Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.h @ dc172ecc

History | View | Annotate | Download (46.2 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   24
43
#define CHROMA_DC_BLOCK_INDEX 25
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
    int constraint_set_flags;          ///< constraint_set[0-3]_flag
213
}SPS;
214

    
215
/**
216
 * Picture parameter set
217
 */
218
typedef struct PPS{
219
    unsigned int sps_id;
220
    int cabac;                  ///< entropy_coding_mode_flag
221
    int pic_order_present;      ///< pic_order_present_flag
222
    int slice_group_count;      ///< num_slice_groups_minus1 + 1
223
    int mb_slice_group_map_type;
224
    unsigned int ref_count[2];  ///< num_ref_idx_l0/1_active_minus1 + 1
225
    int weighted_pred;          ///< weighted_pred_flag
226
    int weighted_bipred_idc;
227
    int init_qp;                ///< pic_init_qp_minus26 + 26
228
    int init_qs;                ///< pic_init_qs_minus26 + 26
229
    int chroma_qp_index_offset[2];
230
    int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
231
    int constrained_intra_pred; ///< constrained_intra_pred_flag
232
    int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
233
    int transform_8x8_mode;     ///< transform_8x8_mode_flag
234
    uint8_t scaling_matrix4[6][16];
235
    uint8_t scaling_matrix8[2][64];
236
    uint8_t chroma_qp_table[2][64];  ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
237
    int chroma_qp_diff;
238
}PPS;
239

    
240
/**
241
 * Memory management control operation opcode.
242
 */
243
typedef enum MMCOOpcode{
244
    MMCO_END=0,
245
    MMCO_SHORT2UNUSED,
246
    MMCO_LONG2UNUSED,
247
    MMCO_SHORT2LONG,
248
    MMCO_SET_MAX_LONG,
249
    MMCO_RESET,
250
    MMCO_LONG,
251
} MMCOOpcode;
252

    
253
/**
254
 * Memory management control operation.
255
 */
256
typedef struct MMCO{
257
    MMCOOpcode opcode;
258
    int short_pic_num;  ///< pic_num without wrapping (pic_num & max_pic_num)
259
    int long_arg;       ///< index, pic_num, or num long refs depending on opcode
260
} MMCO;
261

    
262
/**
263
 * H264Context
264
 */
265
typedef struct H264Context{
266
    MpegEncContext s;
267
    H264DSPContext h264dsp;
268
    int pixel_size;
269
    int chroma_qp[2]; //QPc
270

    
271
    int qp_thresh;      ///< QP threshold to skip loopfilter
272

    
273
    int prev_mb_skipped;
274
    int next_mb_skipped;
275

    
276
    //prediction stuff
277
    int chroma_pred_mode;
278
    int intra16x16_pred_mode;
279

    
280
    int topleft_mb_xy;
281
    int top_mb_xy;
282
    int topright_mb_xy;
283
    int left_mb_xy[2];
284

    
285
    int topleft_type;
286
    int top_type;
287
    int topright_type;
288
    int left_type[2];
289

    
290
    const uint8_t * left_block;
291
    int topleft_partition;
292

    
293
    int8_t intra4x4_pred_mode_cache[5*8];
294
    int8_t (*intra4x4_pred_mode);
295
    H264PredContext hpc;
296
    unsigned int topleft_samples_available;
297
    unsigned int top_samples_available;
298
    unsigned int topright_samples_available;
299
    unsigned int left_samples_available;
300
    uint8_t (*top_borders[2])[(16+2*8)*2];
301

    
302
    /**
303
     * non zero coeff count cache.
304
     * is 64 if not available.
305
     */
306
    DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[6*8];
307

    
308
    /*
309
    .UU.YYYY
310
    .UU.YYYY
311
    .vv.YYYY
312
    .VV.YYYY
313
    */
314
    uint8_t (*non_zero_count)[32];
315

    
316
    /**
317
     * Motion vector cache.
318
     */
319
    DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5*8][2];
320
    DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5*8];
321
#define LIST_NOT_USED -1 //FIXME rename?
322
#define PART_NOT_AVAILABLE -2
323

    
324
    /**
325
     * is 1 if the specific list MV&references are set to 0,0,-2.
326
     */
327
    int mv_cache_clean[2];
328

    
329
    /**
330
     * number of neighbors (top and/or left) that used 8x8 dct
331
     */
332
    int neighbor_transform_size;
333

    
334
    /**
335
     * block_offset[ 0..23] for frame macroblocks
336
     * block_offset[24..47] for field macroblocks
337
     */
338
    int block_offset[2*(16+8)];
339

    
340
    uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
341
    uint32_t *mb2br_xy;
342
    int b_stride; //FIXME use s->b4_stride
343

