Statistics
| Branch: | Revision:

ffmpeg / libavcodec / h264.h @ e1c88a21

History | View | Annotate | Download (56.4 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)[8];
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
    uint32_t *mb2b8_xy;
351
    int b_stride; //FIXME use s->b4_stride
352
    int b8_stride;
353

    
354
    int mb_linesize;   ///< may be equal to s->linesize or s->linesize*2, for mbaff
355
    int mb_uvlinesize;
356

    
357
    int emu_edge_width;
358
    int emu_edge_height;
359

    
360
    int halfpel_flag;
361
    int thirdpel_flag;
362

    
363
    int unknown_svq3_flag;
364
    int next_slice_index;
365

    
366
    SPS *sps_buffers[MAX_SPS_COUNT];
367
    SPS sps; ///< current sps
368

    
369
    PPS *pps_buffers[MAX_PPS_COUNT];
370
    /**
371
     * current pps
372
     */
373
    PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
374

    
375
    uint32_t dequant4_buffer[6][52][16];
376
    uint32_t dequant8_buffer[2][52][64];
377
    uint32_t (*dequant4_coeff[6])[16];
378
    uint32_t (*dequant8_coeff[2])[64];
379
    int dequant_coeff_pps;     ///< reinit tables when pps changes
380

    
381
    int slice_num;
382
    uint16_t *slice_table_base;
383
    uint16_t *slice_table;     ///< slice_table_base + 2*mb_stride + 1
384
    int slice_type;
385
    int slice_type_nos;        ///< S free slice type (SI/SP are remapped to I/P)
386
    int slice_type_fixed;
387

    
388
    //interlacing specific flags
389
    int mb_aff_frame;
390
    int mb_field_decoding_flag;
391
    int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag
392

    
393
    DECLARE_ALIGNED_8(uint16_t, sub_mb_type)[4];
394

    
395
    //POC stuff
396
    int poc_lsb;
397
    int poc_msb;
398
    int delta_poc_bottom;
399
    int delta_poc[2];
400
    int frame_num;
401
    int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
402
    int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
403
    int frame_num_offset;         ///< for POC type 2
404
    int prev_frame_num_offset;    ///< for POC type 2
405
    int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
406

    
407
    /**
408
     * frame_num for frames or 2*frame_num+1 for field pics.
409
     */
410
    int curr_pic_num;
411

    
412
    /**
413
     * max_frame_num or 2*max_frame_num for field pics.
414
     */
415
    int max_pic_num;
416

    
417
    //Weighted pred stuff
418
    int use_weight;
419
    int use_weight_chroma;
420
    int luma_log2_weight_denom;
421
    int chroma_log2_weight_denom;
422
    int luma_weight[2][48];
423
    int luma_offset[2][48];
424
    int chroma_weight[2][48][2];
425
    int chroma_offset[2][48][2];
426
    int implicit_weight[48][48];
427

    
428
    //deblock
429
    int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
430
    int slice_alpha_c0_offset;
431
    int slice_beta_offset;
432

    
433
    int redundant_pic_count;
434

    
435
    int direct_spatial_mv_pred;
436
    int col_parity;
437
    int col_fieldoff;
438
    int dist_scale_factor[16];
439
    int dist_scale_factor_field[2][32];
440
    int map_col_to_list0[2][16+32];
441
    int map_col_to_list0_field[2][2][16+32];
442

    
443
    /**
444
     * num_ref_idx_l0/1_active_minus1 + 1
445
     */
446
    unsigned int ref_count[2];   ///< counts frames or fields, depending on current mb mode
447
    unsigned int list_count;
448
    uint8_t *list_counts;            ///< Array of list_count per MB specifying the slice type
449
    Picture *short_ref[32];
450
    Picture *long_ref[32];
451
    Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
452
    Picture ref_list[2][48];         /**< 0..15: frame refs, 16..47: mbaff field refs.
453
                                          Reordered version of default_ref_list
454
                                          according to picture reordering in slice header */
455
    int ref2frm[MAX_SLICES][2][64];  ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
456
    Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
457
    int outputed_poc;
458

    
459
    /**
460
     * memory management control operations buffer.
461
     */
462
    MMCO mmco[MAX_MMCO_COUNT];
463
    int mmco_index;
464

    
465
    int long_ref_count;  ///< number of actual long term references
466
    int short_ref_count; ///< number of actual short term references
467

