Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.h @ 77c6edb8

History | View | Annotate | Download (55.1 KB)

1
/*
2
 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3
 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4
 *
5
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21

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

    
28
#ifndef AVCODEC_H264_H
29
#define AVCODEC_H264_H
30

    
31
#include "libavutil/intreadwrite.h"
32
#include "dsputil.h"
33
#include "cabac.h"
34
#include "mpegvideo.h"
35
#include "h264pred.h"
36
#include "rectangle.h"
37

    
38
#define interlaced_dct interlaced_dct_is_a_bad_name
39
#define mb_intra mb_intra_is_not_initialized_see_mb_type
40

    
41
#define LUMA_DC_BLOCK_INDEX   25
42
#define CHROMA_DC_BLOCK_INDEX 26
43

    
44
#define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
45
#define COEFF_TOKEN_VLC_BITS           8
46
#define TOTAL_ZEROS_VLC_BITS           9
47
#define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
48
#define RUN_VLC_BITS                   3
49
#define RUN7_VLC_BITS                  6
50

    
51
#define MAX_SPS_COUNT 32
52
#define MAX_PPS_COUNT 256
53

    
54
#define MAX_MMCO_COUNT 66
55

    
56
#define MAX_DELAYED_PIC_COUNT 16
57

    
58
/* Compiling in interlaced support reduces the speed
59
 * of progressive decoding by about 2%. */
60
#define ALLOW_INTERLACE
61

    
62
#define ALLOW_NOCHROMA
63

    
64
#define FMO 0
65

    
66
/**
67
 * The maximum number of slices supported by the decoder.
68
 * must be a power of 2
69
 */
70
#define MAX_SLICES 16
71

    
72
#ifdef ALLOW_INTERLACE
73
#define MB_MBAFF h->mb_mbaff
74
#define MB_FIELD h->mb_field_decoding_flag
75
#define FRAME_MBAFF h->mb_aff_frame
76
#define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
77
#else
78
#define MB_MBAFF 0
79
#define MB_FIELD 0
80
#define FRAME_MBAFF 0
81
#define FIELD_PICTURE 0
82
#undef  IS_INTERLACED
83
#define IS_INTERLACED(mb_type) 0
84
#endif
85
#define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
86

    
87
#ifdef ALLOW_NOCHROMA
88
#define CHROMA h->sps.chroma_format_idc
89
#else
90
#define CHROMA 1
91
#endif
92

    
93
#ifndef CABAC
94
#define CABAC h->pps.cabac
95
#endif
96

    
97
#define EXTENDED_SAR          255
98

    
99
#define MB_TYPE_REF0       MB_TYPE_ACPRED //dirty but it fits in 16 bit
100
#define MB_TYPE_8x8DCT     0x01000000
101
#define IS_REF0(a)         ((a) & MB_TYPE_REF0)
102
#define IS_8x8DCT(a)       ((a) & MB_TYPE_8x8DCT)
103

    
104
/**
105
 * Value of Picture.reference when Picture is not a reference picture, but
106
 * is held for delayed output.
107
 */
108
#define DELAYED_PIC_REF 4
109

    
110

    
111
/* NAL unit types */
112
enum {
113
    NAL_SLICE=1,
114
    NAL_DPA,
115
    NAL_DPB,
116
    NAL_DPC,
117
    NAL_IDR_SLICE,
118
    NAL_SEI,
119
    NAL_SPS,
120
    NAL_PPS,
121
    NAL_AUD,
122
    NAL_END_SEQUENCE,
123
    NAL_END_STREAM,
124
    NAL_FILLER_DATA,
125
    NAL_SPS_EXT,
126
    NAL_AUXILIARY_SLICE=19
127
};
128

    
129
/**
130
 * SEI message types
131
 */
132
typedef enum {
133
    SEI_BUFFERING_PERIOD             =  0, ///< buffering period (H.264, D.1.1)
134
    SEI_TYPE_PIC_TIMING              =  1, ///< picture timing
135
    SEI_TYPE_USER_DATA_UNREGISTERED  =  5, ///< unregistered user data
136
    SEI_TYPE_RECOVERY_POINT          =  6  ///< recovery point (frame # to decoder sync)
137
} SEI_Type;
138

    
139
/**
140
 * pic_struct in picture timing SEI message
141
 */
142
typedef enum {
143
    SEI_PIC_STRUCT_FRAME             = 0, ///<  0: %frame
144
    SEI_PIC_STRUCT_TOP_FIELD         = 1, ///<  1: top field
145
    SEI_PIC_STRUCT_BOTTOM_FIELD      = 2, ///<  2: bottom field
146
    SEI_PIC_STRUCT_TOP_BOTTOM        = 3, ///<  3: top field, bottom field, in that order
147
    SEI_PIC_STRUCT_BOTTOM_TOP        = 4, ///<  4: bottom field, top field, in that order
148
    SEI_PIC_STRUCT_TOP_BOTTOM_TOP    = 5, ///<  5: top field, bottom field, top field repeated, in that order
149
    SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///<  6: bottom field, top field, bottom field repeated, in that order
150
    SEI_PIC_STRUCT_FRAME_DOUBLING    = 7, ///<  7: %frame doubling
151
    SEI_PIC_STRUCT_FRAME_TRIPLING    = 8  ///<  8: %frame tripling
152
} SEI_PicStructType;
153