    
344
    int mb_linesize;   ///< may be equal to s->linesize or s->linesize*2, for mbaff
345
    int mb_uvlinesize;
346

    
347
    int emu_edge_width;
348
    int emu_edge_height;
349

    
350
    SPS sps; ///< current sps
351

    
352
    /**
353
     * current pps
354
     */
355
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
356

    
357
    uint32_t dequant4_buffer[6][52][16]; //FIXME should these be moved down?
358
    uint32_t dequant8_buffer[2][52][64];
359
    uint32_t (*dequant4_coeff[6])[16];
360
    uint32_t (*dequant8_coeff[2])[64];
361

    
362
    int slice_num;
363
    uint16_t *slice_table;     ///< slice_table_base + 2*mb_stride + 1
364
    int slice_type;
365
    int slice_type_nos;        ///< S free slice type (SI/SP are remapped to I/P)
366
    int slice_type_fixed;
367

    
368
    //interlacing specific flags
369
    int mb_aff_frame;
370
    int mb_field_decoding_flag;
371
    int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag
372

    
373
    DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
374

    
375
    //Weighted pred stuff
376
    int use_weight;
377
    int use_weight_chroma;
378
    int luma_log2_weight_denom;
379
    int chroma_log2_weight_denom;
380
    //The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
381
    int luma_weight[48][2][2];
382
    int chroma_weight[48][2][2][2];
383
    int implicit_weight[48][48][2];
384

    
385
    int direct_spatial_mv_pred;
386
    int col_parity;
387
    int col_fieldoff;
388
    int dist_scale_factor[16];
389
    int dist_scale_factor_field[2][32];
390
    int map_col_to_list0[2][16+32];
391
    int map_col_to_list0_field[2][2][16+32];
392

    
393
    /**
394
     * num_ref_idx_l0/1_active_minus1 + 1
395
     */
396
    uint8_t *list_counts;            ///< Array of list_count per MB specifying the slice type
397
    unsigned int ref_count[2];   ///< counts frames or fields, depending on current mb mode
398
    unsigned int list_count;
399
    Picture ref_list[2][48];         /**< 0..15: frame refs, 16..47: mbaff field refs.
400
                                          Reordered version of default_ref_list
401
                                          according to picture reordering in slice header */
402
    int ref2frm[MAX_SLICES][2][64];  ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
403

    
404
    //data partitioning
405
    GetBitContext intra_gb;
406
    GetBitContext inter_gb;
407
    GetBitContext *intra_gb_ptr;
408
    GetBitContext *inter_gb_ptr;
409

    
410
    DECLARE_ALIGNED(16, DCTELEM, mb)[16*24*2]; ///< as a dct coeffecient is int32_t in high depth, we need to reserve twice the space.
411
    DECLARE_ALIGNED(16, DCTELEM, mb_luma_dc)[16*2];
412
    DCTELEM mb_padding[256*2];        ///< 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
413

    
414
    /**
415
     * Cabac
416
     */
417
    CABACContext cabac;
418
    uint8_t      cabac_state[460];
419

    
420
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
421
    uint16_t     *cbp_table;
422
    int cbp;
423
    int top_cbp;
424
    int left_cbp;
425
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
426
    uint8_t     *chroma_pred_mode_table;
427
    int         last_qscale_diff;
428
    uint8_t     (*mvd_table[2])[2];
429
    DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5*8][2];
430
    uint8_t     *direct_table;
431
    uint8_t     direct_cache[5*8];
432

    
433
    uint8_t zigzag_scan[16];
434
    uint8_t zigzag_scan8x8[64];
435
    uint8_t zigzag_scan8x8_cavlc[64];
436
    uint8_t field_scan[16];
437
    uint8_t field_scan8x8[64];
438
    uint8_t field_scan8x8_cavlc[64];
439
    const uint8_t *zigzag_scan_q0;
440
    const uint8_t *zigzag_scan8x8_q0;
441
    const uint8_t *zigzag_scan8x8_cavlc_q0;
442
    const uint8_t *field_scan_q0;
443
    const uint8_t *field_scan8x8_q0;
444
    const uint8_t *field_scan8x8_cavlc_q0;
445

    
446
    int x264_build;
447

    
448
    int mb_xy;
449

    
450
    int is_complex;
451

    
452
    //deblock
453
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
454
    int slice_alpha_c0_offset;
455
    int slice_beta_offset;
456

    
457
//=============================================================
458
    //Things below are not used in the MB or more inner code
459

    
460
    int nal_ref_idc;
461
    int nal_unit_type;
462
    uint8_t *rbsp_buffer[2];
463
    unsigned int rbsp_buffer_size[2];
464

    
465
    /**
466
     * Used to parse AVC variant of h264
467
     */
468
    int is_avc; ///< this flag is != 0 if codec is avc1
469
    int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
470
    int got_first; ///< this flag is != 0 if we've parsed a frame
471

