Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.h @ 2912e87a

History | View | Annotate | Download (46 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 Libav.
6
 *
7
 * Libav 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
 * Libav 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 Libav; 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 chroma_qp[2]; //QPc
269

    
270
    int qp_thresh;      ///< QP threshold to skip loopfilter
271

    
272
    int prev_mb_skipped;
273
    int next_mb_skipped;
274

    
275
    //prediction stuff
276
    int chroma_pred_mode;
277
    int intra16x16_pred_mode;
278

    
279
    int topleft_mb_xy;
280
    int top_mb_xy;
281
    int topright_mb_xy;
282
    int left_mb_xy[2];
283

    
284
    int topleft_type;
285
    int top_type;
286
    int topright_type;
287
    int left_type[2];
288

    
289
    const uint8_t * left_block;
290
    int topleft_partition;
291

    
292
    int8_t intra4x4_pred_mode_cache[5*8];
293
    int8_t (*intra4x4_pred_mode);
294
    H264PredContext hpc;
295
    unsigned int topleft_samples_available;
296
    unsigned int top_samples_available;
297
    unsigned int topright_samples_available;
298
    unsigned int left_samples_available;
299
    uint8_t (*top_borders[2])[16+2*8];
300

    
301
    /**
302
     * non zero coeff count cache.
303
     * is 64 if not available.
304
     */
305
    DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[6*8];
306

    
307
    /*
308
    .UU.YYYY
309
    .UU.YYYY
310
    .vv.YYYY
311
    .VV.YYYY
312
    */
313
    uint8_t (*non_zero_count)[32];
314

    
315
    /**
316
     * Motion vector cache.
317
     */
318
    DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5*8][2];
319
    DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5*8];
320
#define LIST_NOT_USED -1 //FIXME rename?
321
#define PART_NOT_AVAILABLE -2
322

    
323
    /**
324
     * is 1 if the specific list MV&references are set to 0,0,-2.
325
     */
326
    int mv_cache_clean[2];
327

    
328
    /**
329
     * number of neighbors (top and/or left) that used 8x8 dct
330
     */
331
    int neighbor_transform_size;
332

    
333
    /**
334
     * block_offset[ 0..23] for frame macroblocks
335
     * block_offset[24..47] for field macroblocks
336
     */
337
    int block_offset[2*(16+8)];
338

    
339
    uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
340
    uint32_t *mb2br_xy;
341
    int b_stride; //FIXME use s->b4_stride
342

    
343
    int mb_linesize;   ///< may be equal to s->linesize or s->linesize*2, for mbaff
344
    int mb_uvlinesize;
345

    
346
    int emu_edge_width;
347
    int emu_edge_height;
348

    
349
    SPS sps; ///< current sps
350

    
351
    /**
352
     * current pps
353
     */
354
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
355

    
356
    uint32_t dequant4_buffer[6][52][16]; //FIXME should these be moved down?
357
    uint32_t dequant8_buffer[2][52][64];
358
    uint32_t (*dequant4_coeff[6])[16];
359
    uint32_t (*dequant8_coeff[2])[64];
360

    
361
    int slice_num;
362
    uint16_t *slice_table;     ///< slice_table_base + 2*mb_stride + 1
363
    int slice_type;
364
    int slice_type_nos;        ///< S free slice type (SI/SP are remapped to I/P)
365
    int slice_type_fixed;
366

    
367
    //interlacing specific flags
368
    int mb_aff_frame;
369
    int mb_field_decoding_flag;
370
    int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag
371

    
372
    DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
373

    
374
    //Weighted pred stuff
375
    int use_weight;
376
    int use_weight_chroma;
377
    int luma_log2_weight_denom;
378
    int chroma_log2_weight_denom;
379
    //The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
380
    int luma_weight[48][2][2];
381
    int chroma_weight[48][2][2][2];
382
    int implicit_weight[48][48][2];
383

    
384
    int direct_spatial_mv_pred;
385
    int col_parity;
386
    int col_fieldoff;
387
    int dist_scale_factor[16];
388
    int dist_scale_factor_field[2][32];
389
    int map_col_to_list0[2][16+32];
390
    int map_col_to_list0_field[2][2][16+32];
391

