Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.h @ d7f5e520

History | View | Annotate | Download (55.4 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
    SPS *sps_buffers[MAX_SPS_COUNT];
359
    SPS sps; ///< current sps
360

    
361
    PPS *pps_buffers[MAX_PPS_COUNT];
362
    /**
363
     * current pps
364
     */
365
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
366

    
367
    uint32_t dequant4_buffer[6][52][16];
368
    uint32_t dequant8_buffer[2][52][64];
369
    uint32_t (*dequant4_coeff[6])[16];
370
    uint32_t (*dequant8_coeff[2])[64];
371
    int dequant_coeff_pps;     ///< reinit tables when pps changes
372

    
373
    int slice_num;
374
    uint16_t *slice_table_base;
375
    uint16_t *slice_table;     ///< slice_table_base + 2*mb_stride + 1
376
    int slice_type;
377
    int slice_type_nos;        ///< S free slice type (SI/SP are remapped to I/P)
378
    int slice_type_fixed;
379

    
380
    //interlacing specific flags
381
    int mb_aff_frame;
382
    int mb_field_decoding_flag;
383
    int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag
384

    
385
    DECLARE_ALIGNED_8(uint16_t, sub_mb_type)[4];
386

    
387
    //POC stuff
388
    int poc_lsb;
389
    int poc_msb;
390
    int delta_poc_bottom;
391
    int delta_poc[2];
392
    int frame_num;
393
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
394
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
395
    int frame_num_offset;         ///< for POC type 2
396
    int prev_frame_num_offset;    ///< for POC type 2
397
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
398

    
399
    /**
400
     * frame_num for frames or 2*frame_num+1 for field pics.
401
     */
402
    int curr_pic_num;
403

    
404
    /**
405
     * max_frame_num or 2*max_frame_num for field pics.
406
     */
407
    int max_pic_num;
408

    
409
    //Weighted pred stuff
410
    int use_weight;
411
    int use_weight_chroma;
412
    int luma_log2_weight_denom;
413
    int chroma_log2_weight_denom;
414
    int luma_weight[2][48][2];
415
    int chroma_weight[2][48][2][2];
416
    int implicit_weight[48][48];
417

    
418
    //deblock
419
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
420
    int slice_alpha_c0_offset;
421
    int slice_beta_offset;
422

    
423
    int redundant_pic_count;
424

    
425
    int direct_spatial_mv_pred;
426
    int col_parity;
427
    int col_fieldoff;
428
    int dist_scale_factor[16];
429
    int dist_scale_factor_field[2][32];
430
    int map_col_to_list0[2][16+32];
431
    int map_col_to_list0_field[2][2][16+32];
432

    
433
    /**
434
     * num_ref_idx_l0/1_active_minus1 + 1
435
     */
436
    unsigned int ref_count[2];   ///< counts frames or fields, depending on current mb mode
437
    unsigned int list_count;
438
    uint8_t *list_counts;            ///< Array of list_count per MB specifying the slice type
439
    Picture *short_ref[32];
440
    Picture *long_ref[32];
441
    Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
442
    Picture ref_list[2][48];         /**< 0..15: frame refs, 16..47: mbaff field refs.
443
                                          Reordered version of default_ref_list
444
                                          according to picture reordering in slice header */
445
    int ref2frm[MAX_SLICES][2][64];  ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
446
    Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
447
    int outputed_poc;
448

    
449
    /**
450
     * memory management control operations buffer.
451
     */
452
    MMCO mmco[MAX_MMCO_COUNT];
453
    int mmco_index;
454

    
455
    int long_ref_count;  ///< number of actual long term references
456
    int short_ref_count; ///< number of actual short term references
457

    
458
    //data partitioning
459
    GetBitContext intra_gb;
460
    GetBitContext inter_gb;
461
    GetBitContext *intra_gb_ptr;
462
    GetBitContext *inter_gb_ptr;
463

    
464
    DECLARE_ALIGNED_16(DCTELEM, mb)[16*24];
465
    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
466

    
467
    /**
468
     * Cabac
469
     */
470
    CABACContext cabac;
471
    uint8_t      cabac_state[460];
472
    int          cabac_init_idc;
473

    
474
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
475
    uint16_t     *cbp_table;
476
    int cbp;
477
    int top_cbp;
478
    int left_cbp;
479
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
480
    uint8_t     *chroma_pred_mode_table;
481
    int         last_qscale_diff;
482
    uint8_t     (*mvd_table[2])[2];
483
    DECLARE_ALIGNED_16(uint8_t, mvd_cache)[2][5*8][2];
484
    uint8_t     *direct_table;
485
    uint8_t     direct_cache[5*8];
486

