t dequant4_buffer[6][52][16]; //FIXME should these be moved down?
356
    uint32_t dequant8_buffer[2][52][64];
357
    uint32_t (*dequant4_coeff[6])[16];
358
    uint32_t (*dequant8_coeff[2])[64];
359

    
360
    int slice_num;
361
    uint16_t *slice_table;     ///< slice_table_base + 2*mb_stride + 1
362
    int slice_type;
363
    int slice_type_nos;        ///< S free slice type (SI/SP are remapped to I/P)
364
    int slice_type_fixed;
365

    
366
    //interlacing specific flags
367
    int mb_aff_frame;
368
    int mb_field_decoding_flag;
369
    int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag
370

    
371
    DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
372

    
373
    //Weighted pred stuff
374
    int use_weight;
375
    int use_weight_chroma;
376
    int luma_log2_weight_denom;
377
    int chroma_log2_weight_denom;
378
    //The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
379
    int luma_weight[48][2][2];
380
    int chroma_weight[48][2][2][2];
381
    int implicit_weight[48][48][2];
382

    
383
    int direct_spatial_mv_pred;
384
    int col_parity;
385
    int col_fieldoff;
386
    int dist_scale_factor[16];
387
    int dist_scale_factor_field[2][32];
388
    int map_col_to_list0[2][16+32];
389
    int map_col_to_list0_field[2][2][16+32];
390

    
391
    /**
392
     * num_ref_idx_l0/1_active_minus1 + 1
393
     */
394
    unsigned int ref_count[2];   ///< counts frames or fields, depending on current mb mode
395
    unsigned int list_count;
396
    uint8_t *list_counts;            ///< Array of list_count per MB specifying the slice type
397
    Picture ref_list[2][48];         /**< 0..15: frame refs, 16..47: mbaff field refs.
398
                                          Reordered version of default_ref_list
399
                                          according to picture reordering in slice header */
400
    int ref2frm[MAX_SLICES][2][64];  ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
401

    
402
    //data partitioning
403
    GetBitContext intra_gb;
404
    GetBitContext inter_gb;
405
    GetBitContext *intra_gb_ptr;
406
    GetBitContext *inter_gb_ptr;
407

    
408
    DECLARE_ALIGNED(16, DCTELEM, mb)[16*24];
409
    DECLARE_ALIGNED(16, DCTELEM, mb_luma_dc)[16];
410
    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
411

    
412
    /**
413
     * Cabac
414
     */
415
    CABACContext cabac;
416
    uint8_t      cabac_state[460];
417

    
418
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
419
    uint16_t     *cbp_table;
420
    int cbp;
421
    int top_cbp;
422
    int left_cbp;
423
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
424
    uint8_t     *chroma_pred_mode_table;
425
    int         last_qscale_diff;
426
    uint8_t     (*mvd_table[2])[2];
427
    DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5*8][2];
428
    uint8_t     *direct_table;
429
    uint8_t     direct_cache[5*8];
430

    
431
    uint8_t zigzag_scan[16];
432
    uint8_t zigzag_scan8x8[64];
433
    uint8_t zigzag_scan8x8_cavlc[64];
434
    uint8_t field_scan[16];
435
    uint8_t field_scan8x8[64];
436
    uint8_t field_scan8x8_cavlc[64];
437
    const uint8_t *zigzag_scan_q0;
438
    const uint8_t *zigzag_scan8x8_q0;
439
    const uint8_t *zigzag_scan8x8_cavlc_q0;
440
    const uint8_t *field_scan_q0;
441
    const uint8_t *field_scan8x8_q0;
442
    const uint8_t *field_scan8x8_cavlc_q0;
443

    
444
    int x264_build;
445

    
446
    int mb_xy;
447

    
448
    int is_complex;
449

    
450
    //deblock
451
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
452
    int slice_alpha_c0_offset;
453
    int slice_beta_offset;
454

    
455
//=============================================================
456
    //Things below are not used in the MB or more inner code
457