    
154
/**
155
 * Sequence parameter set
156
 */
157
typedef struct SPS{
158

    
159
    int profile_idc;
160
    int level_idc;
161
    int chroma_format_idc;
162
    int transform_bypass;              ///< qpprime_y_zero_transform_bypass_flag
163
    int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
164
    int poc_type;                      ///< pic_order_cnt_type
165
    int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
166
    int delta_pic_order_always_zero_flag;
167
    int offset_for_non_ref_pic;
168
    int offset_for_top_to_bottom_field;
169
    int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
170
    int ref_frame_count;               ///< num_ref_frames
171
    int gaps_in_frame_num_allowed_flag;
172
    int mb_width;                      ///< pic_width_in_mbs_minus1 + 1
173
    int mb_height;                     ///< pic_height_in_map_units_minus1 + 1
174
    int frame_mbs_only_flag;
175
    int mb_aff;                        ///<mb_adaptive_frame_field_flag
176
    int direct_8x8_inference_flag;
177
    int crop;                   ///< frame_cropping_flag
178
    unsigned int crop_left;            ///< frame_cropping_rect_left_offset
179
    unsigned int crop_right;           ///< frame_cropping_rect_right_offset
180
    unsigned int crop_top;             ///< frame_cropping_rect_top_offset
181
    unsigned int crop_bottom;          ///< frame_cropping_rect_bottom_offset
182
    int vui_parameters_present_flag;
183
    AVRational sar;
184
    int video_signal_type_present_flag;
185
    int full_range;
186
    int colour_description_present_flag;
187
    enum AVColorPrimaries color_primaries;
188
    enum AVColorTransferCharacteristic color_trc;
189
    enum AVColorSpace colorspace;
190
    int timing_info_present_flag;
191
    uint32_t num_units_in_tick;
192
    uint32_t time_scale;
193
    int fixed_frame_rate_flag;
194
    short offset_for_ref_frame[256]; //FIXME dyn aloc?
195
    int bitstream_restriction_flag;
196
    int num_reorder_frames;
197
    int scaling_matrix_present;
198
    uint8_t scaling_matrix4[6][16];
199
    uint8_t scaling_matrix8[2][64];
200
    int nal_hrd_parameters_present_flag;
201
    int vcl_hrd_parameters_present_flag;
202
    int pic_struct_present_flag;
203
    int time_offset_length;
204
    int cpb_cnt;                       ///< See H.264 E.1.2
205
    int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1
206
    int cpb_removal_delay_length;      ///< cpb_removal_delay_length_minus1 + 1
207
    int dpb_output_delay_length;       ///< dpb_output_delay_length_minus1 + 1
208
    int bit_depth_luma;                ///< bit_depth_luma_minus8 + 8
209
    int bit_depth_chroma;              ///< bit_depth_chroma_minus8 + 8
210
    int residual_color_transform_flag; ///< residual_colour_transform_flag
211
}SPS;
212

    
213
/**
214
 * Picture parameter set
215
 */
216
typedef struct PPS{
217
    unsigned int sps_id;
218
    int cabac;                  ///< entropy_coding_mode_flag
219
    int pic_order_present;      ///< pic_order_present_flag
220
    int slice_group_count;      ///< num_slice_groups_minus1 + 1
221
    int mb_slice_group_map_type;
222
    unsigned int ref_count[2];  ///< num_ref_idx_l0/1_active_minus1 + 1
223
    int weighted_pred;          ///< weighted_pred_flag
224
    int weighted_bipred_idc;
225
    int init_qp;                ///< pic_init_qp_minus26 + 26
226
    int init_qs;                ///< pic_init_qs_minus26 + 26
227
    int chroma_qp_index_offset[2];
228
    int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
229
    int constrained_intra_pred; ///< constrained_intra_pred_flag
230
    int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
231
    int transform_8x8_mode;     ///< transform_8x8_mode_flag
232
    uint8_t scaling_matrix4[6][16];
233
    uint8_t scaling_matrix8[2][64];
234
    uint8_t chroma_qp_table[2][64];  ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
235
    int chroma_qp_diff;
236
}PPS;
237

    
238
/**
239
 * Memory management control operation opcode.
240
 */
241
typedef enum MMCOOpcode{
242
    MMCO_END=0,
243
    MMCO_SHORT2UNUSED,
244
    MMCO_LONG2UNUSED,
245
    MMCO_SHORT2LONG,
246
    MMCO_SET_MAX_LONG,
247
    MMCO_RESET,
248
    MMCO_LONG,
249
} MMCOOpcode;
250

    
251
/**
252
 * Memory management control operation.
253
 */
254
typedef struct MMCO{
255
    MMCOOpcode opcode;
256
    int short_pic_num;  ///< pic_num without wrapping (pic_num & max_pic_num)
257
    int long_arg;       ///< index, pic_num, or num long refs depending on opcode
258
} MMCO;
259

    
260
/**
261
 * H264Context
262
 */
263
typedef struct H264Context{
264
    MpegEncContext s;
265
    int nal_ref_idc;
266
    int nal_unit_type;
267
    uint8_t *rbsp_buffer[2];
268
    unsigned int rbsp_buffer_size[2];
269

    
270
    /**
271
      * Used to parse AVC variant of h264
272
      */
273
    int is_avc; ///< this flag is != 0 if codec is avc1
274
    int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
275

    
276
    int chroma_qp[2]; //QPc
277

    
278
    int qp_thresh;      ///< QP threshold to skip loopfilter
279

    
280
    int prev_mb_skipped;
281
    int next_mb_skipped;
282

    
283
    //prediction stuff
284
    int chroma_pred_mode;
285
    int intra16x16_pred_mode;
286

    
287
    int topleft_mb_xy;
288
    int top_mb_xy;
289
    int topright_mb_xy;
290
    int left_mb_xy[2];
291

    
292
    int topleft_type;
293
    int top_type;
294
    int topright_type;
295
    int left_type[2];
296

    
297
    const uint8_t * left_block;
298
    int topleft_partition;
299

    
300
    int8_t intra4x4_pred_mode_cache[5*8];
301
    int8_t (*intra4x4_pred_mode);
302
    H264PredContext hpc;
303
    unsigned int topleft_samples_available;
304
    unsigned int top_samples_available;
305
    unsigned int topright_samples_available;
306
    unsigned int left_samples_available;
307
    uint8_t (*top_borders[2])[16+2*8];
308
    uint8_t left_border[2*(17+2*9)];
309

    
310
    /**
311
     * non zero coeff count cache.
312
     * is 64 if not available.
313
     */
314
    DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache)[6*8];
315

    
316
    /*
317
    .UU.YYYY
318
    .UU.YYYY
319
    .vv.YYYY
320
    .VV.YYYY
321
    */
322
    uint8_t (*non_zero_count)[32];
323

    
324
    /**
325
     * Motion vector cache.
326
     */
327
    DECLARE_ALIGNED_16(int16_t, mv_cache)[2][5*8][2];
328
    DECLARE_ALIGNED_8(int8_t, ref_cache)[2][5*8];
329
#define LIST_NOT_USED -1 //FIXME rename?
330
#define PART_NOT_AVAILABLE -2
331