    
468
    //data partitioning
469
    GetBitContext intra_gb;
470
    GetBitContext inter_gb;
471
    GetBitContext *intra_gb_ptr;
472
    GetBitContext *inter_gb_ptr;
473

    
474
    DECLARE_ALIGNED_16(DCTELEM, mb)[16*24];
475
    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
476

    
477
    /**
478
     * Cabac
479
     */
480
    CABACContext cabac;
481
    uint8_t      cabac_state[460];
482
    int          cabac_init_idc;
483

    
484
    /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
485
    uint16_t     *cbp_table;
486
    int cbp;
487
    int top_cbp;
488
    int left_cbp;
489
    /* chroma_pred_mode for i4x4 or i16x16, else 0 */
490
    uint8_t     *chroma_pred_mode_table;
491
    int         last_qscale_diff;
492
    uint8_t     (*mvd_table[2])[2];
493
    DECLARE_ALIGNED_16(uint8_t, mvd_cache)[2][5*8][2];
494
    uint8_t     *direct_table;
495
    uint8_t     direct_cache[5*8];
496

    
497
    uint8_t zigzag_scan[16];
498
    uint8_t zigzag_scan8x8[64];
499
    uint8_t zigzag_scan8x8_cavlc[64];
500
    uint8_t field_scan[16];
501
    uint8_t field_scan8x8[64];
502
    uint8_t field_scan8x8_cavlc[64];
503
    const uint8_t *zigzag_scan_q0;
504
    const uint8_t *zigzag_scan8x8_q0;
505
    const uint8_t *zigzag_scan8x8_cavlc_q0;
506
    const uint8_t *field_scan_q0;
507
    const uint8_t *field_scan8x8_q0;
508
    const uint8_t *field_scan8x8_cavlc_q0;
509

    
510
    int x264_build;
511

    
512
    /**
513
     * @defgroup multithreading Members for slice based multithreading
514
     * @{
515
     */
516
    struct H264Context *thread_context[MAX_THREADS];
517

    
518
    /**
519
     * current slice number, used to initalize slice_num of each thread/context
520
     */
521
    int current_slice;
522

    
523
    /**
524
     * Max number of threads / contexts.
525
     * This is equal to AVCodecContext.thread_count unless
526
     * multithreaded decoding is impossible, in which case it is
527
     * reduced to 1.
528
     */
529
    int max_contexts;
530

    
531
    /**
532
     *  1 if the single thread fallback warning has already been
533
     *  displayed, 0 otherwise.
534
     */
535
    int single_decode_warning;
536

    
537
    int last_slice_type;
538
    /** @} */
539

    
540
    int mb_xy;
541

    
542
    uint32_t svq3_watermark_key;
543

    
544
    /**
545
     * pic_struct in picture timing SEI message
546
     */
547
    SEI_PicStructType sei_pic_struct;
548

    
549
    /**
550
     * Complement sei_pic_struct
551
     * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
552
     * However, soft telecined frames may have these values.
553
     * This is used in an attempt to flag soft telecine progressive.
554
     */
555
    int prev_interlaced_frame;
556

    
557
    /**
558
     * Bit set of clock types for fields/frames in picture timing SEI message.
559
     * For each found ct_type, appropriate bit is set (e.g., bit 1 for
560
     * interlaced).
561
     */
562
    int sei_ct_type;
563

    
564
    /**
565
     * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
566
     */
567
    int sei_dpb_output_delay;
568

    
569
    /**
570
     * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
571
     */
572
    int sei_cpb_removal_delay;
573

    
574
    /**
575
     * recovery_frame_cnt from SEI message
576
     *
577
     * Set to -1 if no recovery point SEI message found or to number of frames
578
     * before playback synchronizes. Frames having recovery point are key
579
     * frames.
580
     */
581
    int sei_recovery_frame_cnt;
582

    
583
    int is_complex;
584

    
585
    int luma_weight_flag[2];   ///< 7.4.3.2 luma_weight_lX_flag
586
    int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
587

    
588
    // Timestamp stuff
589
    int sei_buffering_period_present;  ///< Buffering period SEI flag
590
    int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
591
}H264Context;
592

    
593

    
594
extern const uint8_t ff_h264_chroma_qp[52];
595

    
596
void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
597

    
598
void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
599

    
600
/**
601
 * Decode SEI
602
 */
603
int ff_h264_decode_sei(H264Context *h);
604