    
392
    /**
393
     * num_ref_idx_l0/1_active_minus1 + 1
394
     */
395
    unsigned int ref_count[2];   ///< counts frames or fields, depending on current mb mode
396
    unsigned int list_count;
397
    uint8_t *list_counts;            ///< Array of list_count per MB specifying the slice type
398
    Picture ref_list[2][48];         /**< 0..15: frame refs, 16..47: mbaff field refs.
399
                                          Reordered version of default_ref_list
400
                                          according to picture reordering in slice header */
401
    int ref2frm[MAX_SLICES][2][64];  ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
402

    
403
    //data partitioning
404
    GetBitContext intra_gb;
405
    GetBitContext inter_gb;
406
    GetBitContext *intra_gb_ptr;
407
    GetBitContext *inter_gb_ptr;
408

    
409
    DECLARE_ALIGNED(16, DCTELEM, mb)[16*24];
410
    DECLARE_ALIGNED(16, DCTELEM, mb_luma_dc)[16];
411
    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
412

    
413
    /**
414
     * Cabac
415
     */
416
    CABACContext cabac;
417
    uint8_t      cabac_state[460];
418

    
419
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
420
    uint16_t     *cbp_table;
421
    int cbp;
422
    int top_cbp;
423
    int left_cbp;
424
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
425
    uint8_t     *chroma_pred_mode_table;
426
    int         last_qscale_diff;
427
    uint8_t     (*mvd_table[2])[2];
428
    DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5*8][2];
429
    uint8_t     *direct_table;
430
    uint8_t     direct_cache[5*8];
431

    
432
    uint8_t zigzag_scan[16];
433
    uint8_t zigzag_scan8x8[64];
434
    uint8_t zigzag_scan8x8_cavlc[64];
435
    uint8_t field_scan[16];
436
    uint8_t field_scan8x8[64];
437
    uint8_t field_scan8x8_cavlc[64];
438
    const uint8_t *zigzag_scan_q0;
439
    const uint8_t *zigzag_scan8x8_q0;
440
    const uint8_t *zigzag_scan8x8_cavlc_q0;
441
    const uint8_t *field_scan_q0;
442
    const uint8_t *field_scan8x8_q0;
443
    const uint8_t *field_scan8x8_cavlc_q0;
444

    
445
    int x264_build;
446

    
447
    int mb_xy;
448

    
449
    int is_complex;
450

    
451
    //deblock
452
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
453
    int slice_alpha_c0_offset;
454
    int slice_beta_offset;
455

    
456
//=============================================================
457
    //Things below are not used in the MB or more inner code
458

    
459
    int nal_ref_idc;
460
    int nal_unit_type;
461
    uint8_t *rbsp_buffer[2];
462
    unsigned int rbsp_buffer_size[2];
463

    
464
    /**
465
     * Used to parse AVC variant of h264
466
     */
467
    int is_avc; ///< this flag is != 0 if codec is avc1
468
    int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
469
    int got_first; ///< this flag is != 0 if we've parsed a frame
470

    
471
    SPS *sps_buffers[MAX_SPS_COUNT];
472
    PPS *pps_buffers[MAX_PPS_COUNT];
473

    
474
    int dequant_coeff_pps;     ///< reinit tables when pps changes
475

    
476
    uint16_t *slice_table_base;
477

    
478

    
479
    //POC stuff
480
    int poc_lsb;
481
    int poc_msb;
482
    int delta_poc_bottom;
483
    int delta_poc[2];
484
    int frame_num;
485
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
486
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
487
    int frame_num_offset;         ///< for POC type 2
488
    int prev_frame_num_offset;    ///< for POC type 2
489
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
490

    
491
    /**
492
     * frame_num for frames or 2*frame_num+1 for field pics.
493
     */
494
    int curr_pic_num;
495

    
496
    /**
497
     * max_frame_num or 2*max_frame_num for field pics.
498
     */
499
    int max_pic_num;
500