    
458
    int nal_ref_idc;
459
    int nal_unit_type;
460
    uint8_t *rbsp_buffer[2];
461
    unsigned int rbsp_buffer_size[2];
462

    
463
    /**
464
     * Used to parse AVC variant of h264
465
     */
466
    int is_avc; ///< this flag is != 0 if codec is avc1
467
    int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
468
    int got_first; ///< this flag is != 0 if we've parsed a frame
469

    
470
    SPS *sps_buffers[MAX_SPS_COUNT];
471
    PPS *pps_buffers[MAX_PPS_COUNT];
472

    
473
    int dequant_coeff_pps;     ///< reinit tables when pps changes
474

    
475
    uint16_t *slice_table_base;
476

    
477

    
478
    //POC stuff
479
    int poc_lsb;
480
    int poc_msb;
481
    int delta_poc_bottom;
482
    int delta_poc[2];
483
    int frame_num;
484
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
485
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
486
    int frame_num_offset;         ///< for POC type 2
487
    int prev_frame_num_offset;    ///< for POC type 2
488
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
489

    
490
    /**
491
     * frame_num for frames or 2*frame_num+1 for field pics.
492
     */
493
    int curr_pic_num;
494

    
495
    /**
496
     * max_frame_num or 2*max_frame_num for field pics.
497
     */
498
    int max_pic_num;
499

    
500
    int redundant_pic_count;
501

    
502
    Picture *short_ref[32];
503
    Picture *long_ref[32];
504
    Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
505
    Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
506
    int outputed_poc;
507

    
508
    /**
509
     * memory management control operations buffer.
510
     */
511
    MMCO mmco[MAX_MMCO_COUNT];
512
    int mmco_index;
513

    
514
    int long_ref_count;  ///< number of actual long term references
515
    int short_ref_count; ///< number of actual short term references
516

    
517
    int          cabac_init_idc;
518

    
519
    /**
520
     * @defgroup multithreading Members for slice based multithreading
521
     * @{
522
     */
523
    struct H264Context *thread_context[MAX_THREADS];
524

    
525
    /**
526
     * current slice number, used to initalize slice_num of each thread/context
527
     */
528
    int current_slice;
529

    
530
    /**
531
     * Max number of threads / contexts.
532
     * This is equal to AVCodecContext.thread_count unless
533
     * multithreaded decoding is impossible, in which case it is
534
     * reduced to 1.
535
     */
536
    int max_contexts;
537

    
538
    /**
539
     *  1 if the single thread fallback warning has already been
540
     *  displayed, 0 otherwise.
541
     */
542
    int single_decode_warning;
543

    
544
    int last_slice_type;
545
    /** @} */
546

    
547
    /**
548
     * pic_struct in picture timing SEI message
549
     */
550
    SEI_PicStructType sei_pic_struct;
551

    
552
    /**
553
     * Complement sei_pic_struct
554
     * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
555
     * However, soft telecined frames may have these values.
556
     * This is used in an attempt to flag soft telecine progressive.
557
     */
558
    int prev_interlaced_frame;
559

    
560
    /**
561
     * Bit set of clock types for fields/frames in picture timing SEI message.
562
     * For each found ct_type, appropriate bit is set (e.g., bit 1 for
563
     * interlaced).
564
     */
565
    int sei_ct_type;
566

    
567
    /**
568
     * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
569
     */
570
    int sei_dpb_output_delay;
571

    
572
    /**
573
     * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
574
     */
575
    int sei_cpb_removal_delay;
576

    
577
    /**
578
     * recovery_frame_cnt from SEI message
579
     *
580
     * Set to -1 if no recovery point SEI message found or to number of frames
581
     * before playback synchronizes. Frames having recovery point are key
582
     * frames.
583
     */
584
    int sei_recovery_frame_cnt;
585

    
586
    int luma_weight_flag[2];   ///< 7.4.3.2 luma_weight_lX_flag
587
    int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
588

    
589
    // Timestamp stuff
590
    int sei_buffering_period_present;  ///< Buffering period SEI flag
591
    int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
592