    
332
    /**
333
     * is 1 if the specific list MV&references are set to 0,0,-2.
334
     */
335
    int mv_cache_clean[2];
336

    
337
    /**
338
     * number of neighbors (top and/or left) that used 8x8 dct
339
     */
340
    int neighbor_transform_size;
341

    
342
    /**
343
     * block_offset[ 0..23] for frame macroblocks
344
     * block_offset[24..47] for field macroblocks
345
     */
346
    int block_offset[2*(16+8)];
347

    
348
    uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
349
    uint32_t *mb2br_xy;
350
    int b_stride; //FIXME use s->b4_stride
351

    
352
    int mb_linesize;   ///< may be equal to s->linesize or s->linesize*2, for mbaff
353
    int mb_uvlinesize;
354

    
355
    int emu_edge_width;
356
    int emu_edge_height;
357

    
358
    int halfpel_flag;
359
    int thirdpel_flag;
360

    
361
    int unknown_svq3_flag;
362
    int next_slice_index;
363

    
364
    SPS *sps_buffers[MAX_SPS_COUNT];
365
    SPS sps; ///< current sps
366

    
367
    PPS *pps_buffers[MAX_PPS_COUNT];
368
    /**
369
     * current pps
370
     */
371
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
372

    
373
    uint32_t dequant4_buffer[6][52][16];
374
    uint32_t dequant8_buffer[2][52][64];
375
    uint32_t (*dequant4_coeff[6])[16];
376
    uint32_t (*dequant8_coeff[2])[64];
377
    int dequant_coeff_pps;     ///< reinit tables when pps changes
378

    
379
    int slice_num;
380
    uint16_t *slice_table_base;
381
    uint16_t *slice_table;     ///< slice_table_base + 2*mb_stride + 1
382
    int slice_type;
383
    int slice_type_nos;        ///< S free slice type (SI/SP are remapped to I/P)
384
    int slice_type_fixed;
385

    
386
    //interlacing specific flags
387
    int mb_aff_frame;
388
    int mb_field_decoding_flag;
389
    int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag
390

    
391
    DECLARE_ALIGNED_8(uint16_t, sub_mb_type)[4];
392

    
393
    //POC stuff
394
    int poc_lsb;
395
    int poc_msb;
396
    int delta_poc_bottom;
397
    int delta_poc[2];
398
    int frame_num;
399
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
400
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
401
    int frame_num_offset;         ///< for POC type 2
402
    int prev_frame_num_offset;    ///< for POC type 2
403
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
404

    
405
    /**
406
     * frame_num for frames or 2*frame_num+1 for field pics.
407
     */
408
    int curr_pic_num;
409

    
410
    /**
411
     * max_frame_num or 2*max_frame_num for field pics.
412
     */
413
    int max_pic_num;
414

    
415
    //Weighted pred stuff
416
    int use_weight;
417
    int use_weight_chroma;
418
    int luma_log2_weight_denom;
419
    int chroma_log2_weight_denom;
420
    int luma_weight[2][48];
421
    int luma_offset[2][48];
422
    int chroma_weight[2][48][2];
423
    int chroma_offset[2][48][2];
424
    int implicit_weight[48][48];
425

    
426
    //deblock
427
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
428
    int slice_alpha_c0_offset;
429
    int slice_beta_offset;
430

    
431
    int redundant_pic_count;
432

    
433
    int direct_spatial_mv_pred;
434
    int col_parity;
435
    int col_fieldoff;
436
    int dist_scale_factor[16];
437
    int dist_scale_factor_field[2][32];
438
    int map_col_to_list0[2][16+32];
439
    int map_col_to_list0_field[2][2][16+32];
440

    
441
    /**
442
     * num_ref_idx_l0/1_active_minus1 + 1
443
     */
444
    unsigned int ref_count[2];   ///< counts frames or fields, depending on current mb mode
445
    unsigned int list_count;
446
    uint8_t *list_counts;            ///< Array of list_count per MB specifying the slice type
447
    Picture *short_ref[32];
448
    Picture *long_ref[32];
449
    Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
450
    Picture ref_list[2][48];         /**< 0..15: frame refs, 16..47: mbaff field refs.
451
                                          Reordered version of default_ref_list
452
                                          according to picture reordering in slice header */
453
    int ref2frm[MAX_SLICES][2][64];  ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
454
    Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
455
    int outputed_poc;
456

    
457
    /**
458
     * memory management control operations buffer.
459
     */
460
    MMCO mmco[MAX_MMCO_COUNT];
461
    int mmco_index;
462

    
463
    int long_ref_count;  ///< number of actual long term references
464
    int short_ref_count; ///< number of actual short term references
465

    
466
    //data partitioning
467
    GetBitContext intra_gb;
468
    GetBitContext inter_gb;
469
    GetBitContext *intra_gb_ptr;
470
    GetBitContext *inter_gb_ptr;
471

    
472
    DECLARE_ALIGNED_16(DCTELEM, mb)[16*24];
473
    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
474

    
475
    /**
476
     * Cabac
477
     */
478
    CABACContext cabac;
479
    uint8_t      cabac_state[460];
480
    int          cabac_init_idc;
481

    
482
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
483
    uint16_t     *cbp_table;
484
    int cbp;
485
    int top_cbp;
486
    int left_cbp;
487
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
488
    uint8_t     *chroma_pred_mode_table;
489
    int         last_qscale_diff;
490
    uint8_t     (*mvd_table[2])[2];
491
    DECLARE_ALIGNED_16(uint8_t, mvd_cache)[2][5*8][2];
492
    uint8_t     *direct_table;
493
    uint8_t     direct_cache[5*8];
494

    
495
    uint8_t zigzag_scan[16];
496
    uint8_t zigzag_scan8x8[64];
497
    uint8_t zigzag_scan8x8_cavlc[64];
498
    uint8_t field_scan[16];
499
    uint8_t field_scan8x8[64];
500
    uint8_t field_scan8x8_cavlc[64];
501
    const uint8_t *zigzag_scan_q0;
502
    const uint8_t *zigzag_scan8x8_q0;
503
    const uint8_t *zigzag_scan8x8_cavlc_q0;
504
    const uint8_t *field_scan_q0;
505
    const uint8_t *field_scan8x8_q0;
506
    const uint8_t *field_scan8x8_cavlc_q0;
507