    
605
/**
606
 * Decode SPS
607
 */
608
int ff_h264_decode_seq_parameter_set(H264Context *h);
609

    
610
/**
611
 * Decode PPS
612
 */
613
int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
614

    
615
/**
616
 * Decodes a network abstraction layer unit.
617
 * @param consumed is the number of bytes used as input
618
 * @param length is the length of the array
619
 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
620
 * @returns decoded bytes, might be src+1 if no escapes
621
 */
622
const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
623

    
624
/**
625
 * identifies the exact end of the bitstream
626
 * @return the length of the trailing, or 0 if damaged
627
 */
628
int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
629

    
630
/**
631
 * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
632
 */
633
av_cold void ff_h264_free_context(H264Context *h);
634

    
635
/**
636
 * reconstructs bitstream slice_type.
637
 */
638
int ff_h264_get_slice_type(const H264Context *h);
639

    
640
/**
641
 * allocates tables.
642
 * needs width/height
643
 */
644
int ff_h264_alloc_tables(H264Context *h);
645

    
646
/**
647
 * fills the default_ref_list.
648
 */
649
int ff_h264_fill_default_ref_list(H264Context *h);
650

    
651
int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
652
void ff_h264_fill_mbaff_ref_list(H264Context *h);
653
void ff_h264_remove_all_refs(H264Context *h);
654

    
655
/**
656
 * Executes the reference picture marking (memory management control operations).
657
 */
658
int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
659

    
660
int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
661

    
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_intra4x4_pred_mode(H264Context *h);
667

    
668
/**
669
 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
670
 */
671
int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
672

    
673
void ff_h264_write_back_intra_pred_mode(H264Context *h);
674
void ff_h264_hl_decode_mb(H264Context *h);
675
int ff_h264_frame_start(H264Context *h);
676
av_cold int ff_h264_decode_init(AVCodecContext *avctx);
677
av_cold int ff_h264_decode_end(AVCodecContext *avctx);
678
av_cold void ff_h264_decode_init_vlc(void);
679

    
680
/**
681
 * decodes a macroblock
682
 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
683
 */
684
int ff_h264_decode_mb_cavlc(H264Context *h);
685

    
686
/**
687
 * decodes a CABAC coded macroblock
688
 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
689
 */
690
int ff_h264_decode_mb_cabac(H264Context *h);
691

    
692
void ff_h264_init_cabac_states(H264Context *h);
693

    
694
void ff_h264_direct_dist_scale_factor(H264Context * const h);
695
void ff_h264_direct_ref_list_init(H264Context * const h);
696
void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
697

    
698
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);
699
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);
700

    
701
/**
702
 * Reset SEI values at the beginning of the frame.
703
 *
704
 * @param h H.264 context.
705
 */
706
void ff_h264_reset_sei(H264Context *h);
707

    
708

    
709
/*
710
o-o o-o
711
 / / /
712
o-o o-o
713
 ,---'
714
o-o o-o
715
 / / /
716
o-o o-o
717
*/
718
//This table must be here because scan8[constant] must be known at compiletime
719
static const uint8_t scan8[16 + 2*4]={
720
 4+1*8, 5+1*8, 4+2*8, 5+2*8,
721
 6+1*8, 7+1*8, 6+2*8, 7+2*8,
722
 4+3*8, 5+3*8, 4+4*8, 5+4*8,
723
 6+3*8, 7+3*8, 6+4*8, 7+4*8,
724
 1+1*8, 2+1*8,
725
 1+2*8, 2+2*8,
726
 1+4*8, 2+4*8,
727
 1+5*8, 2+5*8,
728
};
729

    
730
static av_always_inline uint32_t pack16to32(int a, int b){
731
#if HAVE_BIGENDIAN
732
   return (b&0xFFFF) + (a<<16);
733
#else
734
   return (a&0xFFFF) + (b<<16);
735
#endif
736
}
737

    
738
static av_always_inline uint16_t pack8to16(int a, int b){
739
#if HAVE_BIGENDIAN
740
   return (b&0xFF) + (a<<8);
741
#else
742
   return (a&0xFF) + (b<<8);
743
#endif
744
}
745

    
746
/**
747
 * gets the chroma qp.
748
 */
749
static inline int get_chroma_qp(H264Context *h, int t, int qscale){
750
    return h->pps.chroma_qp_table[t][qscale];
751
}
752