    
472
    SPS *sps_buffers[MAX_SPS_COUNT];
473
    PPS *pps_buffers[MAX_PPS_COUNT];
474

    
475
    int dequant_coeff_pps;     ///< reinit tables when pps changes
476

    
477
    uint16_t *slice_table_base;
478

    
479

    
480
    //POC stuff
481
    int poc_lsb;
482
    int poc_msb;
483
    int delta_poc_bottom;
484
    int delta_poc[2];
485
    int frame_num;
486
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
487
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
488
    int frame_num_offset;         ///< for POC type 2
489
    int prev_frame_num_offset;    ///< for POC type 2
490
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
491

    
492
    /**
493
     * frame_num for frames or 2*frame_num+1 for field pics.
494
     */
495
    int curr_pic_num;
496

    
497
    /**
498
     * max_frame_num or 2*max_frame_num for field pics.
499
     */
500
    int max_pic_num;
501

    
502
    int redundant_pic_count;
503

    
504
    Picture *short_ref[32];
505
    Picture *long_ref[32];
506
    Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
507
    Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
508
    Picture *next_output_pic;
509
    int outputed_poc;
510
    int next_outputed_poc;
511

    
512
    /**
513
     * memory management control operations buffer.
514
     */
515
    MMCO mmco[MAX_MMCO_COUNT];
516
    int mmco_index;
517

    
518
    int long_ref_count;  ///< number of actual long term references
519
    int short_ref_count; ///< number of actual short term references
520

    
521
    int          cabac_init_idc;
522

    
523
    /**
524
     * @defgroup multithreading Members for slice based multithreading
525
     * @{
526
     */
527
    struct H264Context *thread_context[MAX_THREADS];
528

    
529
    /**
530
     * current slice number, used to initalize slice_num of each thread/context
531
     */
532
    int current_slice;
533

    
534
    /**
535
     * Max number of threads / contexts.
536
     * This is equal to AVCodecContext.thread_count unless
537
     * multithreaded decoding is impossible, in which case it is
538
     * reduced to 1.
539
     */
540
    int max_contexts;
541

    
542
    /**
543
     *  1 if the single thread fallback warning has already been
544
     *  displayed, 0 otherwise.
545
     */
546
    int single_decode_warning;
547

    
548
    int last_slice_type;
549
    /** @} */
550

    
551
    /**
552
     * pic_struct in picture timing SEI message
553
     */
554
    SEI_PicStructType sei_pic_struct;
555

    
556
    /**
557
     * Complement sei_pic_struct
558
     * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
559
     * However, soft telecined frames may have these values.
560
     * This is used in an attempt to flag soft telecine progressive.
561
     */
562
    int prev_interlaced_frame;
563

    
564
    /**
565
     * Bit set of clock types for fields/frames in picture timing SEI message.
566
     * For each found ct_type, appropriate bit is set (e.g., bit 1 for
567
     * interlaced).
568
     */
569
    int sei_ct_type;
570

    
571
    /**
572
     * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
573
     */
574
    int sei_dpb_output_delay;
575

    
576
    /**
577
     * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
578
     */
579
    int sei_cpb_removal_delay;
580

    
581
    /**
582
     * recovery_frame_cnt from SEI message
583
     *
584
     * Set to -1 if no recovery point SEI message found or to number of frames
585
     * before playback synchronizes. Frames having recovery point are key
586
     * frames.
587
     */
588
    int sei_recovery_frame_cnt;
589

    
590
    int luma_weight_flag[2];   ///< 7.4.3.2 luma_weight_lX_flag
591
    int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
592

    
593
    // Timestamp stuff
594
    int sei_buffering_period_present;  ///< Buffering period SEI flag
595
    int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
596

    
597
    //SVQ3 specific fields
598
    int halfpel_flag;
599
    int thirdpel_flag;
600
    int unknown_svq3_flag;
601
    int next_slice_index;
602
    uint32_t svq3_watermark_key;
603
}H264Context;
604

    
605

    
606
extern const uint8_t ff_h264_chroma_qp[52];
607

    
608
/**
609
 * Decode SEI
610
 */
611
int ff_h264_decode_sei(H264Context *h);
612

    
613
/**
614
 * Decode SPS
615
 */
616
int ff_h264_decode_seq_parameter_set(H264Context *h);
617

    
618
/**
619
 * compute profile from sps
620
 */
621
int ff_h264_get_profile(SPS *sps);
622

    
623
/**
624
 * Decode PPS
625
 */
626
int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
627

    
628
/**
629
 * Decode a network abstraction layer unit.
630
 * @param consumed is the number of bytes used as input
631
 * @param length is the length of the array
632
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
633
 * @return decoded bytes, might be src+1 if no escapes
634
 */
635
const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
636