    
501
    int redundant_pic_count;
502

    
503
    Picture *short_ref[32];
504
    Picture *long_ref[32];
505
    Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
506
    Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
507
    int outputed_poc;
508

    
509
    /**
510
     * memory management control operations buffer.
511
     */
512
    MMCO mmco[MAX_MMCO_COUNT];
513
    int mmco_index;
514

    
515
    int long_ref_count;  ///< number of actual long term references
516
    int short_ref_count; ///< number of actual short term references
517

    
518
    int          cabac_init_idc;
519

    
520
    /**
521
     * @defgroup multithreading Members for slice based multithreading
522
     * @{
523
     */
524
    struct H264Context *thread_context[MAX_THREADS];
525

    
526
    /**
527
     * current slice number, used to initalize slice_num of each thread/context
528
     */
529
    int current_slice;
530

    
531
    /**
532
     * Max number of threads / contexts.
533
     * This is equal to AVCodecContext.thread_count unless
534
     * multithreaded decoding is impossible, in which case it is
535
     * reduced to 1.
536
     */
537
    int max_contexts;
538

    
539
    /**
540
     *  1 if the single thread fallback warning has already been
541
     *  displayed, 0 otherwise.
542
     */
543
    int single_decode_warning;
544

    
545
    int last_slice_type;
546
    /** @} */
547

    
548
    /**
549
     * pic_struct in picture timing SEI message
550
     */
551
    SEI_PicStructType sei_pic_struct;
552

    
553
    /**
554
     * Complement sei_pic_struct
555
     * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
556
     * However, soft telecined frames may have these values.
557
     * This is used in an attempt to flag soft telecine progressive.
558
     */
559
    int prev_interlaced_frame;
560

    
561
    /**
562
     * Bit set of clock types for fields/frames in picture timing SEI message.
563
     * For each found ct_type, appropriate bit is set (e.g., bit 1 for
564
     * interlaced).
565
     */
566
    int sei_ct_type;
567

    
568
    /**
569
     * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
570
     */
571
    int sei_dpb_output_delay;
572

    
573
    /**
574
     * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
575
     */
576
    int sei_cpb_removal_delay;
577

    
578
    /**
579
     * recovery_frame_cnt from SEI message
580
     *
581
     * Set to -1 if no recovery point SEI message found or to number of frames
582
     * before playback synchronizes. Frames having recovery point are key
583
     * frames.
584
     */
585
    int sei_recovery_frame_cnt;
586

    
587
    int luma_weight_flag[2];   ///< 7.4.3.2 luma_weight_lX_flag
588
    int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
589

    
590
    // Timestamp stuff
591
    int sei_buffering_period_present;  ///< Buffering period SEI flag
592
    int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
593

    
594
    //SVQ3 specific fields
595
    int halfpel_flag;
596
    int thirdpel_flag;
597
    int unknown_svq3_flag;
598
    int next_slice_index;
599
    uint32_t svq3_watermark_key;
600
}H264Context;
601

    
602

    
603
extern const uint8_t ff_h264_chroma_qp[52];
604

    
605
/**
606
 * Decode SEI
607
 */
608
int ff_h264_decode_sei(H264Context *h);
609

    
610
/**
611
 * Decode SPS
612
 */
613
int ff_h264_decode_seq_parameter_set(H264Context *h);
614

    
615
/**
616
 * compute profile from sps
617
 */
618
int ff_h264_get_profile(SPS *sps);
619

    
620
/**
621
 * Decode PPS
622
 */
623
int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
624

    
625
/**
626
 * Decode a network abstraction layer unit.
627
 * @param consumed is the number of bytes used as input
628
 * @param length is the length of the array
629
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
630
 * @return decoded bytes, might be src+1 if no escapes
631
 */
632
const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
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

    
726
/* Scan8 organization:
727
 *   0 1 2 3 4 5 6 7
728
 * 0   u u y y y y y
729
 * 1 u U U y Y Y Y Y
730
 * 2 u U U y Y Y Y Y
731
 * 3   v v y Y Y Y Y
732
 * 4 v V V y Y Y Y Y
733
 * 5 v V V   DYDUDV
734
 * DY/DU/DV are for luma/chroma DC.
735
 */
736