    
487
    uint8_t zigzag_scan[16];
488
    uint8_t zigzag_scan8x8[64];
489
    uint8_t zigzag_scan8x8_cavlc[64];
490
    uint8_t field_scan[16];
491
    uint8_t field_scan8x8[64];
492
    uint8_t field_scan8x8_cavlc[64];
493
    const uint8_t *zigzag_scan_q0;
494
    const uint8_t *zigzag_scan8x8_q0;
495
    const uint8_t *zigzag_scan8x8_cavlc_q0;
496
    const uint8_t *field_scan_q0;
497
    const uint8_t *field_scan8x8_q0;
498
    const uint8_t *field_scan8x8_cavlc_q0;
499

    
500
    int x264_build;
501

    
502
    /**
503
     * @defgroup multithreading Members for slice based multithreading
504
     * @{
505
     */
506
    struct H264Context *thread_context[MAX_THREADS];
507

    
508
    /**
509
     * current slice number, used to initalize slice_num of each thread/context
510
     */
511
    int current_slice;
512

    
513
    /**
514
     * Max number of threads / contexts.
515
     * This is equal to AVCodecContext.thread_count unless
516
     * multithreaded decoding is impossible, in which case it is
517
     * reduced to 1.
518
     */
519
    int max_contexts;
520

    
521
    /**
522
     *  1 if the single thread fallback warning has already been
523
     *  displayed, 0 otherwise.
524
     */
525
    int single_decode_warning;
526

    
527
    int last_slice_type;
528
    /** @} */
529

    
530
    int mb_xy;
531

    
532
    /**
533
     * pic_struct in picture timing SEI message
534
     */
535
    SEI_PicStructType sei_pic_struct;
536

    
537
    /**
538
     * Complement sei_pic_struct
539
     * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
540
     * However, soft telecined frames may have these values.
541
     * This is used in an attempt to flag soft telecine progressive.
542
     */
543
    int prev_interlaced_frame;
544

    
545
    /**
546
     * Bit set of clock types for fields/frames in picture timing SEI message.
547
     * For each found ct_type, appropriate bit is set (e.g., bit 1 for
548
     * interlaced).
549
     */
550
    int sei_ct_type;
551

    
552
    /**
553
     * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
554
     */
555
    int sei_dpb_output_delay;
556

    
557
    /**
558
     * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
559
     */
560
    int sei_cpb_removal_delay;
561

    
562
    /**
563
     * recovery_frame_cnt from SEI message
564
     *
565
     * Set to -1 if no recovery point SEI message found or to number of frames
566
     * before playback synchronizes. Frames having recovery point are key
567
     * frames.
568
     */
569
    int sei_recovery_frame_cnt;
570

    
571
    int is_complex;
572

    
573
    int luma_weight_flag[2];   ///< 7.4.3.2 luma_weight_lX_flag
574
    int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
575

    
576
    // Timestamp stuff
577
    int sei_buffering_period_present;  ///< Buffering period SEI flag
578
    int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
579

    
580
    //SVQ3 specific fields
581
    int halfpel_flag;
582
    int thirdpel_flag;
583
    int unknown_svq3_flag;
584
    int next_slice_index;
585
    uint32_t svq3_watermark_key;
586
}H264Context;
587

    
588

    
589
extern const uint8_t ff_h264_chroma_qp[52];
590

    
591
void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
592

    
593
void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
594

    
595
/**
596
 * Decode SEI
597
 */
598
int ff_h264_decode_sei(H264Context *h);
599

    
600
/**
601
 * Decode SPS
602
 */
603
int ff_h264_decode_seq_parameter_set(H264Context *h);
604

    
605
/**
606
 * Decode PPS
607
 */
608
int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
609

    
610
/**
611
 * Decodes a network abstraction layer unit.
612
 * @param consumed is the number of bytes used as input
613
 * @param length is the length of the array
614
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
615
 * @returns decoded bytes, might be src+1 if no escapes
616
 */
617
const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
618

    
619
/**
620
 * identifies the exact end of the bitstream
621
 * @return the length of the trailing, or 0 if damaged
622
 */
623
int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
624

    
625
/**
626
 * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
627
 */
628
av_cold void ff_h264_free_context(H264Context *h);
629

    
630
/**
631
 * reconstructs bitstream slice_type.
632
 */
633
int ff_h264_get_slice_type(const H264Context *h);
634

    
635
/**
636
 * allocates tables.
637
 * needs width/height
638
 */
639
int ff_h264_alloc_tables(H264Context *h);
640