    
637
/**
638
 * Free any data that may have been allocated in the H264 context like SPS, PPS etc.
639
 */
640
av_cold void ff_h264_free_context(H264Context *h);
641

    
642
/**
643
 * Reconstruct bitstream slice_type.
644
 */
645
int ff_h264_get_slice_type(const H264Context *h);
646

    
647
/**
648
 * Allocate tables.
649
 * needs width/height
650
 */
651
int ff_h264_alloc_tables(H264Context *h);
652

    
653
/**
654
 * Fill the default_ref_list.
655
 */
656
int ff_h264_fill_default_ref_list(H264Context *h);
657

    
658
int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
659
void ff_h264_fill_mbaff_ref_list(H264Context *h);
660
void ff_h264_remove_all_refs(H264Context *h);
661

    
662
/**
663
 * Execute the reference picture marking (memory management control operations).
664
 */
665
int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
666

    
667
int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
668

    
669
void ff_generate_sliding_window_mmcos(H264Context *h);
670

    
671

    
672
/**
673
 * Check if the top & left blocks are available if needed & change the dc mode so it only uses the available blocks.
674
 */
675
int ff_h264_check_intra4x4_pred_mode(H264Context *h);
676

    
677
/**
678
 * Check if the top & left blocks are available if needed & change the dc mode so it only uses the available blocks.
679
 */
680
int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
681

    
682
void ff_h264_write_back_intra_pred_mode(H264Context *h);
683
void ff_h264_hl_decode_mb(H264Context *h);
684
int ff_h264_frame_start(H264Context *h);
685
int ff_h264_decode_extradata(H264Context *h);
686
av_cold int ff_h264_decode_init(AVCodecContext *avctx);
687
av_cold int ff_h264_decode_end(AVCodecContext *avctx);
688
av_cold void ff_h264_decode_init_vlc(void);
689

    
690
/**
691
 * Decode a macroblock
692
 * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
693
 */
694
int ff_h264_decode_mb_cavlc(H264Context *h);
695

    
696
/**
697
 * Decode a CABAC coded macroblock
698
 * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
699
 */
700
int ff_h264_decode_mb_cabac(H264Context *h);
701

    
702
void ff_h264_init_cabac_states(H264Context *h);
703

    
704
void ff_h264_direct_dist_scale_factor(H264Context * const h);
705
void ff_h264_direct_ref_list_init(H264Context * const h);
706
void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
707

    
708
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);
709
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);
710

    
711
/**
712
 * Reset SEI values at the beginning of the frame.
713
 *
714
 * @param h H.264 context.
715
 */
716
void ff_h264_reset_sei(H264Context *h);
717

    
718

    
719
/*
720
o-o o-o
721
 / / /
722
o-o o-o
723
 ,---'
724
o-o o-o
725
 / / /
726
o-o o-o
727
*/
728

    
729
/* Scan8 organization:
730
 *   0 1 2 3 4 5 6 7
731
 * 0   u u y y y y y
732
 * 1 u U U y Y Y Y Y
733
 * 2 u U U y Y Y Y Y
734
 * 3   v v y Y Y Y Y
735
 * 4 v V V y Y Y Y Y
736
 * 5 v V V   DYDUDV
737
 * DY/DU/DV are for luma/chroma DC.
738
 */
739

    
740
//This table must be here because scan8[constant] must be known at compiletime
741
static const uint8_t scan8[16 + 2*4 + 3]={
742
 4+1*8, 5+1*8, 4+2*8, 5+2*8,
743
 6+1*8, 7+1*8, 6+2*8, 7+2*8,
744
 4+3*8, 5+3*8, 4+4*8, 5+4*8,
745
 6+3*8, 7+3*8, 6+4*8, 7+4*8,
746
 1+1*8, 2+1*8,
747
 1+2*8, 2+2*8,
748
 1+4*8, 2+4*8,
749
 1+5*8, 2+5*8,
750
 4+5*8, 5+5*8, 6+5*8
751
};
752

    
753
static av_always_inline uint32_t pack16to32(int a, int b){
754
#if HAVE_BIGENDIAN
755
   return (b&0xFFFF) + (a<<16);
756
#else
757
   return (a&0xFFFF) + (b<<16);
758
#endif
759
}
760

    
761
static av_always_inline uint16_t pack8to16(int a, int b){
762
#if HAVE_BIGENDIAN
763
   return (b&0xFF) + (a<<8);
764
#else
765
   return (a&0xFF) + (b<<8);
766
#endif
767
}
768

    
769
/**
770
 * gets the chroma qp.
771
 */
772
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
773
    return h->pps.chroma_qp_table[t][qscale];
774
}
775