    
753
static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
754

    
755
static void fill_decode_neighbors(H264Context *h, int mb_type){
756
    MpegEncContext * const s = &h->s;
757
    const int mb_xy= h->mb_xy;
758
    int topleft_xy, top_xy, topright_xy, left_xy[2];
759
    static const uint8_t left_block_options[4][16]={
760
        {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},
761
        {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},
762
        {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},
763
        {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}
764
    };
765

    
766
    h->topleft_partition= -1;
767

    
768
    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
769

    
770
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
771
     * stuff, I can't imagine that these complex rules are worth it. */
772

    
773
    topleft_xy = top_xy - 1;
774
    topright_xy= top_xy + 1;
775
    left_xy[1] = left_xy[0] = mb_xy-1;
776
    h->left_block = left_block_options[0];
777
    if(FRAME_MBAFF){
778
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
779
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
780
        if(s->mb_y&1){
781
            if (left_mb_field_flag != curr_mb_field_flag) {
782
                left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
783
                if (curr_mb_field_flag) {
784
                    left_xy[1] += s->mb_stride;
785
                    h->left_block = left_block_options[3];
786
                } else {
787
                    topleft_xy += s->mb_stride;
788
                    // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
789
                    h->topleft_partition = 0;
790
                    h->left_block = left_block_options[1];
791
                }
792
            }
793
        }else{
794
            if(curr_mb_field_flag){
795
                topleft_xy  += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
796
                topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
797
                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
798
            }
799
            if (left_mb_field_flag != curr_mb_field_flag) {
800
                left_xy[1] = left_xy[0] = mb_xy - 1;
801
                if (curr_mb_field_flag) {
802
                    left_xy[1] += s->mb_stride;
803
                    h->left_block = left_block_options[3];
804
                } else {
805
                    h->left_block = left_block_options[2];
806
                }
807
            }
808
        }
809
    }
810

    
811
    h->topleft_mb_xy = topleft_xy;
812
    h->top_mb_xy     = top_xy;
813
    h->topright_mb_xy= topright_xy;
814
    h->left_mb_xy[0] = left_xy[0];
815
    h->left_mb_xy[1] = left_xy[1];
816
    //FIXME do we need all in the context?
817
    h->topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
818
    h->top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
819
    h->topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
820
    h->left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
821
    h->left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[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 = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
867
                                    ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
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
                    h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
890
                    h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
891
                    h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
892
                    h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
893
                }else{
894
                    int pred;
895
                    if(!(top_type & type_mask))
896
                        pred= -1;
897
                    else{
898
                        pred= 2;
899
                    }
900
                    h->intra4x4_pred_mode_cache[4+8*0]=
901
                    h->intra4x4_pred_mode_cache[5+8*0]=
902
                    h->intra4x4_pred_mode_cache[6+8*0]=
903
                    h->intra4x4_pred_mode_cache[7+8*0]= pred;
904
                }
905
                for(i=0; i<2; i++){
906
                    if(IS_INTRA4x4(left_type[i])){
907
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
908
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
909
                    }else{
910
                        int pred;
911
                        if(!(left_type[i] & type_mask))
912
                            pred= -1;
913
                        else{
914
                            pred= 2;
915
                        }
916
                        h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
917
                        h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
918
                    }
919
                }
920
            }
921
        }
922

    
923

    
924
/*
925
0 . T T. T T T T
926
1 L . .L . . . .
927
2 L . .L . . . .
928
3 . T TL . . . .
929
4 L . .L . . . .
930
5 L . .. . . . .
931
*/
932
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
933
    if(top_type){
934
        AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
935
            h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
936
            h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
937

    
938
            h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
939
            h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
940
    }else {
941
            h->non_zero_count_cache[1+8*0]=
942
            h->non_zero_count_cache[2+8*0]=
943

    
944
            h->non_zero_count_cache[1+8*3]=
945
            h->non_zero_count_cache[2+8*3]=
946
            AV_WN32A(&h->non_zero_count_cache[4+8*0], CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040);
947
    }
948

    
949
    for (i=0; i<2; i++) {
950
        if(left_type[i]){
951
            h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
952
            h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
953
                h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
954
                h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
955
        }else{
956
                h->non_zero_count_cache[3+8*1 + 2*8*i]=
957
                h->non_zero_count_cache[3+8*2 + 2*8*i]=
958
                h->non_zero_count_cache[0+8*1 +   8*i]=
959
                h->non_zero_count_cache[0+8*4 +   8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
960
        }
961
    }
962