    
508
    int x264_build;
509

    
510
    /**
511
     * @defgroup multithreading Members for slice based multithreading
512
     * @{
513
     */
514
    struct H264Context *thread_context[MAX_THREADS];
515

    
516
    /**
517
     * current slice number, used to initalize slice_num of each thread/context
518
     */
519
    int current_slice;
520

    
521
    /**
522
     * Max number of threads / contexts.
523
     * This is equal to AVCodecContext.thread_count unless
524
     * multithreaded decoding is impossible, in which case it is
525
     * reduced to 1.
526
     */
527
    int max_contexts;
528

    
529
    /**
530
     *  1 if the single thread fallback warning has already been
531
     *  displayed, 0 otherwise.
532
     */
533
    int single_decode_warning;
534

    
535
    int last_slice_type;
536
    /** @} */
537

    
538
    int mb_xy;
539

    
540
    uint32_t svq3_watermark_key;
541

    
542
    /**
543
     * pic_struct in picture timing SEI message
544
     */
545
    SEI_PicStructType sei_pic_struct;
546

    
547
    /**
548
     * Complement sei_pic_struct
549
     * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
550
     * However, soft telecined frames may have these values.
551
     * This is used in an attempt to flag soft telecine progressive.
552
     */
553
    int prev_interlaced_frame;
554

    
555
    /**
556
     * Bit set of clock types for fields/frames in picture timing SEI message.
557
     * For each found ct_type, appropriate bit is set (e.g., bit 1 for
558
     * interlaced).
559
     */
560
    int sei_ct_type;
561

    
562
    /**
563
     * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
564
     */
565
    int sei_dpb_output_delay;
566

    
567
    /**
568
     * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
569
     */
570
    int sei_cpb_removal_delay;
571

    
572
    /**
573
     * recovery_frame_cnt from SEI message
574
     *
575
     * Set to -1 if no recovery point SEI message found or to number of frames
576
     * before playback synchronizes. Frames having recovery point are key
577
     * frames.
578
     */
579
    int sei_recovery_frame_cnt;
580

    
581
    int is_complex;
582

    
583
    int luma_weight_flag[2];   ///< 7.4.3.2 luma_weight_lX_flag
584
    int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
585

    
586
    // Timestamp stuff
587
    int sei_buffering_period_present;  ///< Buffering period SEI flag
588
    int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
589
}H264Context;
590

    
591

    
592
extern const uint8_t ff_h264_chroma_qp[52];
593

    
594
void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
595

    
596
void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
597

    
598
/**
599
 * Decode SEI
600
 */
601
int ff_h264_decode_sei(H264Context *h);
602

    
603
/**
604
 * Decode SPS
605
 */
606
int ff_h264_decode_seq_parameter_set(H264Context *h);
607

    
608
/**
609
 * Decode PPS
610
 */
611
int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
612

    
613
/**
614
 * Decodes a network abstraction layer unit.
615
 * @param consumed is the number of bytes used as input
616
 * @param length is the length of the array
617
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
618
 * @returns decoded bytes, might be src+1 if no escapes
619
 */
620
const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
621

    
622
/**
623
 * identifies the exact end of the bitstream
624
 * @return the length of the trailing, or 0 if damaged
625
 */
626
int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
627

    
628
/**
629
 * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
630
 */
631
av_cold void ff_h264_free_context(H264Context *h);
632

    
633
/**
634
 * reconstructs bitstream slice_type.
635
 */
636
int ff_h264_get_slice_type(const H264Context *h);
637

    
638
/**
639
 * allocates tables.
640
 * needs width/height
641
 */
642
int ff_h264_alloc_tables(H264Context *h);
643

    
644
/**
645
 * fills the default_ref_list.
646
 */
647
int ff_h264_fill_default_ref_list(H264Context *h);
648

    
649
int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
650
void ff_h264_fill_mbaff_ref_list(H264Context *h);
651
void ff_h264_remove_all_refs(H264Context *h);
652

    
653
/**
654
 * Executes the reference picture marking (memory management control operations).
655
 */
656
int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
657

    
658
int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
659

    
660

    
661
/**
662
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
663
 */
664
int ff_h264_check_intra4x4_pred_mode(H264Context *h);
665

    
666
/**
667
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
668
 */
669
int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
670

    
671
void ff_h264_write_back_intra_pred_mode(H264Context *h);
672
void ff_h264_hl_decode_mb(H264Context *h);
673
int ff_h264_frame_start(H264Context *h);
674
av_cold int ff_h264_decode_init(AVCodecContext *avctx);
675
av_cold int ff_h264_decode_end(AVCodecContext *avctx);
676
av_cold void ff_h264_decode_init_vlc(void);
677

    
678
/**
679
 * decodes a macroblock
680
 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
681
 */
682
int ff_h264_decode_mb_cavlc(H264Context *h);
683

    
684
/**
685
 * decodes a CABAC coded macroblock
686
 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
687
 */
688
int ff_h264_decode_mb_cabac(H264Context *h);
689

    
690
void ff_h264_init_cabac_states(H264Context *h);
691

    
692
void ff_h264_direct_dist_scale_factor(H264Context * const h);
693
void ff_h264_direct_ref_list_init(H264Context * const h);
694
void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
695

    
696
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);
697
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);
698

    
699
/**
700
 * Reset SEI values at the beginning of the frame.
701
 *
702
 * @param h H.264 context.
703
 */
704
void ff_h264_reset_sei(H264Context *h);
705

    
706

    
707
/*
708
o-o o-o
709
 / / /
710
o-o o-o
711
 ,---'
712
o-o o-o
713
 / / /
714
o-o o-o
715
*/
716
//This table must be here because scan8[constant] must be known at compiletime
717
static const uint8_t scan8[16 + 2*4]={
718
 4+1*8, 5+1*8, 4+2*8, 5+2*8,
719
 6+1*8, 7+1*8, 6+2*8, 7+2*8,
720
 4+3*8, 5+3*8, 4+4*8, 5+4*8,
721
 6+3*8, 7+3*8, 6+4*8, 7+4*8,
722
 1+1*8, 2+1*8,
723
 1+2*8, 2+2*8,
724
 1+4*8, 2+4*8,
725
 1+5*8, 2+5*8,
726
};
727