    
641
/**
642
 * fills the default_ref_list.
643
 */
644
int ff_h264_fill_default_ref_list(H264Context *h);
645

    
646
int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
647
void ff_h264_fill_mbaff_ref_list(H264Context *h);
648
void ff_h264_remove_all_refs(H264Context *h);
649

    
650
/**
651
 * Executes the reference picture marking (memory management control operations).
652
 */
653
int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
654

    
655
int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
656

    
657

    
658
/**
659
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
660
 */
661
int ff_h264_check_intra4x4_pred_mode(H264Context *h);
662

    
663
/**
664
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
665
 */
666
int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
667

    
668
void ff_h264_write_back_intra_pred_mode(H264Context *h);
669
void ff_h264_hl_decode_mb(H264Context *h);
670
int ff_h264_frame_start(H264Context *h);
671
av_cold int ff_h264_decode_init(AVCodecContext *avctx);
672
av_cold int ff_h264_decode_end(AVCodecContext *avctx);
673
av_cold void ff_h264_decode_init_vlc(void);
674

    
675
/**
676
 * decodes a macroblock
677
 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
678
 */
679
int ff_h264_decode_mb_cavlc(H264Context *h);
680

    
681
/**
682
 * decodes a CABAC coded macroblock
683
 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
684
 */
685
int ff_h264_decode_mb_cabac(H264Context *h);
686

    
687
void ff_h264_init_cabac_states(H264Context *h);
688

    
689
void ff_h264_direct_dist_scale_factor(H264Context * const h);
690
void ff_h264_direct_ref_list_init(H264Context * const h);
691
void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
692

    
693
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);
694
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);
695

    
696
/**
697
 * Reset SEI values at the beginning of the frame.
698
 *
699
 * @param h H.264 context.
700
 */
701
void ff_h264_reset_sei(H264Context *h);
702

    
703

    
704
/*
705
o-o o-o
706
 / / /
707
o-o o-o
708
 ,---'
709
o-o o-o
710
 / / /
711
o-o o-o
712
*/
713
//This table must be here because scan8[constant] must be known at compiletime
714
static const uint8_t scan8[16 + 2*4]={
715
 4+1*8, 5+1*8, 4+2*8, 5+2*8,
716
 6+1*8, 7+1*8, 6+2*8, 7+2*8,
717
 4+3*8, 5+3*8, 4+4*8, 5+4*8,
718
 6+3*8, 7+3*8, 6+4*8, 7+4*8,
719
 1+1*8, 2+1*8,
720
 1+2*8, 2+2*8,
721
 1+4*8, 2+4*8,
722
 1+5*8, 2+5*8,
723
};
724

    
725
static av_always_inline uint32_t pack16to32(int a, int b){
726
#if HAVE_BIGENDIAN
727
   return (b&0xFFFF) + (a<<16);
728
#else
729
   return (a&0xFFFF) + (b<<16);
730
#endif
731
}
732

    
733
static av_always_inline uint16_t pack8to16(int a, int b){
734
#if HAVE_BIGENDIAN
735
   return (b&0xFF) + (a<<8);
736
#else
737
   return (a&0xFF) + (b<<8);
738
#endif
739
}
740

    
741
/**
742
 * gets the chroma qp.
743
 */
744
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
745
    return h->pps.chroma_qp_table[t][qscale];
746
}
747

    
748
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
749

    
750
static void fill_decode_neighbors(H264Context *h, int mb_type){
751
    MpegEncContext * const s = &h->s;
752
    const int mb_xy= h->mb_xy;
753
    int topleft_xy, top_xy, topright_xy, left_xy[2];
754
    static const uint8_t left_block_options[4][16]={
755
        {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},
756
        {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},
757
        {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},
758
        {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}
759
    };
760

    
761
    h->topleft_partition= -1;
762

    
763
    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
764

    
765
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
766
     * stuff, I can't imagine that these complex rules are worth it. */
767