    
593
    //SVQ3 specific fields
594
    int halfpel_flag;
595
    int thirdpel_flag;
596
    int unknown_svq3_flag;
597
    int next_slice_index;
598
    uint32_t svq3_watermark_key;
599
}H264Context;
600

    
601

    
602
extern const uint8_t ff_h264_chroma_qp[52];
603

    
604
/**
605
 * Decode SEI
606
 */
607
int ff_h264_decode_sei(H264Context *h);
608

    
609
/**
610
 * Decode SPS
611
 */
612
int ff_h264_decode_seq_parameter_set(H264Context *h);
613

    
614
/**
615
 * Decode PPS
616
 */
617
int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
618

    
619
/**
620
 * Decode a network abstraction layer unit.
621
 * @param consumed is the number of bytes used as input
622
 * @param length is the length of the array
623
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
624
 * @return decoded bytes, might be src+1 if no escapes
625
 */
626
const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
627

    
628
/**
629
 * Identify the exact end of the bitstream
630
 * @return the length of the trailing, or 0 if damaged
631
 */
632
int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
633

    
634
/**
635
 * Free any data that may have been allocated in the H264 context like SPS, PPS etc.
636
 */
637
av_cold void ff_h264_free_context(H264Context *h);
638

    
639
/**
640
 * Reconstruct bitstream slice_type.
641
 */
642
int ff_h264_get_slice_type(const H264Context *h);
643

    
644
/**
645
 * Allocate tables.
646
 * needs width/height
647
 */
648
int ff_h264_alloc_tables(H264Context *h);
649

    
650
/**
651
 * Fill the default_ref_list.
652
 */
653
int ff_h264_fill_default_ref_list(H264Context *h);
654

    
655
int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
656
void ff_h264_fill_mbaff_ref_list(H264Context *h);
657
void ff_h264_remove_all_refs(H264Context *h);
658

    
659
/**
660
 * Execute the reference picture marking (memory management control operations).
661
 */
662
int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
663

    
664
int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
665

    
666
void ff_generate_sliding_window_mmcos(H264Context *h);
667

    
668

    
669
/**
670
 * Check if the top & left blocks are available if needed & change the dc mode so it only uses the available blocks.
671
 */
672
int ff_h264_check_intra4x4_pred_mode(H264Context *h);
673

    
674
/**
675
 * Check if the top & left blocks are available if needed & change the dc mode so it only uses the available blocks.
676
 */
677
int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
678

    
679
void ff_h264_write_back_intra_pred_mode(H264Context *h);
680
void ff_h264_hl_decode_mb(H264Context *h);
681
int ff_h264_frame_start(H264Context *h);
682
int ff_h264_decode_extradata(H264Context *h);
683
av_cold int ff_h264_decode_init(AVCodecContext *avctx);
684
av_cold int ff_h264_decode_end(AVCodecContext *avctx);
685
av_cold void ff_h264_decode_init_vlc(void);
686

    
687
/**
688
 * Decode a macroblock
689
 * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
690
 */
691
int ff_h264_decode_mb_cavlc(H264Context *h);
692

    
693
/**
694
 * Decode a CABAC coded macroblock
695
 * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
696
 */
697
int ff_h264_decode_mb_cabac(H264Context *h);
698

    
699
void ff_h264_init_cabac_states(H264Context *h);
700

    
701
void ff_h264_direct_dist_scale_factor(H264Context * const h);
702
void ff_h264_direct_ref_list_init(H264Context * const h);
703
void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
704

    
705
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);
706
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);
707

    
708
/**
709
 * Reset SEI values at the beginning of the frame.
710
 *
711
 * @param h H.264 context.
712
 */
713
void ff_h264_reset_sei(H264Context *h);
714