    
728
static av_always_inline uint32_t pack16to32(int a, int b){
729
#if HAVE_BIGENDIAN
730
   return (b&0xFFFF) + (a<<16);
731
#else
732
   return (a&0xFFFF) + (b<<16);
733
#endif
734
}
735

    
736
static av_always_inline uint16_t pack8to16(int a, int b){
737
#if HAVE_BIGENDIAN
738
   return (b&0xFF) + (a<<8);
739
#else
740
   return (a&0xFF) + (b<<8);
741
#endif
742
}
743

    
744
/**
745
 * gets the chroma qp.
746
 */
747
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
748
    return h->pps.chroma_qp_table[t][qscale];
749
}
750

    
751
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
752

    
753
static void fill_decode_neighbors(H264Context *h, int mb_type){
754
    MpegEncContext * const s = &h->s;
755
    const int mb_xy= h->mb_xy;
756
    int topleft_xy, top_xy, topright_xy, left_xy[2];
757
    static const uint8_t left_block_options[4][16]={
758
        {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},
759
        {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},
760
        {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},
761
        {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}
762
    };
763

    
764
    h->topleft_partition= -1;
765

    
766
    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
767

    
768
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
769
     * stuff, I can't imagine that these complex rules are worth it. */
770

    
771
    topleft_xy = top_xy - 1;
772
    topright_xy= top_xy + 1;
773
    left_xy[1] = left_xy[0] = mb_xy-1;
774
    h->left_block = left_block_options[0];
775
    if(FRAME_MBAFF){
776
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
777
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
778
        if(s->mb_y&1){
779
            if (left_mb_field_flag != curr_mb_field_flag) {
780
                left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
781
                if (curr_mb_field_flag) {
782
                    left_xy[1] += s->mb_stride;
783
                    h->left_block = left_block_options[3];
784
                } else {
785
                    topleft_xy += s->mb_stride;
786
                    // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
787
                    h->topleft_partition = 0;
788
                    h->left_block = left_block_options[1];
789
                }
790
            }
791
        }else{
792
            if(curr_mb_field_flag){
793
                topleft_xy  += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
794
                topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
795
                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
796
            }
797
            if (left_mb_field_flag != curr_mb_field_flag) {
798
                if (curr_mb_field_flag) {
799
                    left_xy[1] += s->mb_stride;
800
                    h->left_block = left_block_options[3];
801
                } else {
802
                    h->left_block = left_block_options[2];
803
                }
804
            }
805
        }
806
    }
807

    
808
    h->topleft_mb_xy = topleft_xy;
809
    h->top_mb_xy     = top_xy;
810
    h->topright_mb_xy= topright_xy;
811
    h->left_mb_xy[0] = left_xy[0];
812
    h->left_mb_xy[1] = left_xy[1];
813
    //FIXME do we need all in the context?
814
    h->topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
815
    h->top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
816
    h->topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
817
    h->left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
818
    h->left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
819
}
820

    
821
static void fill_decode_caches(H264Context *h, int mb_type){
822
    MpegEncContext * const s = &h->s;
823
    int topleft_xy, top_xy, topright_xy, left_xy[2];
824
    int topleft_type, top_type, topright_type, left_type[2];
825
    const uint8_t * left_block= h->left_block;
826
    int i;
827

    
828
    topleft_xy   = h->topleft_mb_xy ;
829
    top_xy       = h->top_mb_xy     ;
830
    topright_xy  = h->topright_mb_xy;
831
    left_xy[0]   = h->left_mb_xy[0] ;
832
    left_xy[1]   = h->left_mb_xy[1] ;
833
    topleft_type = h->topleft_type  ;
834
    top_type     = h->top_type      ;
835
    topright_type= h->topright_type ;
836
    left_type[0] = h->left_type[0]  ;
837
    left_type[1] = h->left_type[1]  ;
838

    
839
    if(!IS_SKIP(mb_type)){
840
        if(IS_INTRA(mb_type)){
841
            int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
842
            h->topleft_samples_available=
843
            h->top_samples_available=
844
            h->left_samples_available= 0xFFFF;
845
            h->topright_samples_available= 0xEEEA;
846

    
847
            if(!(top_type & type_mask)){
848
                h->topleft_samples_available= 0xB3FF;
849
                h->top_samples_available= 0x33FF;
850
                h->topright_samples_available= 0x26EA;
851
            }
852
            if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
853
                if(IS_INTERLACED(mb_type)){
854
                    if(!(left_type[0] & type_mask)){
855
                        h->topleft_samples_available&= 0xDFFF;
856
                        h->left_samples_available&= 0x5FFF;
857
                    }
858
                    if(!(left_type[1] & type_mask)){
859
                        h->topleft_samples_available&= 0xFF5F;
860
                        h->left_samples_available&= 0xFF5F;
861
                    }
862
                }else{
863
                    int left_typei = s->current_picture.mb_type[left_xy[0] + s->mb_stride];
864

    
865
                    assert(left_xy[0] == left_xy[1]);
866
                    if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
867
                        h->topleft_samples_available&= 0xDF5F;
868
                        h->left_samples_available&= 0x5F5F;
869
                    }
870
                }
871
            }else{
872
                if(!(left_type[0] & type_mask)){
873
                    h->topleft_samples_available&= 0xDF5F;
874
                    h->left_samples_available&= 0x5F5F;
875
                }
876
            }
877

    
878
            if(!(topleft_type & type_mask))
879
                h->topleft_samples_available&= 0x7FFF;
880

    
881
            if(!(topright_type & type_mask))
882
                h->topright_samples_available&= 0xFBFF;
883

    
884
            if(IS_INTRA4x4(mb_type)){
885
                if(IS_INTRA4x4(top_type)){
886
                    AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]);
887
                }else{
888
                    h->intra4x4_pred_mode_cache[4+8*0]=
889
                    h->intra4x4_pred_mode_cache[5+8*0]=
890
                    h->intra4x4_pred_mode_cache[6+8*0]=
891
                    h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask);
892
                }
893
                for(i=0; i<2; i++){
894
                    if(IS_INTRA4x4(left_type[i])){
895
                        int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]];
896
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]];
897
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]];
898
                    }else{
899
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
900
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[i] & type_mask);
901
                    }
902
                }
903
            }
904
        }
905