    
737
//This table must be here because scan8[constant] must be known at compiletime
738
static const uint8_t scan8[16 + 2*4 + 3]={
739
 4+1*8, 5+1*8, 4+2*8, 5+2*8,
740
 6+1*8, 7+1*8, 6+2*8, 7+2*8,
741
 4+3*8, 5+3*8, 4+4*8, 5+4*8,
742
 6+3*8, 7+3*8, 6+4*8, 7+4*8,
743
 1+1*8, 2+1*8,
744
 1+2*8, 2+2*8,
745
 1+4*8, 2+4*8,
746
 1+5*8, 2+5*8,
747
 4+5*8, 5+5*8, 6+5*8
748
};
749

    
750
static av_always_inline uint32_t pack16to32(int a, int b){
751
#if HAVE_BIGENDIAN
752
   return (b&0xFFFF) + (a<<16);
753
#else
754
   return (a&0xFFFF) + (b<<16);
755
#endif
756
}
757

    
758
static av_always_inline uint16_t pack8to16(int a, int b){
759
#if HAVE_BIGENDIAN
760
   return (b&0xFF) + (a<<8);
761
#else
762
   return (a&0xFF) + (b<<8);
763
#endif
764
}
765

    
766
/**
767
 * gets the chroma qp.
768
 */
769
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
770
    return h->pps.chroma_qp_table[t][qscale];
771
}
772

    
773
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
774

    
775
static void fill_decode_neighbors(H264Context *h, int mb_type){
776
    MpegEncContext * const s = &h->s;
777
    const int mb_xy= h->mb_xy;
778
    int topleft_xy, top_xy, topright_xy, left_xy[2];
779
    static const uint8_t left_block_options[4][16]={
780
        {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},
781
        {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},
782
        {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},
783
        {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}
784
    };
785

    
786
    h->topleft_partition= -1;
787

    
788
    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
789

    
790
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
791
     * stuff, I can't imagine that these complex rules are worth it. */
792

    
793
    topleft_xy = top_xy - 1;
794
    topright_xy= top_xy + 1;
795
    left_xy[1] = left_xy[0] = mb_xy-1;
796
    h->left_block = left_block_options[0];
797
    if(FRAME_MBAFF){
798
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
799
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
800
        if(s->mb_y&1){
801
            if (left_mb_field_flag != curr_mb_field_flag) {
802
                left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
803
                if (curr_mb_field_flag) {
804
                    left_xy[1] += s->mb_stride;
805
                    h->left_block = left_block_options[3];
806
                } else {
807
                    topleft_xy += s->mb_stride;
808
                    // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
809
                    h->topleft_partition = 0;
810
                    h->left_block = left_block_options[1];
811
                }
812
            }
813
        }else{
814
            if(curr_mb_field_flag){
815
                topleft_xy  += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
816
                topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
817
                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
818
            }
819
            if (left_mb_field_flag != curr_mb_field_flag) {
820
                if (curr_mb_field_flag) {
821
                    left_xy[1] += s->mb_stride;
822
                    h->left_block = left_block_options[3];
823
                } else {
824
                    h->left_block = left_block_options[2];
825
                }
826
            }
827
        }
828
    }
829

    
830
    h->topleft_mb_xy = topleft_xy;
831
    h->top_mb_xy     = top_xy;
832
    h->topright_mb_xy= topright_xy;
833
    h->left_mb_xy[0] = left_xy[0];
834
    h->left_mb_xy[1] = left_xy[1];
835
    //FIXME do we need all in the context?
836

    
837
    h->topleft_type = s->current_picture.mb_type[topleft_xy] ;
838
    h->top_type     = s->current_picture.mb_type[top_xy]     ;
839
    h->topright_type= s->current_picture.mb_type[topright_xy];
840
    h->left_type[0] = s->current_picture.mb_type[left_xy[0]] ;
841
    h->left_type[1] = s->current_picture.mb_type[left_xy[1]] ;
842