    
768
    topleft_xy = top_xy - 1;
769
    topright_xy= top_xy + 1;
770
    left_xy[1] = left_xy[0] = mb_xy-1;
771
    h->left_block = left_block_options[0];
772
    if(FRAME_MBAFF){
773
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
774
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
775
        if(s->mb_y&1){
776
            if (left_mb_field_flag != curr_mb_field_flag) {
777
                left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
778
                if (curr_mb_field_flag) {
779
                    left_xy[1] += s->mb_stride;
780
                    h->left_block = left_block_options[3];
781
                } else {
782
                    topleft_xy += s->mb_stride;
783
                    // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
784
                    h->topleft_partition = 0;
785
                    h->left_block = left_block_options[1];
786
                }
787
            }
788
        }else{
789
            if(curr_mb_field_flag){
790
                topleft_xy  += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
791
                topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
792
                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
793
            }
794
            if (left_mb_field_flag != curr_mb_field_flag) {
795
                if (curr_mb_field_flag) {
796
                    left_xy[1] += s->mb_stride;
797
                    h->left_block = left_block_options[3];
798
                } else {
799
                    h->left_block = left_block_options[2];
800
                }
801
            }
802
        }
803
    }
804

    
805
    h->topleft_mb_xy = topleft_xy;
806
    h->top_mb_xy     = top_xy;
807
    h->topright_mb_xy= topright_xy;
808
    h->left_mb_xy[0] = left_xy[0];
809
    h->left_mb_xy[1] = left_xy[1];
810
    //FIXME do we need all in the context?
811

    
812
    h->topleft_type = s->current_picture.mb_type[topleft_xy] ;
813
    h->top_type     = s->current_picture.mb_type[top_xy]     ;
814
    h->topright_type= s->current_picture.mb_type[topright_xy];
815
    h->left_type[0] = s->current_picture.mb_type[left_xy[0]] ;
816
    h->left_type[1] = s->current_picture.mb_type[left_xy[1]] ;
817

    
818
    if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0;
819
    if(h->slice_table[top_xy     ] != h->slice_num) h->top_type     = 0;
820
    if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
821
    if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
822
}
823

    
824
static void fill_decode_caches(H264Context *h, int mb_type){
825
    MpegEncContext * const s = &h->s;
826
    int topleft_xy, top_xy, topright_xy, left_xy[2];
827
    int topleft_type, top_type, topright_type, left_type[2];
828
    const uint8_t * left_block= h->left_block;
829
    int i;
830

    
831
    topleft_xy   = h->topleft_mb_xy ;
832
    top_xy       = h->top_mb_xy     ;
833
    topright_xy  = h->topright_mb_xy;
834
    left_xy[0]   = h->left_mb_xy[0] ;
835
    left_xy[1]   = h->left_mb_xy[1] ;
836
    topleft_type = h->topleft_type  ;
837
    top_type     = h->top_type      ;
838
    topright_type= h->topright_type ;
839
    left_type[0] = h->left_type[0]  ;
840
    left_type[1] = h->left_type[1]  ;
841

    
842
    if(!IS_SKIP(mb_type)){
843
        if(IS_INTRA(mb_type)){
844
            int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
845
            h->topleft_samples_available=
846
            h->top_samples_available=
847
            h->left_samples_available= 0xFFFF;
848
            h->topright_samples_available= 0xEEEA;
849

    
850
            if(!(top_type & type_mask)){
851
                h->topleft_samples_available= 0xB3FF;
852
                h->top_samples_available= 0x33FF;
853
                h->topright_samples_available= 0x26EA;
854
            }
855
            if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
856
                if(IS_INTERLACED(mb_type)){
857
                    if(!(left_type[0] & type_mask)){
858
                        h->topleft_samples_available&= 0xDFFF;
859
                        h->left_samples_available&= 0x5FFF;
860
                    }
861
                    if(!(left_type[1] & type_mask)){
862
                        h->topleft_samples_available&= 0xFF5F;
863
                        h->left_samples_available&= 0xFF5F;
864
                    }
865
                }else{
866
                    int left_typei = s->current_picture.mb_type[left_xy[0] + s->mb_stride];
867

    
868
                    assert(left_xy[0] == left_xy[1]);
869
                    if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
870
                        h->topleft_samples_available&= 0xDF5F;
871
                        h->left_samples_available&= 0x5F5F;
872
                    }
873
                }
874
            }else{
875
                if(!(left_type[0] & type_mask)){
876
                    h->topleft_samples_available&= 0xDF5F;
877
                    h->left_samples_available&= 0x5F5F;
878
                }
879
            }
880

    
881
            if(!(topleft_type & type_mask))
882
                h->topleft_samples_available&= 0x7FFF;
883

    
884
            if(!(topright_type & type_mask))
885
                h->topright_samples_available&= 0xFBFF;
886