    
715

    
716
/*
717
o-o o-o
718
 / / /
719
o-o o-o
720
 ,---'
721
o-o o-o
722
 / / /
723
o-o o-o
724
*/
725
//This table must be here because scan8[constant] must be known at compiletime
726
static const uint8_t scan8[16 + 2*4]={
727
 4+1*8, 5+1*8, 4+2*8, 5+2*8,
728
 6+1*8, 7+1*8, 6+2*8, 7+2*8,
729
 4+3*8, 5+3*8, 4+4*8, 5+4*8,
730
 6+3*8, 7+3*8, 6+4*8, 7+4*8,
731
 1+1*8, 2+1*8,
732
 1+2*8, 2+2*8,
733
 1+4*8, 2+4*8,
734
 1+5*8, 2+5*8,
735
};
736

    
737
static av_always_inline uint32_t pack16to32(int a, int b){
738
#if HAVE_BIGENDIAN
739
   return (b&0xFFFF) + (a<<16);
740
#else
741
   return (a&0xFFFF) + (b<<16);
742
#endif
743
}
744

    
745
static av_always_inline uint16_t pack8to16(int a, int b){
746
#if HAVE_BIGENDIAN
747
   return (b&0xFF) + (a<<8);
748
#else
749
   return (a&0xFF) + (b<<8);
750
#endif
751
}
752

    
753
/**
754
 * gets the chroma qp.
755
 */
756
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
757
    return h->pps.chroma_qp_table[t][qscale];
758
}
759

    
760
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
761

    
762
static void fill_decode_neighbors(H264Context *h, int mb_type){
763
    MpegEncContext * const s = &h->s;
764
    const int mb_xy= h->mb_xy;
765
    int topleft_xy, top_xy, topright_xy, left_xy[2];
766
    static const uint8_t left_block_options[4][16]={
767
        {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},
768
        {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},
769
        {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},
770
        {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}
771
    };
772

    
773
    h->topleft_partition= -1;
774

    
775
    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
776

    
777
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
778
     * stuff, I can't imagine that these complex rules are worth it. */
779

    
780
    topleft_xy = top_xy - 1;
781
    topright_xy= top_xy + 1;
782
    left_xy[1] = left_xy[0] = mb_xy-1;
783
    h->left_block = left_block_options[0];
784
    if(FRAME_MBAFF){
785
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
786
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
787
        if(s->mb_y&1){
788
            if (left_mb_field_flag != curr_mb_field_flag) {
789
                left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 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
                    topleft_xy += s->mb_stride;
795
                    // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
796
                    h->topleft_partition = 0;
797
                    h->left_block = left_block_options[1];
798
                }
799
            }
800
        }else{
801
            if(curr_mb_field_flag){
802
                topleft_xy  += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
803
                topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
804
                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
805
            }
806
            if (left_mb_field_flag != curr_mb_field_flag) {
807
                if (curr_mb_field_flag) {
808
                    left_xy[1] += s->mb_stride;
809
                    h->left_block = left_block_options[3];
810
                } else {
811
                    h->left_block = left_block_options[2];
812
                }
813
            }
814
        }
815
    }
816

    
817
    h->topleft_mb_xy = topleft_xy;
818
    h->top_mb_xy     = top_xy;
819
    h->topright_mb_xy= topright_xy;
820
    h->left_mb_xy[0] = left_xy[0];
821
    h->left_mb_xy[1] = left_xy[1];
822
    //FIXME do we need all in the context?
823

    
824
    h->topleft_type = s->current_picture.mb_type[topleft_xy] ;
825
    h->top_type     = s->current_picture.mb_type[top_xy]     ;
826
    h->topright_type= s->current_picture.mb_type[topright_xy];
827
    h->left_type[0] = s->current_picture.mb_type[left_xy[0]] ;
828
    h->left_type[1] = s->current_picture.mb_type[left_xy[1]] ;
829

    
830
    if(FMO){
831
    if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0;
832
    if(h->slice_table[top_xy     ] != h->slice_num) h->top_type     = 0;
833
    if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
834
    }else{
835
        if(h->slice_table[topleft_xy ] != h->slice_num){
836
            h->topleft_type = 0;
837
            if(h->slice_table[top_xy     ] != h->slice_num) h->top_type     = 0;
838
            if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
839
        }
840
    }
841
    if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
842
}
843