    
843
    if(FMO){
844
    if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0;
845
    if(h->slice_table[top_xy     ] != h->slice_num) h->top_type     = 0;
846
    if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
847
    }else{
848
        if(h->slice_table[topleft_xy ] != h->slice_num){
849
            h->topleft_type = 0;
850
            if(h->slice_table[top_xy     ] != h->slice_num) h->top_type     = 0;
851
            if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
852
        }
853
    }
854
    if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
855
}
856

    
857
static void fill_decode_caches(H264Context *h, int mb_type){
858
    MpegEncContext * const s = &h->s;
859
    int topleft_xy, top_xy, topright_xy, left_xy[2];
860
    int topleft_type, top_type, topright_type, left_type[2];
861
    const uint8_t * left_block= h->left_block;
862
    int i;
863

    
864
    topleft_xy   = h->topleft_mb_xy ;
865
    top_xy       = h->top_mb_xy     ;
866
    topright_xy  = h->topright_mb_xy;
867
    left_xy[0]   = h->left_mb_xy[0] ;
868
    left_xy[1]   = h->left_mb_xy[1] ;
869
    topleft_type = h->topleft_type  ;
870
    top_type     = h->top_type      ;
871
    topright_type= h->topright_type ;
872
    left_type[0] = h->left_type[0]  ;
873
    left_type[1] = h->left_type[1]  ;
874

    
875
    if(!IS_SKIP(mb_type)){
876
        if(IS_INTRA(mb_type)){
877
            int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
878
            h->topleft_samples_available=
879
            h->top_samples_available=
880
            h->left_samples_available= 0xFFFF;
881
            h->topright_samples_available= 0xEEEA;
882

    
883
            if(!(top_type & type_mask)){
884
                h->topleft_samples_available= 0xB3FF;
885
                h->top_samples_available= 0x33FF;
886
                h->topright_samples_available= 0x26EA;
887
            }
888
            if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
889
                if(IS_INTERLACED(mb_type)){
890
                    if(!(left_type[0] & type_mask)){
891
                        h->topleft_samples_available&= 0xDFFF;
892
                        h->left_samples_available&= 0x5FFF;
893
                    }
894
                    if(!(left_type[1] & type_mask)){
895
                        h->topleft_samples_available&= 0xFF5F;
896
                        h->left_samples_available&= 0xFF5F;
897
                    }
898
                }else{
899
                    int left_typei = s->current_picture.mb_type[left_xy[0] + s->mb_stride];
900

    
901
                    assert(left_xy[0] == left_xy[1]);
902
                    if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
903
                        h->topleft_samples_available&= 0xDF5F;
904
                        h->left_samples_available&= 0x5F5F;
905
                    }
906
                }
907
            }else{
908
                if(!(left_type[0] & type_mask)){
909
                    h->topleft_samples_available&= 0xDF5F;
910
                    h->left_samples_available&= 0x5F5F;
911
                }
912
            }
913

    
914
            if(!(topleft_type & type_mask))
915
                h->topleft_samples_available&= 0x7FFF;
916

    
917
            if(!(topright_type & type_mask))
918
                h->topright_samples_available&= 0xFBFF;
919

    
920
            if(IS_INTRA4x4(mb_type)){
921
                if(IS_INTRA4x4(top_type)){
922
                    AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]);
923
                }else{
924
                    h->intra4x4_pred_mode_cache[4+8*0]=
925
                    h->intra4x4_pred_mode_cache[5+8*0]=
926
                    h->intra4x4_pred_mode_cache[6+8*0]=
927
                    h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask);
928
                }
929
                for(i=0; i<2; i++){
930
                    if(IS_INTRA4x4(left_type[i])){
931
                        int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]];
932
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]];
933
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]];
934
                    }else{
935
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
936
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[i] & type_mask);
937
                    }
938
                }
939
            }
940
        }
941

    
942

    
943
/*
944
0 . T T. T T T T
945
1 L . .L . . . .
946
2 L . .L . . . .
947
3 . T TL . . . .
948
4 L . .L . . . .
949
5 L . .. . . . .
950
*/
951
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
952
    if(top_type){
953
        AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
954
            h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
955
            h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
956