    
887
            if(IS_INTRA4x4(mb_type)){
888
                if(IS_INTRA4x4(top_type)){
889
                    AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]);
890
                }else{
891
                    h->intra4x4_pred_mode_cache[4+8*0]=
892
                    h->intra4x4_pred_mode_cache[5+8*0]=
893
                    h->intra4x4_pred_mode_cache[6+8*0]=
894
                    h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask);
895
                }
896
                for(i=0; i<2; i++){
897
                    if(IS_INTRA4x4(left_type[i])){
898
                        int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]];
899
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]];
900
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]];
901
                    }else{
902
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
903
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[i] & type_mask);
904
                    }
905
                }
906
            }
907
        }
908

    
909

    
910
/*
911
0 . T T. T T T T
912
1 L . .L . . . .
913
2 L . .L . . . .
914
3 . T TL . . . .
915
4 L . .L . . . .
916
5 L . .. . . . .
917
*/
918
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
919
    if(top_type){
920
        AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
921
            h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
922
            h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
923

    
924
            h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
925
            h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
926
    }else {
927
            h->non_zero_count_cache[1+8*0]=
928
            h->non_zero_count_cache[2+8*0]=
929

    
930
            h->non_zero_count_cache[1+8*3]=
931
            h->non_zero_count_cache[2+8*3]=
932
            AV_WN32A(&h->non_zero_count_cache[4+8*0], CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040);
933
    }
934

    
935
    for (i=0; i<2; i++) {
936
        if(left_type[i]){
937
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
938
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
939
                h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
940
                h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
941
        }else{
942
                h->non_zero_count_cache[3+8*1 + 2*8*i]=
943
                h->non_zero_count_cache[3+8*2 + 2*8*i]=
944
                h->non_zero_count_cache[0+8*1 +   8*i]=
945
                h->non_zero_count_cache[0+8*4 +   8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
946
        }
947
    }
948

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

    
967
#if 1
968
    if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
969
        int list;
970
        for(list=0; list<h->list_count; list++){
971
            if(!USES_LIST(mb_type, list)){
972
                /*if(!h->mv_cache_clean[list]){
973
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
974
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
975
                    h->mv_cache_clean[list]= 1;
976
                }*/
977
                continue;
978
            }
979
            assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
980

    
981
            h->mv_cache_clean[list]= 0;
982

    
983
            if(USES_LIST(top_type, list)){
984
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
985
                AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
986
                    h->ref_cache[list][scan8[0] + 0 - 1*8]=
987
                    h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 2];
988
                    h->ref_cache[list][scan8[0] + 2 - 1*8]=
989
                    h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 3];
990
            }else{
991
                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
992
                AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
993
            }
994

    
995
            if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){
996
            for(i=0; i<2; i++){
997
                int cache_idx = scan8[0] - 1 + i*2*8;
998
                if(USES_LIST(left_type[i], list)){
999
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1000
                    const int b8_xy= 4*left_xy[i] + 1;
1001
                    AV_COPY32(h->mv_cache[list][cache_idx  ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]);
1002
                    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]]);
1003
                        h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + (left_block[0+i*2]&~1)];
1004
                        h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + (left_block[1+i*2]&~1)];
1005
                }else{
1006
                    AV_ZERO32(h->mv_cache [list][cache_idx  ]);
1007
                    AV_ZERO32(h->mv_cache [list][cache_idx+8]);
1008
                    h->ref_cache[list][cache_idx  ]=
1009
                    h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1010
                }
1011
            }
1012
            }else{
1013
                if(USES_LIST(left_type[0], list)){
1014
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1015
                    const int b8_xy= 4*left_xy[0] + 1;
1016
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]]);
1017
                    h->ref_cache[list][scan8[0] - 1]= s->current_picture.ref_index[list][b8_xy + (left_block[0]&~1)];
1018
                }else{
1019
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1]);
1020
                    h->ref_cache[list][scan8[0] - 1]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1021
                }
1022
            }
1023

    
1024
            if(USES_LIST(topright_type, list)){
1025
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1026
                AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]);
1027
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][4*topright_xy + 2];
1028
            }else{
1029
                AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]);
1030
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1031
            }
1032
            if(h->ref_cache[list][scan8[0] + 4 - 1*8] < 0){
1033
                if(USES_LIST(topleft_type, list)){
1034
                    const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
1035
                    const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2);
1036
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]);
1037
                    h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1038
                }else{
1039
                    AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]);
1040
                    h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1041
                }
1042
            }
1043

    
1044
            if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
1045
                continue;
1046

    
1047
            if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
1048
            h->ref_cache[list][scan8[4 ]] =
1049
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1050
            AV_ZERO32(h->mv_cache [list][scan8[4 ]]);
1051
            AV_ZERO32(h->mv_cache [list][scan8[12]]);
1052