    
844
static void fill_decode_caches(H264Context *h, int mb_type){
845
    MpegEncContext * const s = &h->s;
846
    int topleft_xy, top_xy, topright_xy, left_xy[2];
847
    int topleft_type, top_type, topright_type, left_type[2];
848
    const uint8_t * left_block= h->left_block;
849
    int i;
850

    
851
    topleft_xy   = h->topleft_mb_xy ;
852
    top_xy       = h->top_mb_xy     ;
853
    topright_xy  = h->topright_mb_xy;
854
    left_xy[0]   = h->left_mb_xy[0] ;
855
    left_xy[1]   = h->left_mb_xy[1] ;
856
    topleft_type = h->topleft_type  ;
857
    top_type     = h->top_type      ;
858
    topright_type= h->topright_type ;
859
    left_type[0] = h->left_type[0]  ;
860
    left_type[1] = h->left_type[1]  ;
861

    
862
    if(!IS_SKIP(mb_type)){
863
        if(IS_INTRA(mb_type)){
864
            int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
865
            h->topleft_samples_available=
866
            h->top_samples_available=
867
            h->left_samples_available= 0xFFFF;
868
            h->topright_samples_available= 0xEEEA;
869

    
870
            if(!(top_type & type_mask)){
871
                h->topleft_samples_available= 0xB3FF;
872
                h->top_samples_available= 0x33FF;
873
                h->topright_samples_available= 0x26EA;
874
            }
875
            if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
876
                if(IS_INTERLACED(mb_type)){
877
                    if(!(left_type[0] & type_mask)){
878
                        h->topleft_samples_available&= 0xDFFF;
879
                        h->left_samples_available&= 0x5FFF;
880
                    }
881
                    if(!(left_type[1] & type_mask)){
882
                        h->topleft_samples_available&= 0xFF5F;
883
                        h->left_samples_available&= 0xFF5F;
884
                    }
885
                }else{
886
                    int left_typei = s->current_picture.mb_type[left_xy[0] + s->mb_stride];
887

    
888
                    assert(left_xy[0] == left_xy[1]);
889
                    if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
890
                        h->topleft_samples_available&= 0xDF5F;
891
                        h->left_samples_available&= 0x5F5F;
892
                    }
893
                }
894
            }else{
895
                if(!(left_type[0] & type_mask)){
896
                    h->topleft_samples_available&= 0xDF5F;
897
                    h->left_samples_available&= 0x5F5F;
898
                }
899
            }
900

    
901
            if(!(topleft_type & type_mask))
902
                h->topleft_samples_available&= 0x7FFF;
903

    
904
            if(!(topright_type & type_mask))
905
                h->topright_samples_available&= 0xFBFF;
906

    
907
            if(IS_INTRA4x4(mb_type)){
908
                if(IS_INTRA4x4(top_type)){
909
                    AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]);
910
                }else{
911
                    h->intra4x4_pred_mode_cache[4+8*0]=
912
                    h->intra4x4_pred_mode_cache[5+8*0]=
913
                    h->intra4x4_pred_mode_cache[6+8*0]=
914
                    h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask);
915
                }
916
                for(i=0; i<2; i++){
917
                    if(IS_INTRA4x4(left_type[i])){
918
                        int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]];
919
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]];
920
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]];
921
                    }else{
922
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
923
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[i] & type_mask);
924
                    }
925
                }
926
            }
927
        }
928

    
929

    
930
/*
931
0 . T T. T T T T
932
1 L . .L . . . .
933
2 L . .L . . . .
934
3 . T TL . . . .
935
4 L . .L . . . .
936
5 L . .. . . . .
937
*/
938
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
939
    if(top_type){
940
        AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
941
            h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
942
            h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
943

    
944
            h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
945
            h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
946
    }else {
947
            h->non_zero_count_cache[1+8*0]=
948
            h->non_zero_count_cache[2+8*0]=
949