    
957
            h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
958
            h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
959
    }else {
960
            h->non_zero_count_cache[1+8*0]=
961
            h->non_zero_count_cache[2+8*0]=
962

    
963
            h->non_zero_count_cache[1+8*3]=
964
            h->non_zero_count_cache[2+8*3]=
965
            AV_WN32A(&h->non_zero_count_cache[4+8*0], CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040);
966
    }
967

    
968
    for (i=0; i<2; i++) {
969
        if(left_type[i]){
970
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
971
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
972
                h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
973
                h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
974
        }else{
975
                h->non_zero_count_cache[3+8*1 + 2*8*i]=
976
                h->non_zero_count_cache[3+8*2 + 2*8*i]=
977
                h->non_zero_count_cache[0+8*1 +   8*i]=
978
                h->non_zero_count_cache[0+8*4 +   8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
979
        }
980
    }
981

    
982
    if( CABAC ) {
983
        // top_cbp
984
        if(top_type) {
985
            h->top_cbp = h->cbp_table[top_xy];
986
        } else {
987
            h->top_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
988
        }
989
        // left_cbp
990
        if (left_type[0]) {
991
            h->left_cbp = (h->cbp_table[left_xy[0]] & 0x1f0)
992
                        |  ((h->cbp_table[left_xy[0]]>>(left_block[0]&(~1)))&2)
993
                        | (((h->cbp_table[left_xy[1]]>>(left_block[2]&(~1)))&2) << 2);
994
        } else {
995
            h->left_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
996
        }
997
    }
998
    }
999

    
1000
#if 1
1001
    if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
1002
        int list;
1003
        for(list=0; list<h->list_count; list++){
1004
            if(!USES_LIST(mb_type, list)){
1005
                /*if(!h->mv_cache_clean[list]){
1006
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
1007
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
1008
                    h->mv_cache_clean[list]= 1;
1009
                }*/
1010
                continue;
1011
            }
1012
            assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
1013

    
1014
            h->mv_cache_clean[list]= 0;
1015

    
1016
            if(USES_LIST(top_type, list)){
1017
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1018
                AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1019
                    h->ref_cache[list][scan8[0] + 0 - 1*8]=
1020
                    h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 2];
1021
                    h->ref_cache[list][scan8[0] + 2 - 1*8]=
1022
                    h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 3];
1023
            }else{
1024
                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1025
                AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
1026
            }
1027

    
1028
            if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){
1029
            for(i=0; i<2; i++){
1030
                int cache_idx = scan8[0] - 1 + i*2*8;
1031
                if(USES_LIST(left_type[i], list)){
1032
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1033
                    const int b8_xy= 4*left_xy[i] + 1;
1034
                    AV_COPY32(h->mv_cache[list][cache_idx  ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]);
1035
                    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]]);
1036
                        h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + (left_block[0+i*2]&~1)];
1037
                        h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + (left_block[1+i*2]&~1)];
1038
                }else{
1039
                    AV_ZERO32(h->mv_cache [list][cache_idx  ]);
1040
                    AV_ZERO32(h->mv_cache [list][cache_idx+8]);
1041
                    h->ref_cache[list][cache_idx  ]=
1042
                    h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1043
                }
1044
            }
1045
            }else{
1046
                if(USES_LIST(left_type[0], list)){
1047
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1048
                    const int b8_xy= 4*left_xy[0] + 1;
1049
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]]);
1050
                    h->ref_cache[list][scan8[0] - 1]= s->current_picture.ref_index[list][b8_xy + (left_block[0]&~1)];
1051
                }else{
1052
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1]);
1053
                    h->ref_cache[list][scan8[0] - 1]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1054
                }
1055
            }
1056

    
1057
            if(USES_LIST(topright_type, list)){
1058
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1059
                AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]);
1060
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][4*topright_xy + 2];
1061
            }else{
1062
                AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]);
1063
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1064
            }
1065
            if(h->ref_cache[list][scan8[0] + 4 - 1*8] < 0){
1066
                if(USES_LIST(topleft_type, list)){
1067
                    const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
1068
                    const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2);
1069
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]);
1070
                    h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1071
                }else{
1072
                    AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]);
1073
                    h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1074
                }
1075
            }
1076