    
776
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
777

    
778
static void fill_decode_neighbors(H264Context *h, int mb_type){
779
    MpegEncContext * const s = &h->s;
780
    const int mb_xy= h->mb_xy;
781
    int topleft_xy, top_xy, topright_xy, left_xy[2];
782
    static const uint8_t left_block_options[4][16]={
783
        {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},
784
        {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},
785
        {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},
786
        {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}
787
    };
788

    
789
    h->topleft_partition= -1;
790

    
791
    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
792

    
793
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
794
     * stuff, I can't imagine that these complex rules are worth it. */
795

    
796
    topleft_xy = top_xy - 1;
797
    topright_xy= top_xy + 1;
798
    left_xy[1] = left_xy[0] = mb_xy-1;
799
    h->left_block = left_block_options[0];
800
    if(FRAME_MBAFF){
801
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
802
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
803
        if(s->mb_y&1){
804
            if (left_mb_field_flag != curr_mb_field_flag) {
805
                left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
806
                if (curr_mb_field_flag) {
807
                    left_xy[1] += s->mb_stride;
808
                    h->left_block = left_block_options[3];
809
                } else {
810
                    topleft_xy += s->mb_stride;
811
                    // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
812
                    h->topleft_partition = 0;
813
                    h->left_block = left_block_options[1];
814
                }
815
            }
816
        }else{
817
            if(curr_mb_field_flag){
818
                topleft_xy  += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
819
                topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
820
                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
821
            }
822
            if (left_mb_field_flag != curr_mb_field_flag) {
823
                if (curr_mb_field_flag) {
824
                    left_xy[1] += s->mb_stride;
825
                    h->left_block = left_block_options[3];
826
                } else {
827
                    h->left_block = left_block_options[2];
828
                }
829
            }
830
        }
831
    }
832

    
833
    h->topleft_mb_xy = topleft_xy;
834
    h->top_mb_xy     = top_xy;
835
    h->topright_mb_xy= topright_xy;
836
    h->left_mb_xy[0] = left_xy[0];
837
    h->left_mb_xy[1] = left_xy[1];
838
    //FIXME do we need all in the context?
839

    
840
    h->topleft_type = s->current_picture.mb_type[topleft_xy] ;
841
    h->top_type     = s->current_picture.mb_type[top_xy]     ;
842
    h->topright_type= s->current_picture.mb_type[topright_xy];
843
    h->left_type[0] = s->current_picture.mb_type[left_xy[0]] ;
844
    h->left_type[1] = s->current_picture.mb_type[left_xy[1]] ;
845

    
846
    if(FMO){
847
    if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0;
848
    if(h->slice_table[top_xy     ] != h->slice_num) h->top_type     = 0;
849
    if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
850
    }else{
851
        if(h->slice_table[topleft_xy ] != h->slice_num){
852
            h->topleft_type = 0;
853
            if(h->slice_table[top_xy     ] != h->slice_num) h->top_type     = 0;
854
            if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
855
        }
856
    }
857
    if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
858
}
859

    
860
static void fill_decode_caches(H264Context *h, int mb_type){
861
    MpegEncContext * const s = &h->s;
862
    int topleft_xy, top_xy, topright_xy, left_xy[2];
863
    int topleft_type, top_type, topright_type, left_type[2];
864
    const uint8_t * left_block= h->left_block;
865
    int i;
866

    
867
    topleft_xy   = h->topleft_mb_xy ;
868
    top_xy       = h->top_mb_xy     ;
869
    topright_xy  = h->topright_mb_xy;
870
    left_xy[0]   = h->left_mb_xy[0] ;
871
    left_xy[1]   = h->left_mb_xy[1] ;
872
    topleft_type = h->topleft_type  ;
873
    top_type     = h->top_type      ;
874
    topright_type= h->topright_type ;
875
    left_type[0] = h->left_type[0]  ;
876
    left_type[1] = h->left_type[1]  ;
877

    
878
    if(!IS_SKIP(mb_type)){
879
        if(IS_INTRA(mb_type)){
880
            int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
881
            h->topleft_samples_available=
882
            h->top_samples_available=
883
            h->left_samples_available= 0xFFFF;
884
            h->topright_samples_available= 0xEEEA;
885

    
886
            if(!(top_type & type_mask)){
887
                h->topleft_samples_available= 0xB3FF;
888
                h->top_samples_available= 0x33FF;
889
                h->topright_samples_available= 0x26EA;
890
            }
891
            if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
892
                if(IS_INTERLACED(mb_type)){
893
                    if(!(left_type[0] & type_mask)){
894
                        h->topleft_samples_available&= 0xDFFF;
895
                        h->left_samples_available&= 0x5FFF;
896
                    }
897
                    if(!(left_type[1] & type_mask)){
898
                        h->topleft_samples_available&= 0xFF5F;
899
                        h->left_samples_available&= 0xFF5F;
900
                    }
901
                }else{
902
                    int left_typei = s->current_picture.mb_type[left_xy[0] + s->mb_stride];
903