    
1053
            if( CABAC ) {
1054
                /* XXX beurk, Load mvd */
1055
                if(USES_LIST(top_type, list)){
1056
                    const int b_xy= h->mb2br_xy[top_xy];
1057
                    AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1058
                }else{
1059
                    AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1060
                }
1061
                if(USES_LIST(left_type[0], list)){
1062
                    const int b_xy= h->mb2br_xy[left_xy[0]] + 6;
1063
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy - left_block[0]]);
1064
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy - left_block[1]]);
1065
                }else{
1066
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]);
1067
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]);
1068
                }
1069
                if(USES_LIST(left_type[1], list)){
1070
                    const int b_xy= h->mb2br_xy[left_xy[1]] + 6;
1071
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy - left_block[2]]);
1072
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy - left_block[3]]);
1073
                }else{
1074
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]);
1075
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]);
1076
                }
1077
                AV_ZERO16(h->mvd_cache [list][scan8[4 ]]);
1078
                AV_ZERO16(h->mvd_cache [list][scan8[12]]);
1079
                if(h->slice_type_nos == FF_B_TYPE){
1080
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
1081

    
1082
                    if(IS_DIRECT(top_type)){
1083
                        AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_DIRECT2>>1));
1084
                    }else if(IS_8X8(top_type)){
1085
                        int b8_xy = 4*top_xy;
1086
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy + 2];
1087
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 3];
1088
                    }else{
1089
                        AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1));
1090
                    }
1091

    
1092
                    if(IS_DIRECT(left_type[0]))
1093
                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
1094
                    else if(IS_8X8(left_type[0]))
1095
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[4*left_xy[0] + 1 + (left_block[0]&~1)];
1096
                    else
1097
                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
1098

    
1099
                    if(IS_DIRECT(left_type[1]))
1100
                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
1101
                    else if(IS_8X8(left_type[1]))
1102
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[4*left_xy[1] + 1 + (left_block[2]&~1)];
1103
                    else
1104
                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
1105
                }
1106
            }
1107
            }
1108
            if(FRAME_MBAFF){
1109
#define MAP_MVS\
1110
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1111
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1112
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1113
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1114
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1115
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1116
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1117
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1118
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1119
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1120
                if(MB_FIELD){
1121
#define MAP_F2F(idx, mb_type)\
1122
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1123
                        h->ref_cache[list][idx] <<= 1;\
1124
                        h->mv_cache[list][idx][1] /= 2;\
1125
                        h->mvd_cache[list][idx][1] >>=1;\
1126
                    }
1127
                    MAP_MVS
1128
#undef MAP_F2F
1129
                }else{
1130
#define MAP_F2F(idx, mb_type)\
1131
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1132
                        h->ref_cache[list][idx] >>= 1;\
1133
                        h->mv_cache[list][idx][1] <<= 1;\
1134
                        h->mvd_cache[list][idx][1] <<= 1;\
1135
                    }
1136
                    MAP_MVS
1137
#undef MAP_F2F
1138
                }
1139
            }
1140
        }
1141
    }
1142
#endif
1143

    
1144
        h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1145
}
1146

    
1147
/**
1148
 *
1149
 * @returns non zero if the loop filter can be skiped
1150
 */
1151
static int fill_filter_caches(H264Context *h, int mb_type){
1152
    MpegEncContext * const s = &h->s;
1153
    const int mb_xy= h->mb_xy;
1154
    int top_xy, left_xy[2];
1155
    int top_type, left_type[2];
1156

    
1157
    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
1158

    
1159
    //FIXME deblocking could skip the intra and nnz parts.
1160

    
1161
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
1162
     * stuff, I can't imagine that these complex rules are worth it. */
1163

    
1164
    left_xy[1] = left_xy[0] = mb_xy-1;
1165
    if(FRAME_MBAFF){
1166
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
1167
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
1168
        if(s->mb_y&1){
1169
            if (left_mb_field_flag != curr_mb_field_flag) {
1170
                left_xy[0] -= s->mb_stride;
1171
            }
1172
        }else{
1173
            if(curr_mb_field_flag){
1174
                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
1175
            }
1176
            if (left_mb_field_flag != curr_mb_field_flag) {
1177
                left_xy[1] += s->mb_stride;
1178
            }
1179
        }
1180
    }
1181