    
963
    if( CABAC ) {
964
        // top_cbp
965
        if(top_type) {
966
            h->top_cbp = h->cbp_table[top_xy];
967
        } else if(IS_INTRA(mb_type)) {
968
            h->top_cbp = 0x1CF;
969
        } else {
970
            h->top_cbp = 0x00F;
971
        }
972
        // left_cbp
973
        if (left_type[0]) {
974
            h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
975
        } else if(IS_INTRA(mb_type)) {
976
            h->left_cbp = 0x1CF;
977
        } else {
978
            h->left_cbp = 0x00F;
979
        }
980
        if (left_type[0]) {
981
            h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
982
        }
983
        if (left_type[1]) {
984
            h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
985
        }
986
    }
987
    }
988

    
989
#if 1
990
    if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
991
        int list;
992
        for(list=0; list<h->list_count; list++){
993
            if(!USES_LIST(mb_type, list)){
994
                /*if(!h->mv_cache_clean[list]){
995
                    memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
996
                    memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
997
                    h->mv_cache_clean[list]= 1;
998
                }*/
999
                continue;
1000
            }
1001
            assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
1002

    
1003
            h->mv_cache_clean[list]= 0;
1004

    
1005
            if(USES_LIST(top_type, list)){
1006
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1007
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
1008
                AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1009
                    h->ref_cache[list][scan8[0] + 0 - 1*8]=
1010
                    h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
1011
                    h->ref_cache[list][scan8[0] + 2 - 1*8]=
1012
                    h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
1013
            }else{
1014
                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1015
                AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
1016
            }
1017

    
1018
            for(i=0; i<2; i++){
1019
                int cache_idx = scan8[0] - 1 + i*2*8;
1020
                if(USES_LIST(left_type[i], list)){
1021
                    const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1022
                    const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
1023
                    AV_COPY32(h->mv_cache[list][cache_idx  ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]);
1024
                    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]]);
1025
                        h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
1026
                        h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
1027
                }else{
1028
                    AV_ZERO32(h->mv_cache [list][cache_idx  ]);
1029
                    AV_ZERO32(h->mv_cache [list][cache_idx+8]);
1030
                    h->ref_cache[list][cache_idx  ]=
1031
                    h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1032
                }
1033
            }
1034

    
1035
            if(USES_LIST(topleft_type, list)){
1036
                const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
1037
                const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (h->topleft_partition & h->b8_stride);
1038
                AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]);
1039
                h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1040
            }else{
1041
                AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]);
1042
                h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1043
            }
1044

    
1045
            if(USES_LIST(topright_type, list)){
1046
                const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1047
                const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
1048
                AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]);
1049
                h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1050
            }else{
1051
                AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]);
1052
                h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1053
            }
1054

    
1055
            if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
1056
                continue;
1057

    
1058
            if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
1059
            h->ref_cache[list][scan8[5 ]+1] =
1060
            h->ref_cache[list][scan8[7 ]+1] =
1061
            h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
1062
            h->ref_cache[list][scan8[4 ]] =
1063
            h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1064
            AV_ZERO32(h->mv_cache [list][scan8[5 ]+1]);
1065
            AV_ZERO32(h->mv_cache [list][scan8[7 ]+1]);
1066
            AV_ZERO32(h->mv_cache [list][scan8[13]+1]); //FIXME remove past 3 (init somewhere else)
1067
            AV_ZERO32(h->mv_cache [list][scan8[4 ]]);
1068
            AV_ZERO32(h->mv_cache [list][scan8[12]]);
1069

    
1070
            if( CABAC ) {
1071
                /* XXX beurk, Load mvd */
1072
                if(USES_LIST(top_type, list)){
1073
                    const int b_xy= h->mb2br_xy[top_xy];
1074
                    AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1075
                }else{
1076
                    AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1077
                }
1078
                if(USES_LIST(left_type[0], list)){
1079
                    const int b_xy= h->mb2br_xy[left_xy[0]] + 6;
1080
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy - left_block[0]]);
1081
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy - left_block[1]]);
1082
                }else{
1083
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]);
1084
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]);
1085
                }
1086
                if(USES_LIST(left_type[1], list)){
1087
                    const int b_xy= h->mb2br_xy[left_xy[1]] + 6;
1088
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy - left_block[2]]);
1089
                    AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy - left_block[3]]);
1090
                }else{
1091
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]);
1092
                    AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]);
1093
                }
1094
                AV_ZERO16(h->mvd_cache [list][scan8[5 ]+1]);
1095
                AV_ZERO16(h->mvd_cache [list][scan8[7 ]+1]);
1096
                AV_ZERO16(h->mvd_cache [list][scan8[13]+1]); //FIXME remove past 3 (init somewhere else)
1097
                AV_ZERO16(h->mvd_cache [list][scan8[4 ]]);
1098
                AV_ZERO16(h->mvd_cache [list][scan8[12]]);
1099
                if(h->slice_type_nos == FF_B_TYPE){
1100
                    fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
1101