    
950
            h->non_zero_count_cache[1+8*3]=
951
            h->non_zero_count_cache[2+8*3]=
952
            AV_WN32A(&h->non_zero_count_cache[4+8*0], CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040);
953
    }
954

    
955
    for (i=0; i<2; i++) {
956
        if(left_type[i]){
957
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
958
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
959
                h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
960
                h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
961
        }else{
962
                h->non_zero_count_cache[3+8*1 + 2*8*i]=
963
                h->non_zero_count_cache[3+8*2 + 2*8*i]=
964
                h->non_zero_count_cache[0+8*1 +   8*i]=
965
                h->non_zero_count_cache[0+8*4 +   8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
966
        }
967
    }
968

    
969
    if( CABAC ) {
970
        // top_cbp
971
        if(top_type) {
972
            h->top_cbp = h->cbp_table[top_xy];
973
        } else {
974
            h->top_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
975
        }
976
        // left_cbp
977
        if (left_type[0]) {
978
            h->left_cbp = (h->cbp_table[left_xy[0]] & 0x1f0)
979
                        |  ((h->cbp_table[left_xy[0]]>>(left_block[0]&(~1)))&2)
980
                        | (((h->cbp_table[left_xy[1]]>>(left_block[2]&(~1)))&2) << 2);
981
        } else {
982
            h->left_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
983
        }
984
    }
985
    }
986

    
987
#if 1
988
    if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
989
        int list;
990
        for(list=0; list<h->list_count; list++){
991
            if(!USES_LIST(mb_type, list)){
992
                /*if(!h->mv_cache_clean[list]){
993
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
994
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
995
                    h->mv_cache_clean[list]= 1;
996
                }*/
997
                continue;
998
            }
999
            assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
1000

    
1001
            h->mv_cache_clean[list]= 0;
1002

    
1003
            if(USES_LIST(top_type, list)){
1004
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1005
                AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1006
                    h->ref_cache[list][scan8[0] + 0 - 1*8]=
1007
                    h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 2];
1008
                    h->ref_cache[list][scan8[0] + 2 - 1*8]=
1009
                    h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 3];
1010
            }else{
1011
                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1012
                AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
1013
            }
1014

    
1015
            if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){
1016
            for(i=0; i<2; i++){
1017
                int cache_idx = scan8[0] - 1 + i*2*8;
1018
                if(USES_LIST(left_type[i], list)){
1019
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1020
                    const int b8_xy= 4*left_xy[i] + 1;
1021
                    AV_COPY32(h->mv_cache[list][cache_idx  ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]);
1022
                    AV_COPY32(h->mv_cache[list][cache_idx+8], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]]);
1023
                        h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + (left_block[0+i*2]&~1)];
1024
                        h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + (left_block[1+i*2]&~1)];
1025
                }else{
1026
                    AV_ZERO32(h->mv_cache [list][cache_idx  ]);
1027
                    AV_ZERO32(h->mv_cache [list][cache_idx+8]);
1028
                    h->ref_cache[list][cache_idx  ]=
1029
                    h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1030
                }
1031
            }
1032
            }else{
1033
                if(USES_LIST(left_type[0], list)){
1034
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1035
                    const int b8_xy= 4*left_xy[0] + 1;
1036
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]]);
1037
                    h->ref_cache[list][scan8[0] - 1]= s->current_picture.ref_index[list][b8_xy + (left_block[0]&~1)];
1038
                }else{
1039
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1]);
1040
                    h->ref_cache[list][scan8[0] - 1]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1041
                }
1042
            }
1043

    
1044
            if(USES_LIST(topright_type, list)){
1045
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1046
                AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]);
1047
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][4*topright_xy + 2];
1048
            }else{
1049
                AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]);
1050
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1051
            }
1052
            if(h->ref_cache[list][scan8[0] + 4 - 1*8] < 0){
1053
                if(USES_LIST(topleft_type, list)){
1054
                    const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
1055
                    const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2);
1056
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]);
1057
                    h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1058
                }else{
1059
                    AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]);
1060
                    h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1061
                }
1062
            }
1063