    
906

    
907
/*
908
0 . T T. T T T T
909
1 L . .L . . . .
910
2 L . .L . . . .
911
3 . T TL . . . .
912
4 L . .L . . . .
913
5 L . .. . . . .
914
*/
915
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
916
    if(top_type){
917
        AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
918
            h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
919
            h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
920

    
921
            h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
922
            h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
923
    }else {
924
            h->non_zero_count_cache[1+8*0]=
925
            h->non_zero_count_cache[2+8*0]=
926

    
927
            h->non_zero_count_cache[1+8*3]=
928
            h->non_zero_count_cache[2+8*3]=
929
            AV_WN32A(&h->non_zero_count_cache[4+8*0], CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040);
930
    }
931

    
932
    for (i=0; i<2; i++) {
933
        if(left_type[i]){
934
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
935
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
936
                h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
937
                h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
938
        }else{
939
                h->non_zero_count_cache[3+8*1 + 2*8*i]=
940
                h->non_zero_count_cache[3+8*2 + 2*8*i]=
941
                h->non_zero_count_cache[0+8*1 +   8*i]=
942
                h->non_zero_count_cache[0+8*4 +   8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
943
        }
944
    }
945

    
946
    if( CABAC ) {
947
        // top_cbp
948
        if(top_type) {
949
            h->top_cbp = h->cbp_table[top_xy];
950
        } else if(IS_INTRA(mb_type)) {
951
            h->top_cbp = 0x1CF;
952
        } else {
953
            h->top_cbp = 0x00F;
954
        }
955
        // left_cbp
956
        if (left_type[0]) {
957
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
958
        } else if(IS_INTRA(mb_type)) {
959
            h->left_cbp = 0x1CF;
960
        } else {
961
            h->left_cbp = 0x00F;
962
        }
963
        if (left_type[0]) {
964
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
965
        }
966
        if (left_type[1]) {
967
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
968
        }
969
    }
970
    }
971

    
972
#if 1
973
    if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
974
        int list;
975
        for(list=0; list<h->list_count; list++){
976
            if(!USES_LIST(mb_type, list)){
977
                /*if(!h->mv_cache_clean[list]){
978
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
979
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
980
                    h->mv_cache_clean[list]= 1;
981
                }*/
982
                continue;
983
            }
984
            assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
985

    
986
            h->mv_cache_clean[list]= 0;
987

    
988
            if(USES_LIST(top_type, list)){
989
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
990
                AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
991
                    h->ref_cache[list][scan8[0] + 0 - 1*8]=
992
                    h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 2];
993
                    h->ref_cache[list][scan8[0] + 2 - 1*8]=
994
                    h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 3];
995
            }else{
996
                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
997
                AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
998
            }
999

    
1000
            for(i=0; i<2; i++){
1001
                int cache_idx = scan8[0] - 1 + i*2*8;
1002
                if(USES_LIST(left_type[i], list)){
1003
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1004
                    const int b8_xy= 4*left_xy[i] + 1;
1005
                    AV_COPY32(h->mv_cache[list][cache_idx  ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]);
1006
                    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]]);
1007
                        h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + (left_block[0+i*2]&~1)];
1008
                        h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + (left_block[1+i*2]&~1)];
1009
                }else{
1010
                    AV_ZERO32(h->mv_cache [list][cache_idx  ]);
1011
                    AV_ZERO32(h->mv_cache [list][cache_idx+8]);
1012
                    h->ref_cache[list][cache_idx  ]=
1013
                    h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1014
                }
1015
            }
1016

    
1017
            if(USES_LIST(topleft_type, list)){
1018
                const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
1019
                const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2);
1020
                AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]);
1021
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1022
            }else{
1023
                AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]);
1024
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1025
            }
1026

    
1027
            if(USES_LIST(topright_type, list)){
1028
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1029
                AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]);
1030
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][4*topright_xy + 2];
1031
            }else{
1032
                AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]);
1033
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1034
            }
1035

    
1036
            if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
1037
                continue;
1038

    
1039
            if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
1040
            h->ref_cache[list][scan8[5 ]+1] =
1041
            h->ref_cache[list][scan8[7 ]+1] =
1042
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
1043
            h->ref_cache[list][scan8[4 ]] =
1044
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1045
            AV_ZERO32(h->mv_cache [list][scan8[4 ]]);
1046
            AV_ZERO32(h->mv_cache [list][scan8[12]]);
1047

    
1048
            if( CABAC ) {
1049
                /* XXX beurk, Load mvd */
1050
                if(USES_LIST(top_type, list)){
1051
                    const int b_xy= h->mb2br_xy[top_xy];
1052
                    AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1053
                }else{
1054
                    AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1055
                }
1056
                if(USES_LIST(left_type[0], list)){
1057
                    const int b_xy= h->mb2br_xy[left_xy[0]] + 6;
1058
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy - left_block[0]]);
1059
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy - left_block[1]]);
1060
                }else{
1061
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]);
1062
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]);
1063
                }
1064
                if(USES_LIST(left_type[1], list)){
1065
                    const int b_xy= h->mb2br_xy[left_xy[1]] + 6;
1066
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy - left_block[2]]);
1067
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy - left_block[3]]);
1068
                }else{
1069
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]);
1070
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]);
1071
                }
1072
                AV_ZERO16(h->mvd_cache [list][scan8[5 ]+1]);
1073
                AV_ZERO16(h->mvd_cache [list][scan8[7 ]+1]);
1074
                AV_ZERO16(h->mvd_cache [list][scan8[13]+1]); //FIXME remove past 3 (init somewhere else)
1075
                AV_ZERO16(h->mvd_cache [list][scan8[4 ]]);
1076
                AV_ZERO16(h->mvd_cache [list][scan8[12]]);
1077
                if(h->slice_type_nos == FF_B_TYPE){
1078
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
1079