    
1102
                    if(IS_DIRECT(top_type)){
1103
                        AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_DIRECT2>>1));
1104
                    }else if(IS_8X8(top_type)){
1105
                        int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
1106
                        h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
1107
                        h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
1108
                    }else{
1109
                        AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1));
1110
                    }
1111

    
1112
                    if(IS_DIRECT(left_type[0]))
1113
                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
1114
                    else if(IS_8X8(left_type[0]))
1115
                        h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
1116
                    else
1117
                        h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
1118

    
1119
                    if(IS_DIRECT(left_type[1]))
1120
                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
1121
                    else if(IS_8X8(left_type[1]))
1122
                        h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
1123
                    else
1124
                        h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
1125
                }
1126
            }
1127
            }
1128
            if(FRAME_MBAFF){
1129
#define MAP_MVS\
1130
                    MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1131
                    MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1132
                    MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1133
                    MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1134
                    MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1135
                    MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1136
                    MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1137
                    MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1138
                    MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1139
                    MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1140
                if(MB_FIELD){
1141
#define MAP_F2F(idx, mb_type)\
1142
                    if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1143
                        h->ref_cache[list][idx] <<= 1;\
1144
                        h->mv_cache[list][idx][1] /= 2;\
1145
                        h->mvd_cache[list][idx][1] >>=1;\
1146
                    }
1147
                    MAP_MVS
1148
#undef MAP_F2F
1149
                }else{
1150
#define MAP_F2F(idx, mb_type)\
1151
                    if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1152
                        h->ref_cache[list][idx] >>= 1;\
1153
                        h->mv_cache[list][idx][1] <<= 1;\
1154
                        h->mvd_cache[list][idx][1] <<= 1;\
1155
                    }
1156
                    MAP_MVS
1157
#undef MAP_F2F
1158
                }
1159
            }
1160
        }
1161
    }
1162
#endif
1163

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

    
1167
/**
1168
 *
1169
 * @returns non zero if the loop filter can be skiped
1170
 */
1171
static int fill_filter_caches(H264Context *h, int mb_type){
1172
    MpegEncContext * const s = &h->s;
1173
    const int mb_xy= h->mb_xy;
1174
    int top_xy, left_xy[2];
1175
    int top_type, left_type[2];
1176

    
1177
    top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
1178

    
1179
    //FIXME deblocking could skip the intra and nnz parts.
1180

    
1181
    /* Wow, what a mess, why didn't they simplify the interlacing & intra
1182
     * stuff, I can't imagine that these complex rules are worth it. */
1183

    
1184
    left_xy[1] = left_xy[0] = mb_xy-1;
1185
    if(FRAME_MBAFF){
1186
        const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
1187
        const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
1188
        if(s->mb_y&1){
1189
            if (left_mb_field_flag != curr_mb_field_flag) {
1190
                left_xy[0] -= s->mb_stride;
1191
            }
1192
        }else{
1193
            if(curr_mb_field_flag){
1194
                top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
1195
            }
1196
            if (left_mb_field_flag != curr_mb_field_flag) {
1197
                left_xy[1] += s->mb_stride;
1198
            }
1199
        }
1200
    }
1201

    
1202
    h->top_mb_xy = top_xy;
1203
    h->left_mb_xy[0] = left_xy[0];
1204
    h->left_mb_xy[1] = left_xy[1];
1205
    {
1206
        //for sufficiently low qp, filtering wouldn't do anything
1207
        //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
1208
        int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
1209
        int qp = s->current_picture.qscale_table[mb_xy];
1210
        if(qp <= qp_thresh
1211
           && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
1212
           && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
1213
            if(!FRAME_MBAFF)
1214
                return 1;
1215
            if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
1216
               && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
1217
                return 1;
1218
        }
1219
    }
1220