    
1064
            if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
1065
                continue;
1066

    
1067
            if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
1068
            h->ref_cache[list][scan8[4 ]] =
1069
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1070
            AV_ZERO32(h->mv_cache [list][scan8[4 ]]);
1071
            AV_ZERO32(h->mv_cache [list][scan8[12]]);
1072

    
1073
            if( CABAC ) {
1074
                /* XXX beurk, Load mvd */
1075
                if(USES_LIST(top_type, list)){
1076
                    const int b_xy= h->mb2br_xy[top_xy];
1077
                    AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1078
                }else{
1079
                    AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1080
                }
1081
                if(USES_LIST(left_type[0], list)){
1082
                    const int b_xy= h->mb2br_xy[left_xy[0]] + 6;
1083
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy - left_block[0]]);
1084
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy - left_block[1]]);
1085
                }else{
1086
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]);
1087
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]);
1088
                }
1089
                if(USES_LIST(left_type[1], list)){
1090
                    const int b_xy= h->mb2br_xy[left_xy[1]] + 6;
1091
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy - left_block[2]]);
1092
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy - left_block[3]]);
1093
                }else{
1094
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]);
1095
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]);
1096
                }
1097
                AV_ZERO16(h->mvd_cache [list][scan8[4 ]]);
1098
                AV_ZERO16(h->mvd_cache [list][scan8[12]]);
1099
                if(h->slice_type_nos == FF_B_TYPE){
1100
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
1101

    
1102
                    if(IS_DIRECT(top_type)){
1103
                        AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101u*(MB_TYPE_DIRECT2>>1));
1104
                    }else if(IS_8X8(top_type)){
1105
                        int b8_xy = 4*top_xy;
1106
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy + 2];
1107
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 3];
1108
                    }else{
1109
                        AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1));
1110
                    }
1111

    
1112
                    if(IS_DIRECT(left_type[0]))
1113
                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
1114
                    else if(IS_8X8(left_type[0]))
1115
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[4*left_xy[0] + 1 + (left_block[0]&~1)];
1116
                    else
1117
                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
1118

    
1119
                    if(IS_DIRECT(left_type[1]))
1120
                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
1121
                    else if(IS_8X8(left_type[1]))
1122
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[4*left_xy[1] + 1 + (left_block[2]&~1)];
1123
                    else
1124
                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
1125
                }
1126
            }
1127
            }
1128
            if(FRAME_MBAFF){
1129
#define MAP_MVS\
1130
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1131
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1132
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1133
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1134
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1135
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1136
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1137
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1138
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1139
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1140
                if(MB_FIELD){
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] /= 2;\
1145
                        h->mvd_cache[list][idx][1] >>=1;\
1146
                    }
1147
                    MAP_MVS
1148
#undef MAP_F2F
1149
                }else{
1150
#define MAP_F2F(idx, mb_type)\
1151
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1152
                        h->ref_cache[list][idx] >>= 1;\
1153
                        h->mv_cache[list][idx][1] <<= 1;\
1154
                        h->mvd_cache[list][idx][1] <<= 1;\
1155
                    }
1156
                    MAP_MVS
1157
#undef MAP_F2F
1158
                }
1159
            }
1160
        }
1161
    }
1162
#endif
1163

    
1164
        h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1165
}
1166

    
1167
/**
1168
 * gets the predicted intra4x4 prediction mode.
1169
 */
1170
static inline int pred_intra_mode(H264Context *h, int n){
1171
    const int index8= scan8[n];
1172
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1173
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1174
    const int min= FFMIN(left, top);
1175

    
1176
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1177

    
1178
    if(min<0) return DC_PRED;
1179
    else      return min;
1180
}
1181

    
1182
static inline void write_back_non_zero_count(H264Context *h){
1183
    const int mb_xy= h->mb_xy;
1184

    
1185
    AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1186
    AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1187
    AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[0+8*5]);
1188
    AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8*3]);
1189
    AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1190
}
1191