    
1080
                    if(IS_DIRECT(top_type)){
1081
                        AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_DIRECT2>>1));
1082
                    }else if(IS_8X8(top_type)){
1083
                        int b8_xy = 4*top_xy;
1084
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy + 2];
1085
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 3];
1086
                    }else{
1087
                        AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1));
1088
                    }
1089

    
1090
                    if(IS_DIRECT(left_type[0]))
1091
                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
1092
                    else if(IS_8X8(left_type[0]))
1093
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[4*left_xy[0] + 1 + (left_block[0]&~1)];
1094
                    else
1095
                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
1096

    
1097
                    if(IS_DIRECT(left_type[1]))
1098
                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
1099
                    else if(IS_8X8(left_type[1]))
1100
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[4*left_xy[1] + 1 + (left_block[2]&~1)];
1101
                    else
1102
                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
1103
                }
1104
            }
1105
            }
1106
            if(FRAME_MBAFF){
1107
#define MAP_MVS\
1108
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1109
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1110
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1111
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1112
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1113
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1114
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1115
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1116
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1117
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1118
                if(MB_FIELD){
1119
#define MAP_F2F(idx, mb_type)\
1120
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1121
                        h->ref_cache[list][idx] <<= 1;\
1122
                        h->mv_cache[list][idx][1] /= 2;\
1123
                        h->mvd_cache[list][idx][1] >>=1;\
1124
                    }
1125
                    MAP_MVS
1126
#undef MAP_F2F
1127
                }else{
1128
#define MAP_F2F(idx, mb_type)\
1129
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1130
                        h->ref_cache[list][idx] >>= 1;\
1131
                        h->mv_cache[list][idx][1] <<= 1;\
1132
                        h->mvd_cache[list][idx][1] <<= 1;\
1133
                    }
1134
                    MAP_MVS
1135
#undef MAP_F2F
1136
                }
1137
            }
1138
        }
1139
    }
1140
#endif
1141

    
1142
        h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1143
}
1144

    
1145
/**
1146
 *
1147
 * @returns non zero if the loop filter can be skiped
1148
 */
1149
static int fill_filter_caches(H264Context *h, int mb_type){
1150
    MpegEncContext * const s = &h->s;
1151
    const int mb_xy= h->mb_xy;
1152
    int top_xy, left_xy[2];
1153
    int top_type, left_type[2];
1154

    
1155
    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
1156

    
1157
    //FIXME deblocking could skip the intra and nnz parts.
1158

    
1159
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
1160
     * stuff, I can't imagine that these complex rules are worth it. */
1161

    
1162
    left_xy[1] = left_xy[0] = mb_xy-1;
1163
    if(FRAME_MBAFF){
1164
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
1165
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
1166
        if(s->mb_y&1){
1167
            if (left_mb_field_flag != curr_mb_field_flag) {
1168
                left_xy[0] -= s->mb_stride;
1169
            }
1170
        }else{
1171
            if(curr_mb_field_flag){
1172
                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
1173
            }
1174
            if (left_mb_field_flag != curr_mb_field_flag) {
1175
                left_xy[1] += s->mb_stride;
1176
            }
1177
        }
1178
    }
1179

    
1180
    h->top_mb_xy = top_xy;
1181
    h->left_mb_xy[0] = left_xy[0];
1182
    h->left_mb_xy[1] = left_xy[1];
1183
    {
1184
        //for sufficiently low qp, filtering wouldn't do anything
1185
        //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
1186
        int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
1187
        int qp = s->current_picture.qscale_table[mb_xy];
1188
        if(qp <= qp_thresh
1189
           && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
1190
           && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
1191
            if(!FRAME_MBAFF)
1192
                return 1;
1193
            if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
1194
               && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
1195
                return 1;
1196
        }
1197
    }
1198

    
1199
    if(h->deblocking_filter == 2){
1200
        h->top_type    = top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
1201
        h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
1202
        h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
1203
    }else{
1204
        h->top_type    = top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
1205
        h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
1206
        h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
1207
    }
1208
    if(IS_INTRA(mb_type))
1209
        return 0;
1210

    
1211
    AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
1212
    AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
1213
    AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
1214
    AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
1215
    AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
1216

    
1217
    h->cbp= h->cbp_table[mb_xy];
1218

    
1219
    {
1220
        int list;
1221
        for(list=0; list<h->list_count; list++){
1222
            int8_t *ref;
1223
            int y, b_stride;
1224
            int16_t (*mv_dst)[2];
1225
            int16_t (*mv_src)[2];
1226

    
1227
            if(!USES_LIST(mb_type, list)){
1228
                fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
1229
                AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1230
                AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1231
                AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1232
                AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1233
                continue;
1234
            }
1235

    
1236
            ref = &s->current_picture.ref_index[list][4*mb_xy];
1237
            {
1238
                int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1239
                AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1240
                AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1241
                ref += 2;
1242
                AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1243
                AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1244
            }
1245

    
1246
            b_stride = h->b_stride;
1247
            mv_dst   = &h->mv_cache[list][scan8[0]];
1248
            mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
1249
            for(y=0; y<4; y++){
1250
                AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
1251
            }
1252

    
1253
        }
1254
    }
1255

    
1256

    
1257
/*
1258
0 . T T. T T T T
1259
1 L . .L . . . .
1260
2 L . .L . . . .
1261
3 . T TL . . . .
1262
4 L . .L . . . .
1263
5 L . .. . . . .
1264
*/
1265
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
1266
    if(top_type){
1267
        AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
1268
    }
1269

    
1270
    if(left_type[0]){
1271
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
1272
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
1273
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
1274
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
1275
    }
1276

    
1277
    // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
1278
    if(!CABAC && h->pps.transform_8x8_mode){
1279
        if(IS_8x8DCT(top_type)){
1280
            h->non_zero_count_cache[4+8*0]=
1281
            h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
1282
            h->non_zero_count_cache[6+8*0]=
1283
            h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
1284
        }
1285
        if(IS_8x8DCT(left_type[0])){
1286
            h->non_zero_count_cache[3+8*1]=
1287
            h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
1288
        }
1289
        if(IS_8x8DCT(left_type[1])){
1290
            h->non_zero_count_cache[3+8*3]=
1291
            h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
1292
        }
1293