    
1077
            if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
1078
                continue;
1079

    
1080
            if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
1081
            h->ref_cache[list][scan8[4 ]] =
1082
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1083
            AV_ZERO32(h->mv_cache [list][scan8[4 ]]);
1084
            AV_ZERO32(h->mv_cache [list][scan8[12]]);
1085

    
1086
            if( CABAC ) {
1087
                /* XXX beurk, Load mvd */
1088
                if(USES_LIST(top_type, list)){
1089
                    const int b_xy= h->mb2br_xy[top_xy];
1090
                    AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1091
                }else{
1092
                    AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1093
                }
1094
                if(USES_LIST(left_type[0], list)){
1095
                    const int b_xy= h->mb2br_xy[left_xy[0]] + 6;
1096
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy - left_block[0]]);
1097
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy - left_block[1]]);
1098
                }else{
1099
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]);
1100
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]);
1101
                }
1102
                if(USES_LIST(left_type[1], list)){
1103
                    const int b_xy= h->mb2br_xy[left_xy[1]] + 6;
1104
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy - left_block[2]]);
1105
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy - left_block[3]]);
1106
                }else{
1107
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]);
1108
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]);
1109
                }
1110
                AV_ZERO16(h->mvd_cache [list][scan8[4 ]]);
1111
                AV_ZERO16(h->mvd_cache [list][scan8[12]]);
1112
                if(h->slice_type_nos == FF_B_TYPE){
1113
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
1114

    
1115
                    if(IS_DIRECT(top_type)){
1116
                        AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101u*(MB_TYPE_DIRECT2>>1));
1117
                    }else if(IS_8X8(top_type)){
1118
                        int b8_xy = 4*top_xy;
1119
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy + 2];
1120
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 3];
1121
                    }else{
1122
                        AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1));
1123
                    }
1124

    
1125
                    if(IS_DIRECT(left_type[0]))
1126
                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
1127
                    else if(IS_8X8(left_type[0]))
1128
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[4*left_xy[0] + 1 + (left_block[0]&~1)];
1129
                    else
1130
                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
1131

    
1132
                    if(IS_DIRECT(left_type[1]))
1133
                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
1134
                    else if(IS_8X8(left_type[1]))
1135
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[4*left_xy[1] + 1 + (left_block[2]&~1)];
1136
                    else
1137
                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
1138
                }
1139
            }
1140
            }
1141
            if(FRAME_MBAFF){
1142
#define MAP_MVS\
1143
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1144
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1145
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1146
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1147
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1148
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1149
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1150
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1151
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1152
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1153
                if(MB_FIELD){
1154
#define MAP_F2F(idx, mb_type)\
1155
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1156
                        h->ref_cache[list][idx] <<= 1;\
1157
                        h->mv_cache[list][idx][1] /= 2;\
1158
                        h->mvd_cache[list][idx][1] >>=1;\
1159
                    }
1160
                    MAP_MVS
1161
#undef MAP_F2F
1162
                }else{
1163
#define MAP_F2F(idx, mb_type)\
1164
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1165
                        h->ref_cache[list][idx] >>= 1;\
1166
                        h->mv_cache[list][idx][1] <<= 1;\
1167
                        h->mvd_cache[list][idx][1] <<= 1;\
1168
                    }
1169
                    MAP_MVS
1170
#undef MAP_F2F
1171
                }
1172
            }
1173
        }
1174
    }
1175
#endif
1176

    
1177
        h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1178
}
1179

    
1180
/**
1181
 * gets the predicted intra4x4 prediction mode.
1182
 */
1183
static inline int pred_intra_mode(H264Context *h, int n){
1184
    const int index8= scan8[n];
1185
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1186
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1187
    const int min= FFMIN(left, top);
1188

    
1189
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1190

    
1191
    if(min<0) return DC_PRED;
1192
    else      return min;
1193
}
1194

    
1195
static inline void write_back_non_zero_count(H264Context *h){
1196
    const int mb_xy= h->mb_xy;
1197