    
1221
    if(h->deblocking_filter == 2){
1222
        h->top_type    = top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
1223
        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;
1224
        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;
1225
    }else{
1226
        h->top_type    = top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
1227
        h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
1228
        h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
1229
    }
1230
    if(IS_INTRA(mb_type))
1231
        return 0;
1232

    
1233
    AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
1234
    AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
1235
    AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
1236
    AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
1237
    AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
1238

    
1239
    h->cbp= h->cbp_table[mb_xy];
1240

    
1241
    {
1242
        int list;
1243
        for(list=0; list<h->list_count; list++){
1244
            int8_t *ref;
1245
            int y, b_stride;
1246
            int16_t (*mv_dst)[2];
1247
            int16_t (*mv_src)[2];
1248

    
1249
            if(!USES_LIST(mb_type, list)){
1250
                fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
1251
                AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1252
                AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1253
                AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1254
                AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1255
                continue;
1256
            }
1257

    
1258
            ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
1259
            {
1260
                int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1261
                AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1262
                AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1263
                ref += h->b8_stride;
1264
                AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1265
                AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1266
            }
1267

    
1268
            b_stride = h->b_stride;
1269
            mv_dst   = &h->mv_cache[list][scan8[0]];
1270
            mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
1271
            for(y=0; y<4; y++){
1272
                AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
1273
            }
1274

    
1275
        }
1276
    }
1277

    
1278

    
1279
/*
1280
0 . T T. T T T T
1281
1 L . .L . . . .
1282
2 L . .L . . . .
1283
3 . T TL . . . .
1284
4 L . .L . . . .
1285
5 L . .. . . . .
1286
*/
1287
//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
1288
    if(top_type){
1289
        AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
1290
    }
1291

    
1292
    if(left_type[0]){
1293
        h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
1294
        h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
1295
        h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
1296
        h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
1297
    }
1298

    
1299
    // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
1300
    if(!CABAC && h->pps.transform_8x8_mode){
1301
        if(IS_8x8DCT(top_type)){
1302
            h->non_zero_count_cache[4+8*0]=
1303
            h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
1304
            h->non_zero_count_cache[6+8*0]=
1305
            h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
1306
        }
1307
        if(IS_8x8DCT(left_type[0])){
1308
            h->non_zero_count_cache[3+8*1]=
1309
            h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
1310
        }
1311
        if(IS_8x8DCT(left_type[1])){
1312
            h->non_zero_count_cache[3+8*3]=
1313
            h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
1314
        }
1315

    
1316
        if(IS_8x8DCT(mb_type)){
1317
            h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
1318
            h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
1319

    
1320
            h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
1321
            h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
1322

    
1323
            h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
1324
            h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
1325

    
1326
            h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
1327
            h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
1328
        }
1329
    }
1330

    
1331
    if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
1332
        int list;
1333
        for(list=0; list<h->list_count; list++){
1334
            if(USES_LIST(top_type, list)){
1335
                const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1336
                const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
1337
                int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1338
                AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1339
                h->ref_cache[list][scan8[0] + 0 - 1*8]=
1340
                h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
1341
                h->ref_cache[list][scan8[0] + 2 - 1*8]=
1342
                h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
1343
            }else{
1344
                AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1345
                AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1346
            }
1347

    
1348
            if(!IS_INTERLACED(mb_type^left_type[0])){
1349
                if(USES_LIST(left_type[0], list)){
1350
                    const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1351
                    const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
1352
                    int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1353
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
1354
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
1355
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
1356
                    AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
1357
                    h->ref_cache[list][scan8[0] - 1 + 0 ]=
1358
                    h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + h->b8_stride*0]];
1359
                    h->ref_cache[list][scan8[0] - 1 +16 ]=
1360
                    h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + h->b8_stride*1]];
1361
                }else{
1362
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
1363
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
1364
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
1365
                    AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
1366
                    h->ref_cache[list][scan8[0] - 1 + 0  ]=
1367
                    h->ref_cache[list][scan8[0] - 1 + 8  ]=
1368
                    h->ref_cache[list][scan8[0] - 1 + 16 ]=
1369
                    h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
1370
                }
1371
            }
1372
        }
1373
    }
1374

    
1375
    return 0;
1376
}
1377

    
1378
/**
1379
 * gets the predicted intra4x4 prediction mode.
1380
 */
1381
static inline int pred_intra_mode(H264Context *h, int n){
1382
    const int index8= scan8[n];
1383
    const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1384
    const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1385
    const int min= FFMIN(left, top);
1386