    
1294
        if(IS_8x8DCT(mb_type)){
1295
            h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
1296
            h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
1297

    
1298
            h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
1299
            h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
1300

    
1301
            h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
1302
            h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
1303

    
1304
            h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
1305
            h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
1306
        }
1307
    }
1308

    
1309
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
1310
        int list;
1311
        for(list=0; list<h->list_count; list++){
1312
            if(USES_LIST(top_type, list)){
1313
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1314
                const int b8_xy= 4*top_xy + 2;
1315
                int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1316
                AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1317
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
1318
                h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
1319
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
1320
                h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
1321
            }else{
1322
                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1323
                AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1324
            }
1325

    
1326
            if(!IS_INTERLACED(mb_type^left_type[0])){
1327
                if(USES_LIST(left_type[0], list)){
1328
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1329
                    const int b8_xy= 4*left_xy[0] + 1;
1330
                    int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1331
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
1332
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
1333
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
1334
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
1335
                    h->ref_cache[list][scan8[0] - 1 + 0 ]=
1336
                    h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
1337
                    h->ref_cache[list][scan8[0] - 1 +16 ]=
1338
                    h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
1339
                }else{
1340
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
1341
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
1342
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
1343
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
1344
                    h->ref_cache[list][scan8[0] - 1 + 0  ]=
1345
                    h->ref_cache[list][scan8[0] - 1 + 8  ]=
1346
                    h->ref_cache[list][scan8[0] - 1 + 16 ]=
1347
                    h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
1348
                }
1349
            }
1350
        }
1351
    }
1352

    
1353
    return 0;
1354
}
1355

    
1356
/**
1357
 * gets the predicted intra4x4 prediction mode.
1358
 */
1359
static inline int pred_intra_mode(H264Context *h, int n){
1360
    const int index8= scan8[n];
1361
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1362
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1363
    const int min= FFMIN(left, top);
1364

    
1365
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1366

    
1367
    if(min<0) return DC_PRED;
1368
    else      return min;
1369
}
1370

    
1371
static inline void write_back_non_zero_count(H264Context *h){
1372
    const int mb_xy= h->mb_xy;
1373

    
1374
    AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1375
    AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1376
    AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[0+8*5]);
1377
    AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8*3]);
1378
    AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1379
}
1380

    
1381
static inline void write_back_motion(H264Context *h, int mb_type){
1382
    MpegEncContext * const s = &h->s;
1383
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
1384
    const int b8_xy= 4*h->mb_xy;
1385
    int list;
1386

    
1387
    if(!USES_LIST(mb_type, 0))
1388
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1389

    
1390
    for(list=0; list<h->list_count; list++){
1391
        int y, b_stride;
1392
        int16_t (*mv_dst)[2];
1393
        int16_t (*mv_src)[2];
1394

    
1395
        if(!USES_LIST(mb_type, list))
1396
            continue;
1397

    
1398
        b_stride = h->b_stride;
1399
        mv_dst   = &s->current_picture.motion_val[list][b_xy];
1400
        mv_src   = &h->mv_cache[list][scan8[0]];
1401
        for(y=0; y<4; y++){
1402
            AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1403
        }
1404
        if( CABAC ) {
1405
            uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
1406
            uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1407
            if(IS_SKIP(mb_type))
1408
                AV_ZERO128(mvd_dst);
1409
            else{
1410
            AV_COPY64(mvd_dst, mvd_src + 8*3);
1411
                AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
1412
                AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
1413
                AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
1414
            }
1415
        }
1416

    
1417
        {
1418
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1419
            ref_index[0+0*2]= h->ref_cache[list][scan8[0]];
1420
            ref_index[1+0*2]= h->ref_cache[list][scan8[4]];
1421
            ref_index[0+1*2]= h->ref_cache[list][scan8[8]];
1422
            ref_index[1+1*2]= h->ref_cache[list][scan8[12]];
1423
        }
1424
    }
1425

    
1426
    if(h->slice_type_nos == FF_B_TYPE && CABAC){
1427
        if(IS_8X8(mb_type)){
1428
            uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
1429
            direct_table[1] = h->sub_mb_type[1]>>1;
1430
            direct_table[2] = h->sub_mb_type[2]>>1;
1431
            direct_table[3] = h->sub_mb_type[3]>>1;
1432
        }
1433
    }
1434
}
1435

    
1436
static inline int get_dct8x8_allowed(H264Context *h){
1437
    if(h->sps.direct_8x8_inference_flag)
1438
        return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
1439
    else
1440
        return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1441
}
1442

    
1443
/**
1444
 * decodes a P_SKIP or B_SKIP macroblock
1445
 */
1446
static void decode_mb_skip(H264Context *h){
1447
    MpegEncContext * const s = &h->s;
1448
    const int mb_xy= h->mb_xy;
1449
    int mb_type=0;
1450

    
1451
    memset(h->non_zero_count[mb_xy], 0, 32);
1452
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1453

    
1454
    if(MB_FIELD)
1455
        mb_type|= MB_TYPE_INTERLACED;
1456

    
1457
    if( h->slice_type_nos == FF_B_TYPE )
1458
    {
1459
        // just for fill_caches. pred_direct_motion will set the real mb_type
1460
        mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1461
        if(h->direct_spatial_mv_pred){
1462
            fill_decode_neighbors(h, mb_type);
1463
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1464
        }
1465
        ff_h264_pred_direct_motion(h, &mb_type);
1466
        mb_type|= MB_TYPE_SKIP;
1467
    }
1468
    else
1469
    {
1470
        int mx, my;
1471
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1472

    
1473
        fill_decode_neighbors(h, mb_type);
1474
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1475
        pred_pskip_motion(h, &mx, &my);
1476
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1477
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1478
    }
1479

    
1480
    write_back_motion(h, mb_type);
1481
    s->current_picture.mb_type[mb_xy]= mb_type;
1482
    s->current_picture.qscale_table[mb_xy]= s->qscale;
1483
    h->slice_table[ mb_xy ]= h->slice_num;
1484
    h->prev_mb_skipped= 1;
1485
}
1486

    
1487
#include "h264_mvpred.h" //For pred_pskip_motion()
1488

    
1489
#endif /* AVCODEC_H264_H */