    
1182
    h->top_mb_xy = top_xy;
1183
    h->left_mb_xy[0] = left_xy[0];
1184
    h->left_mb_xy[1] = left_xy[1];
1185
    {
1186
        //for sufficiently low qp, filtering wouldn't do anything
1187
        //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
1188
        int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
1189
        int qp = s->current_picture.qscale_table[mb_xy];
1190
        if(qp <= qp_thresh
1191
           && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
1192
           && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
1193
            if(!FRAME_MBAFF)
1194
                return 1;
1195
            if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
1196
               && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
1197
                return 1;
1198
        }
1199
    }
1200

    
1201
    if(h->deblocking_filter == 2){
1202
        h->top_type    = top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
1203
        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;
1204
        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;
1205
    }else{
1206
        h->top_type    = top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
1207
        h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
1208
        h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
1209
    }
1210
    if(IS_INTRA(mb_type))
1211
        return 0;
1212

    
1213
    AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
1214
    AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
1215
    AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
1216
    AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
1217
    AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
1218

    
1219
    h->cbp= h->cbp_table[mb_xy];
1220

    
1221
    {
1222
        int list;
1223
        for(list=0; list<h->list_count; list++){
1224
            int8_t *ref;
1225
            int y, b_stride;
1226
            int16_t (*mv_dst)[2];
1227
            int16_t (*mv_src)[2];
1228

    
1229
            if(!USES_LIST(mb_type, list)){
1230
                fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
1231
                AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1232
                AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1233
                AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1234
                AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1235
                continue;
1236
            }
1237

    
1238
            ref = &s->current_picture.ref_index[list][4*mb_xy];
1239
            {
1240
                int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1241
                AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1242
                AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1243
                ref += 2;
1244
                AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1245
                AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1246
            }
1247

    
1248
            b_stride = h->b_stride;
1249
            mv_dst   = &h->mv_cache[list][scan8[0]];
1250
            mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
1251
            for(y=0; y<4; y++){
1252
                AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
1253
            }
1254

    
1255
        }
1256
    }
1257

    
1258

    
1259
/*
1260
0 . T T. T T T T
1261
1 L . .L . . . .
1262
2 L . .L . . . .
1263
3 . T TL . . . .
1264
4 L . .L . . . .
1265
5 L . .. . . . .
1266
*/
1267
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
1268
    if(top_type){
1269
        AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
1270
    }
1271

    
1272
    if(left_type[0]){
1273
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
1274
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
1275
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
1276
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
1277
    }
1278

    
1279
    // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
1280
    if(!CABAC && h->pps.transform_8x8_mode){
1281
        if(IS_8x8DCT(top_type)){
1282
            h->non_zero_count_cache[4+8*0]=
1283
            h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
1284
            h->non_zero_count_cache[6+8*0]=
1285
            h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
1286
        }
1287
        if(IS_8x8DCT(left_type[0])){
1288
            h->non_zero_count_cache[3+8*1]=
1289
            h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
1290
        }
1291
        if(IS_8x8DCT(left_type[1])){
1292
            h->non_zero_count_cache[3+8*3]=
1293
            h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
1294
        }
1295

    
1296
        if(IS_8x8DCT(mb_type)){
1297
            h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
1298
            h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
1299

    
1300
            h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
1301
            h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
1302

    
1303
            h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
1304
            h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
1305

    
1306
            h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
1307
            h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
1308
        }
1309
    }
1310

    
1311
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
1312
        int list;
1313
        for(list=0; list<h->list_count; list++){
1314
            if(USES_LIST(top_type, list)){
1315
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1316
                const int b8_xy= 4*top_xy + 2;
1317
                int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1318
                AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1319
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
1320
                h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
1321
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
1322
                h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
1323
            }else{
1324
                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1325
                AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1326
            }
1327

    
1328
            if(!IS_INTERLACED(mb_type^left_type[0])){
1329
                if(USES_LIST(left_type[0], list)){
1330
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1331
                    const int b8_xy= 4*left_xy[0] + 1;
1332
                    int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1333
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
1334
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
1335
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
1336
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
1337
                    h->ref_cache[list][scan8[0] - 1 + 0 ]=
1338
                    h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
1339
                    h->ref_cache[list][scan8[0] - 1 +16 ]=
1340
                    h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
1341
                }else{
1342
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
1343
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
1344
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
1345
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
1346
                    h->ref_cache[list][scan8[0] - 1 + 0  ]=
1347
                    h->ref_cache[list][scan8[0] - 1 + 8  ]=
1348
                    h->ref_cache[list][scan8[0] - 1 + 16 ]=
1349
                    h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
1350
                }
1351
            }
1352
        }
1353
    }
1354