    
1198
    AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1199
    AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1200
    AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[0+8*5]);
1201
    AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8*3]);
1202
    AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1203
}
1204

    
1205
static inline void write_back_motion(H264Context *h, int mb_type){
1206
    MpegEncContext * const s = &h->s;
1207
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
1208
    const int b8_xy= 4*h->mb_xy;
1209
    int list;
1210

    
1211
    if(!USES_LIST(mb_type, 0))
1212
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1213

    
1214
    for(list=0; list<h->list_count; list++){
1215
        int y, b_stride;
1216
        int16_t (*mv_dst)[2];
1217
        int16_t (*mv_src)[2];
1218

    
1219
        if(!USES_LIST(mb_type, list))
1220
            continue;
1221

    
1222
        b_stride = h->b_stride;
1223
        mv_dst   = &s->current_picture.motion_val[list][b_xy];
1224
        mv_src   = &h->mv_cache[list][scan8[0]];
1225
        for(y=0; y<4; y++){
1226
            AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1227
        }
1228
        if( CABAC ) {
1229
            uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
1230
            uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1231
            if(IS_SKIP(mb_type))
1232
                AV_ZERO128(mvd_dst);
1233
            else{
1234
            AV_COPY64(mvd_dst, mvd_src + 8*3);
1235
                AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
1236
                AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
1237
                AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
1238
            }
1239
        }
1240

    
1241
        {
1242
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1243
            ref_index[0+0*2]= h->ref_cache[list][scan8[0]];
1244
            ref_index[1+0*2]= h->ref_cache[list][scan8[4]];
1245
            ref_index[0+1*2]= h->ref_cache[list][scan8[8]];
1246
            ref_index[1+1*2]= h->ref_cache[list][scan8[12]];
1247
        }
1248
    }
1249

    
1250
    if(h->slice_type_nos == FF_B_TYPE && CABAC){
1251
        if(IS_8X8(mb_type)){
1252
            uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
1253
            direct_table[1] = h->sub_mb_type[1]>>1;
1254
            direct_table[2] = h->sub_mb_type[2]>>1;
1255
            direct_table[3] = h->sub_mb_type[3]>>1;
1256
        }
1257
    }
1258
}
1259

    
1260
static inline int get_dct8x8_allowed(H264Context *h){
1261
    if(h->sps.direct_8x8_inference_flag)
1262
        return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
1263
    else
1264
        return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1265
}
1266

    
1267
/**
1268
 * decodes a P_SKIP or B_SKIP macroblock
1269
 */
1270
static void av_unused decode_mb_skip(H264Context *h){
1271
    MpegEncContext * const s = &h->s;
1272
    const int mb_xy= h->mb_xy;
1273
    int mb_type=0;
1274

    
1275
    memset(h->non_zero_count[mb_xy], 0, 32);
1276
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1277

    
1278
    if(MB_FIELD)
1279
        mb_type|= MB_TYPE_INTERLACED;
1280

    
1281
    if( h->slice_type_nos == FF_B_TYPE )
1282
    {
1283
        // just for fill_caches. pred_direct_motion will set the real mb_type
1284
        mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1285
        if(h->direct_spatial_mv_pred){
1286
            fill_decode_neighbors(h, mb_type);
1287
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1288
        }
1289
        ff_h264_pred_direct_motion(h, &mb_type);
1290
        mb_type|= MB_TYPE_SKIP;
1291
    }
1292
    else
1293
    {
1294
        int mx, my;
1295
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1296

    
1297
        fill_decode_neighbors(h, mb_type);
1298
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1299
        pred_pskip_motion(h, &mx, &my);
1300
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1301
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1302
    }
1303

    
1304
    write_back_motion(h, mb_type);
1305
    s->current_picture.mb_type[mb_xy]= mb_type;
1306
    s->current_picture.qscale_table[mb_xy]= s->qscale;
1307
    h->slice_table[ mb_xy ]= h->slice_num;
1308
    h->prev_mb_skipped= 1;
1309
}
1310

    
1311
#include "h264_mvpred.h" //For pred_pskip_motion()
1312

    
1313
#endif /* AVCODEC_H264_H */