    
904
                    assert(left_xy[0] == left_xy[1]);
905
                    if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
906
                        h->topleft_samples_available&= 0xDF5F;
907
                        h->left_samples_available&= 0x5F5F;
908
                    }
909
                }
910
            }else{
911
                if(!(left_type[0] & type_mask)){
912
                    h->topleft_samples_available&= 0xDF5F;
913
                    h->left_samples_available&= 0x5F5F;
914
                }
915
            }
916

    
917
            if(!(topleft_type & type_mask))
918
                h->topleft_samples_available&= 0x7FFF;
919

    
920
            if(!(topright_type & type_mask))
921
                h->topright_samples_available&= 0xFBFF;
922

    
923
            if(IS_INTRA4x4(mb_type)){
924
                if(IS_INTRA4x4(top_type)){
925
                    AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]);
926
                }else{
927
                    h->intra4x4_pred_mode_cache[4+8*0]=
928
                    h->intra4x4_pred_mode_cache[5+8*0]=
929
                    h->intra4x4_pred_mode_cache[6+8*0]=
930
                    h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask);
931
                }
932
                for(i=0; i<2; i++){
933
                    if(IS_INTRA4x4(left_type[i])){
934
                        int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]];
935
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]];
936
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]];
937
                    }else{
938
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
939
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[i] & type_mask);
940
                    }
941
                }
942
            }
943
        }
944

    
945

    
946
/*
947
0 . T T. T T T T
948
1 L . .L . . . .
949
2 L . .L . . . .
950
3 . T TL . . . .
951
4 L . .L . . . .
952
5 L . .. . . . .
953
*/
954
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
955
    if(top_type){
956
        AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
957
            h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
958
            h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
959

    
960
            h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
961
            h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
962
    }else {
963
            h->non_zero_count_cache[1+8*0]=
964
            h->non_zero_count_cache[2+8*0]=
965

    
966
            h->non_zero_count_cache[1+8*3]=
967
            h->non_zero_count_cache[2+8*3]=
968
            AV_WN32A(&h->non_zero_count_cache[4+8*0], CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040);
969
    }
970

    
971
    for (i=0; i<2; i++) {
972
        if(left_type[i]){
973
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
974
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
975
                h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
976
                h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
977
        }else{
978
                h->non_zero_count_cache[3+8*1 + 2*8*i]=
979
                h->non_zero_count_cache[3+8*2 + 2*8*i]=
980
                h->non_zero_count_cache[0+8*1 +   8*i]=
981
                h->non_zero_count_cache[0+8*4 +   8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
982
        }
983
    }
984

    
985
    if( CABAC ) {
986
        // top_cbp
987
        if(top_type) {
988
            h->top_cbp = h->cbp_table[top_xy];
989
        } else {
990
            h->top_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
991
        }
992
        // left_cbp
993
        if (left_type[0]) {
994
            h->left_cbp = (h->cbp_table[left_xy[0]] & 0x1f0)
995
                        |  ((h->cbp_table[left_xy[0]]>>(left_block[0]&(~1)))&2)
996
                        | (((h->cbp_table[left_xy[1]]>>(left_block[2]&(~1)))&2) << 2);
997
        } else {
998
            h->left_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
999
        }
1000
    }
1001
    }
1002

    
1003
#if 1
1004
    if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
1005
        int list;
1006
        for(list=0; list<h->list_count; list++){
1007
            if(!USES_LIST(mb_type, list)){
1008
                /*if(!h->mv_cache_clean[list]){
1009
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
1010
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
1011
                    h->mv_cache_clean[list]= 1;
1012
                }*/
1013
                continue;
1014
            }
1015
            assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
1016

    
1017
            h->mv_cache_clean[list]= 0;
1018

    
1019
            if(USES_LIST(top_type, list)){
1020
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1021
                AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1022
                    h->ref_cache[list][scan8[0] + 0 - 1*8]=
1023
                    h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 2];
1024
                    h->ref_cache[list][scan8[0] + 2 - 1*8]=
1025
                    h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 3];
1026
            }else{
1027
                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1028
                AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
1029
            }
1030