    
1355
    return 0;
1356
}
1357

    
1358
/**
1359
 * gets the predicted intra4x4 prediction mode.
1360
 */
1361
static inline int pred_intra_mode(H264Context *h, int n){
1362
    const int index8= scan8[n];
1363
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1364
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1365
    const int min= FFMIN(left, top);
1366

    
1367
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1368

    
1369
    if(min<0) return DC_PRED;
1370
    else      return min;
1371
}
1372

    
1373
static inline void write_back_non_zero_count(H264Context *h){
1374
    const int mb_xy= h->mb_xy;
1375

    
1376
    AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1377
    AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1378
    AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[0+8*5]);
1379
    AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8*3]);
1380
    AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1381
}
1382

    
1383
static inline void write_back_motion(H264Context *h, int mb_type){
1384
    MpegEncContext * const s = &h->s;
1385
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
1386
    const int b8_xy= 4*h->mb_xy;
1387
    int list;
1388

    
1389
    if(!USES_LIST(mb_type, 0))
1390
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1391

    
1392
    for(list=0; list<h->list_count; list++){
1393
        int y, b_stride;
1394
        int16_t (*mv_dst)[2];
1395
        int16_t (*mv_src)[2];
1396

    
1397
        if(!USES_LIST(mb_type, list))
1398
            continue;
1399

    
1400
        b_stride = h->b_stride;
1401
        mv_dst   = &s->current_picture.motion_val[list][b_xy];
1402
        mv_src   = &h->mv_cache[list][scan8[0]];
1403
        for(y=0; y<4; y++){
1404
            AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1405
        }
1406
        if( CABAC ) {
1407
            uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
1408
            uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1409
            if(IS_SKIP(mb_type))
1410
                AV_ZERO128(mvd_dst);
1411
            else{
1412
            AV_COPY64(mvd_dst, mvd_src + 8*3);
1413
                AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
1414
                AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
1415
                AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
1416
            }
1417
        }
1418

    
1419
        {
1420
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1421
            ref_index[0+0*2]= h->ref_cache[list][scan8[0]];
1422
            ref_index[1+0*2]= h->ref_cache[list][scan8[4]];
1423
            ref_index[0+1*2]= h->ref_cache[list][scan8[8]];
1424
            ref_index[1+1*2]= h->ref_cache[list][scan8[12]];
1425
        }
1426
    }
1427

    
1428
    if(h->slice_type_nos == FF_B_TYPE && CABAC){
1429
        if(IS_8X8(mb_type)){
1430
            uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
1431
            direct_table[1] = h->sub_mb_type[1]>>1;
1432
            direct_table[2] = h->sub_mb_type[2]>>1;
1433
            direct_table[3] = h->sub_mb_type[3]>>1;
1434
        }
1435
    }
1436
}
1437

    
1438
static inline int get_dct8x8_allowed(H264Context *h){
1439
    if(h->sps.direct_8x8_inference_flag)
1440
        return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
1441
    else
1442
        return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1443
}
1444

    
1445
/**
1446
 * decodes a P_SKIP or B_SKIP macroblock
1447
 */
1448
static void decode_mb_skip(H264Context *h){
1449
    MpegEncContext * const s = &h->s;
1450
    const int mb_xy= h->mb_xy;
1451
    int mb_type=0;
1452

    
1453
    memset(h->non_zero_count[mb_xy], 0, 32);
1454
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1455

    
1456
    if(MB_FIELD)
1457
        mb_type|= MB_TYPE_INTERLACED;
1458

    
1459
    if( h->slice_type_nos == FF_B_TYPE )
1460
    {
1461
        // just for fill_caches. pred_direct_motion will set the real mb_type
1462
        mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1463
        if(h->direct_spatial_mv_pred){
1464
            fill_decode_neighbors(h, mb_type);
1465
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1466
        }
1467
        ff_h264_pred_direct_motion(h, &mb_type);
1468
        mb_type|= MB_TYPE_SKIP;
1469
    }
1470
    else
1471
    {
1472
        int mx, my;
1473
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1474

    
1475
        fill_decode_neighbors(h, mb_type);
1476
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1477
        pred_pskip_motion(h, &mx, &my);
1478
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1479
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1480
    }
1481

    
1482
    write_back_motion(h, mb_type);
1483
    s->current_picture.mb_type[mb_xy]= mb_type;
1484
    s->current_picture.qscale_table[mb_xy]= s->qscale;
1485
    h->slice_table[ mb_xy ]= h->slice_num;
1486
    h->prev_mb_skipped= 1;
1487
}
1488

    
1489
#include "h264_mvpred.h" //For pred_pskip_motion()
1490

    
1491
#endif /* AVCODEC_H264_H */