    
1387
    tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1388

    
1389
    if(min<0) return DC_PRED;
1390
    else      return min;
1391
}
1392

    
1393
static inline void write_back_non_zero_count(H264Context *h){
1394
    const int mb_xy= h->mb_xy;
1395

    
1396
    AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1397
    AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1398
    AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[0+8*5]);
1399
    AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8*3]);
1400
    AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1401
}
1402

    
1403
static inline void write_back_motion(H264Context *h, int mb_type){
1404
    MpegEncContext * const s = &h->s;
1405
    const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1406
    const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1407
    int list;
1408

    
1409
    if(!USES_LIST(mb_type, 0))
1410
        fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1411

    
1412
    for(list=0; list<h->list_count; list++){
1413
        int y, b_stride;
1414
        int16_t (*mv_dst)[2];
1415
        int16_t (*mv_src)[2];
1416

    
1417
        if(!USES_LIST(mb_type, list))
1418
            continue;
1419

    
1420
        b_stride = h->b_stride;
1421
        mv_dst   = &s->current_picture.motion_val[list][b_xy];
1422
        mv_src   = &h->mv_cache[list][scan8[0]];
1423
        for(y=0; y<4; y++){
1424
            AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1425
        }
1426
        if( CABAC ) {
1427
            uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
1428
            uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1429
            if(IS_SKIP(mb_type))
1430
                AV_ZERO128(mvd_dst);
1431
            else{
1432
            AV_COPY64(mvd_dst, mvd_src + 8*3);
1433
            for(y=0; y<3; y++){
1434
                AV_COPY16(mvd_dst + 3 + 3 - y, mvd_src + 3 + 8*y);
1435
            }
1436
            }
1437
        }
1438

    
1439
        {
1440
            int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1441
            ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1442
            ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1443
            ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1444
            ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1445
        }
1446
    }
1447

    
1448
    if(h->slice_type_nos == FF_B_TYPE && CABAC){
1449
        if(IS_8X8(mb_type)){
1450
            uint8_t *direct_table = &h->direct_table[b8_xy];
1451
            direct_table[1+0*h->b8_stride] = h->sub_mb_type[1]>>1;
1452
            direct_table[0+1*h->b8_stride] = h->sub_mb_type[2]>>1;
1453
            direct_table[1+1*h->b8_stride] = h->sub_mb_type[3]>>1;
1454
        }
1455
    }
1456
}
1457

    
1458
static inline int get_dct8x8_allowed(H264Context *h){
1459
    if(h->sps.direct_8x8_inference_flag)
1460
        return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
1461
    else
1462
        return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1463
}
1464

    
1465
/**
1466
 * decodes a P_SKIP or B_SKIP macroblock
1467
 */
1468
static void decode_mb_skip(H264Context *h){
1469
    MpegEncContext * const s = &h->s;
1470
    const int mb_xy= h->mb_xy;
1471
    int mb_type=0;
1472

    
1473
    memset(h->non_zero_count[mb_xy], 0, 32);
1474
    memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1475

    
1476
    if(MB_FIELD)
1477
        mb_type|= MB_TYPE_INTERLACED;
1478

    
1479
    if( h->slice_type_nos == FF_B_TYPE )
1480
    {
1481
        // just for fill_caches. pred_direct_motion will set the real mb_type
1482
        mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1483
        if(h->direct_spatial_mv_pred){
1484
            fill_decode_neighbors(h, mb_type);
1485
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1486
        }
1487
        ff_h264_pred_direct_motion(h, &mb_type);
1488
        mb_type|= MB_TYPE_SKIP;
1489
    }
1490
    else
1491
    {
1492
        int mx, my;
1493
        mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1494

    
1495
        fill_decode_neighbors(h, mb_type);
1496
        fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1497
        pred_pskip_motion(h, &mx, &my);
1498
        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1499
        fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1500
    }
1501

    
1502
    write_back_motion(h, mb_type);
1503
    s->current_picture.mb_type[mb_xy]= mb_type;
1504
    s->current_picture.qscale_table[mb_xy]= s->qscale;
1505
    h->slice_table[ mb_xy ]= h->slice_num;
1506
    h->prev_mb_skipped= 1;
1507
}
1508

    
1509
#include "h264_mvpred.h" //For pred_pskip_motion()
1510

    
1511
#endif /* AVCODEC_H264_H */