    
1031
            if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){
1032
            for(i=0; i<2; i++){
1033
                int cache_idx = scan8[0] - 1 + i*2*8;
1034
                if(USES_LIST(left_type[i], list)){
1035
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1036
                    const int b8_xy= 4*left_xy[i] + 1;
1037
                    AV_COPY32(h->mv_cache[list][cache_idx  ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]);
1038
                    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]]);
1039
                        h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + (left_block[0+i*2]&~1)];
1040
                        h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + (left_block[1+i*2]&~1)];
1041
                }else{
1042
                    AV_ZERO32(h->mv_cache [list][cache_idx  ]);
1043
                    AV_ZERO32(h->mv_cache [list][cache_idx+8]);
1044
                    h->ref_cache[list][cache_idx  ]=
1045
                    h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1046
                }
1047
            }
1048
            }else{
1049
                if(USES_LIST(left_type[0], list)){
1050
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1051
                    const int b8_xy= 4*left_xy[0] + 1;
1052
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]]);
1053
                    h->ref_cache[list][scan8[0] - 1]= s->current_picture.ref_index[list][b8_xy + (left_block[0]&~1)];
1054
                }else{
1055
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1]);
1056
                    h->ref_cache[list][scan8[0] - 1]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1057
                }
1058
            }
1059

    
1060
            if(USES_LIST(topright_type, list)){
1061
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1062
                AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]);
1063
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][4*topright_xy + 2];
1064
            }else{
1065
                AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]);
1066
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1067
            }
1068
            if(h->ref_cache[list][scan8[0] + 4 - 1*8] < 0){
1069
                if(USES_LIST(topleft_type, list)){
1070
                    const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
1071
                    const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2);
1072
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]);
1073
                    h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1074
                }else{
1075
                    AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]);
1076
                    h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1077
                }
1078
            }
1079

    
1080
            if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
1081
                continue;
1082

    
1083
            if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
1084
            h->ref_cache[list][scan8[4 ]] =
1085
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1086
            AV_ZERO32(h->mv_cache [list][scan8[4 ]]);
1087
            AV_ZERO32(h->mv_cache [list][scan8[12]]);
1088

    
1089
            if( CABAC ) {
1090
                /* XXX beurk, Load mvd */
1091
                if(USES_LIST(top_type, list)){
1092
                    const int b_xy= h->mb2br_xy[top_xy];
1093
                    AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1094
                }else{
1095
                    AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1096
                }
1097
                if(USES_LIST(left_type[0], list)){
1098
                    const int b_xy= h->mb2br_xy[left_xy[0]] + 6;
1099
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy - left_block[0]]);
1100
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy - left_block[1]]);
1101
                }else{
1102
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]);
1103
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]);
1104
                }
1105
                if(USES_LIST(left_type[1], list)){
1106
                    const int b_xy= h->mb2br_xy[left_xy[1]] + 6;
1107
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy - left_block[2]]);
1108
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy - left_block[3]]);
1109
                }else{
1110
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]);
1111
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]);
1112
                }
1113
                AV_ZERO16(h->mvd_cache [list][scan8[4 ]]);
1114
                AV_ZERO16(h->mvd_cache [list][scan8[12]]);
1115
                if(h->slice_type_nos == FF_B_TYPE){
1116
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
1117

    
1118
                    if(IS_DIRECT(top_type)){
1119
                        AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101u*(MB_TYPE_DIRECT2>>1));
1120
                    }else if(IS_8X8(top_type)){
1121
                        int b8_xy = 4*top_xy;
1122
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy + 2];
1123
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 3];
1124
                    }else{
1125
                        AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1));
1126
                    }
1127

    
1128
                    if(IS_DIRECT(left_type[0]))
1129
                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
1130
                    else if(IS_8X8(left_type[0]))
1131
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[4*left_xy[0] + 1 + (left_block[0]&~1)];
1132
                    else
1133
                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
1134

    
1135
                    if(IS_DIRECT(left_type[1]))
1136
                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
1137
                    else if(IS_8X8(left_type[1]))
1138
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[4*left_xy[1] + 1 + (left_block[2]&~1)];
1139
                    else
1140
                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
1141
                }
1142
            }
1143
            }
1144
            if(FRAME_MBAFF){
1145
#define MAP_MVS\
1146
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1147
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1148
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1149
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1150
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1151
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1152
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1153
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1154
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1155
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1156
                if(MB_FIELD){
1157
#define MAP_F2F(idx, mb_type)\
1158
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1159
                        h->ref_cache[list][idx] <<= 1;\
1160
                        h->mv_cache[list][idx][1] /= 2;\
1161
                        h->mvd_cache[list][idx][1] >>=1;\
1162
                    }
1163
                    MAP_MVS
1164
#undef MAP_F2F
1165
                }else{
1166
#define MAP_F2F(idx, mb_type)\
1167
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1168
                        h->ref_cache[list][idx] >>= 1;\
1169
                        h->mv_cache[list][idx][1] <<= 1;\
1170
                        h->mvd_cache[list][idx][1] <<= 1;\
1171
                    }
1172
                    MAP_MVS
1173
#undef MAP_F2F
1174
                }
1175
            }
1176
        }
1177
    }
1178
#endif
1179