    
1192
static inline void write_back_motion(H264Context *h, int mb_type){
1193
    MpegEncContext * const s = &h->s;
1194
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
1195
    const int b8_xy= 4*h->mb_xy;
1196
    int list;
1197

    
1198
    if(!USES_LIST(mb_type, 0))
1199
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1200

    
1201
    for(list=0; list<h->list_count; list++){
1202
        int y, b_stride;
1203
        int16_t (*mv_dst)[2];
1204
        int16_t (*mv_src)[2];
1205

    
1206
        if(!USES_LIST(mb_type, list))
1207
            continue;
1208

    
1209
        b_stride = h->b_stride;
1210
        mv_dst   = &s->current_picture.motion_val[list][b_xy];
1211
        mv_src   = &h->mv_cache[list][scan8[0]];
1212
        for(y=0; y<4; y++){
1213
            AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1214
        }
1215
        if( CABAC ) {
1216
            uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
1217
            uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1218
            if(IS_SKIP(mb_type))
1219
                AV_ZERO128(mvd_dst);
1220
            else{
1221
            AV_COPY64(mvd_dst, mvd_src + 8*3);
1222
                AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
1223
                AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
1224
                AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
1225
            }
1226
        }
1227

    
1228
        {
1229
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1230
            ref_index[0+0*2]= h->ref_cache[list][scan8[0]];
1231
            ref_index[1+0*2]= h->ref_cache[list][scan8[4]];
1232
            ref_index[0+1*2]= h->ref_cache[list][scan8[8]];
1233
            ref_index[1+1*2]= h->ref_cache[list][scan8[12]];
1234
        }
1235
    }
1236

    
1237
    if(h->slice_type_nos == FF_B_TYPE && CABAC){
1238
        if(IS_8X8(mb_type)){
1239
            uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
1240
            direct_table[1] = h->sub_mb_type[1]>>1;
1241
            direct_table[2] = h->sub_mb_type[2]>>1;
1242
            direct_table[3] = h->sub_mb_type[3]>>1;
1243
        }
1244
    }
1245
}
1246

    
1247
static inline int get_dct8x8_allowed(H264Context *h){
1248
    if(h->sps.direct_8x8_inference_flag)
1249
        return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
1250
    else
1251
        return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1252
}
1253

    
1254
/**
1255
 * decodes a P_SKIP or B_SKIP macroblock
1256
 */
1257
static void av_unused decode_mb_skip(H264Context *h){
1258
    MpegEncContext * const s = &h->s;
1259
    const int mb_xy= h->mb_xy;
1260
    int mb_type=0;
1261

    
1262
    memset(h->non_zero_count[mb_xy], 0, 32);
1263
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1264

    
1265
    if(MB_FIELD)
1266
        mb_type|= MB_TYPE_INTERLACED;
1267

    
1268
    if( h->slice_type_nos == FF_B_TYPE )
1269
    {
1270
        // just for fill_caches. pred_direct_motion will set the real mb_type
1271
        mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1272
        if(h->direct_spatial_mv_pred){
1273
            fill_decode_neighbors(h, mb_type);
1274
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1275
        }
1276
        ff_h264_pred_direct_motion(h, &mb_type);
1277
        mb_type|= MB_TYPE_SKIP;
1278
    }
1279
    else
1280
    {
1281
        int mx, my;
1282
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1283

    
1284
        fill_decode_neighbors(h, mb_type);
1285
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1286
        pred_pskip_motion(h, &mx, &my);
1287
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1288
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1289
    }
1290

    
1291
    write_back_motion(h, mb_type);
1292
    s->current_picture.mb_type[mb_xy]= mb_type;
1293
    s->current_picture.qscale_table[mb_xy]= s->qscale;
1294
    h->slice_table[ mb_xy ]= h->slice_num;
1295
    h->prev_mb_skipped= 1;
1296
}
1297

    
1298
#include "h264_mvpred.h" //For pred_pskip_motion()
1299

    
1300
#endif /* AVCODEC_H264_H */