    
1180
        h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1181
}
1182

    
1183
/**
1184
 * gets the predicted intra4x4 prediction mode.
1185
 */
1186
static inline int pred_intra_mode(H264Context *h, int n){
1187
    const int index8= scan8[n];
1188
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1189
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1190
    const int min= FFMIN(left, top);
1191

    
1192
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1193

    
1194
    if(min<0) return DC_PRED;
1195
    else      return min;
1196
}
1197

    
1198
static inline void write_back_non_zero_count(H264Context *h){
1199
    const int mb_xy= h->mb_xy;
1200

    
1201
    AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1202
    AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1203
    AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[0+8*5]);
1204
    AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8*3]);
1205
    AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1206
}
1207

    
1208
static inline void write_back_motion(H264Context *h, int mb_type){
1209
    MpegEncContext * const s = &h->s;
1210
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
1211
    const int b8_xy= 4*h->mb_xy;
1212
    int list;
1213

    
1214
    if(!USES_LIST(mb_type, 0))
1215
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1216

    
1217
    for(list=0; list<h->list_count; list++){
1218
        int y, b_stride;
1219
        int16_t (*mv_dst)[2];
1220
        int16_t (*mv_src)[2];
1221

    
1222
        if(!USES_LIST(mb_type, list))
1223
            continue;
1224

    
1225
        b_stride = h->b_stride;
1226
        mv_dst   = &s->current_picture.motion_val[list][b_xy];
1227
        mv_src   = &h->mv_cache[list][scan8[0]];
1228
        for(y=0; y<4; y++){
1229
            AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1230
        }
1231
        if( CABAC ) {
1232
            uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
1233
            uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1234
            if(IS_SKIP(mb_type))
1235
                AV_ZERO128(mvd_dst);
1236
            else{
1237
            AV_COPY64(mvd_dst, mvd_src + 8*3);
1238
                AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
1239
                AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
1240
                AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
1241
            }
1242
        }
1243

    
1244
        {
1245
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1246
            ref_index[0+0*2]= h->ref_cache[list][scan8[0]];
1247
            ref_index[1+0*2]= h->ref_cache[list][scan8[4]];
1248
            ref_index[0+1*2]= h->ref_cache[list][scan8[8]];
1249
            ref_index[1+1*2]= h->ref_cache[list][scan8[12]];
1250
        }
1251
    }
1252

    
1253
    if(h->slice_type_nos == FF_B_TYPE && CABAC){
1254
        if(IS_8X8(mb_type)){
1255
            uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
1256
            direct_table[1] = h->sub_mb_type[1]>>1;
1257
            direct_table[2] = h->sub_mb_type[2]>>1;
1258
            direct_table[3] = h->sub_mb_type[3]>>1;
1259
        }
1260
    }
1261
}
1262

    
1263
static inline int get_dct8x8_allowed(H264Context *h){
1264
    if(h->sps.direct_8x8_inference_flag)
1265
        return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
1266
    else
1267
        return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1268
}
1269

    
1270
/**
1271
 * decodes a P_SKIP or B_SKIP macroblock
1272
 */
1273
static void av_unused decode_mb_skip(H264Context *h){
1274
    MpegEncContext * const s = &h->s;
1275
    const int mb_xy= h->mb_xy;
1276
    int mb_type=0;
1277

    
1278
    memset(h->non_zero_count[mb_xy], 0, 32);
1279
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1280

    
1281
    if(MB_FIELD)
1282
        mb_type|= MB_TYPE_INTERLACED;
1283

    
1284
    if( h->slice_type_nos == FF_B_TYPE )
1285
    {
1286
        // just for fill_caches. pred_direct_motion will set the real mb_type
1287
        mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1288
        if(h->direct_spatial_mv_pred){
1289
            fill_decode_neighbors(h, mb_type);
1290
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1291
        }
1292
        ff_h264_pred_direct_motion(h, &mb_type);
1293
        mb_type|= MB_TYPE_SKIP;
1294
    }
1295
    else
1296
    {
1297
        int mx, my;
1298
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1299

    
1300
        fill_decode_neighbors(h, mb_type);
1301
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1302
        pred_pskip_motion(h, &mx, &my);
1303
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1304
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1305
    }
1306

    
1307
    write_back_motion(h, mb_type);
1308
    s->current_picture.mb_type[mb_xy]= mb_type;
1309
    s->current_picture.qscale_table[mb_xy]= s->qscale;
1310
    h->slice_table[ mb_xy ]= h->slice_num;
1311
    h->prev_mb_skipped= 1;
1312
}
1313

    
1314
#include "h264_mvpred.h" //For pred_pskip_motion()
1315

    
1316
#endif /* AVCODEC_H264_H */