Rearchitecturing the stiched up goose part 1
[ffmpeg.git] / libavcodec / h264.h
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 "dsputil.h"
32 #include "cabac.h"
33 #include "mpegvideo.h"
34 #include "h264pred.h"
35 #include "rectangle.h"
36
37 #define interlaced_dct interlaced_dct_is_a_bad_name
38 #define mb_intra mb_intra_is_not_initialized_see_mb_type
39
40 #define LUMA_DC_BLOCK_INDEX   25
41 #define CHROMA_DC_BLOCK_INDEX 26
42
43 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
44 #define COEFF_TOKEN_VLC_BITS           8
45 #define TOTAL_ZEROS_VLC_BITS           9
46 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
47 #define RUN_VLC_BITS                   3
48 #define RUN7_VLC_BITS                  6
49
50 #define MAX_SPS_COUNT 32
51 #define MAX_PPS_COUNT 256
52
53 #define MAX_MMCO_COUNT 66
54
55 #define MAX_DELAYED_PIC_COUNT 16
56
57 /* Compiling in interlaced support reduces the speed
58  * of progressive decoding by about 2%. */
59 #define ALLOW_INTERLACE
60
61 #define ALLOW_NOCHROMA
62
63 /**
64  * The maximum number of slices supported by the decoder.
65  * must be a power of 2
66  */
67 #define MAX_SLICES 16
68
69 #ifdef ALLOW_INTERLACE
70 #define MB_MBAFF h->mb_mbaff
71 #define MB_FIELD h->mb_field_decoding_flag
72 #define FRAME_MBAFF h->mb_aff_frame
73 #define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
74 #else
75 #define MB_MBAFF 0
76 #define MB_FIELD 0
77 #define FRAME_MBAFF 0
78 #define FIELD_PICTURE 0
79 #undef  IS_INTERLACED
80 #define IS_INTERLACED(mb_type) 0
81 #endif
82 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
83
84 #ifdef ALLOW_NOCHROMA
85 #define CHROMA h->sps.chroma_format_idc
86 #else
87 #define CHROMA 1
88 #endif
89
90 #ifndef CABAC
91 #define CABAC h->pps.cabac
92 #endif
93
94 #define EXTENDED_SAR          255
95
96 #define MB_TYPE_REF0       MB_TYPE_ACPRED //dirty but it fits in 16 bit
97 #define MB_TYPE_8x8DCT     0x01000000
98 #define IS_REF0(a)         ((a) & MB_TYPE_REF0)
99 #define IS_8x8DCT(a)       ((a) & MB_TYPE_8x8DCT)
100
101 /**
102  * Value of Picture.reference when Picture is not a reference picture, but
103  * is held for delayed output.
104  */
105 #define DELAYED_PIC_REF 4
106
107
108 /* NAL unit types */
109 enum {
110     NAL_SLICE=1,
111     NAL_DPA,
112     NAL_DPB,
113     NAL_DPC,
114     NAL_IDR_SLICE,
115     NAL_SEI,
116     NAL_SPS,
117     NAL_PPS,
118     NAL_AUD,
119     NAL_END_SEQUENCE,
120     NAL_END_STREAM,
121     NAL_FILLER_DATA,
122     NAL_SPS_EXT,
123     NAL_AUXILIARY_SLICE=19
124 };
125
126 /**
127  * SEI message types
128  */
129 typedef enum {
130     SEI_BUFFERING_PERIOD             =  0, ///< buffering period (H.264, D.1.1)
131     SEI_TYPE_PIC_TIMING              =  1, ///< picture timing
132     SEI_TYPE_USER_DATA_UNREGISTERED  =  5, ///< unregistered user data
133     SEI_TYPE_RECOVERY_POINT          =  6  ///< recovery point (frame # to decoder sync)
134 } SEI_Type;
135
136 /**
137  * pic_struct in picture timing SEI message
138  */
139 typedef enum {
140     SEI_PIC_STRUCT_FRAME             = 0, ///<  0: %frame
141     SEI_PIC_STRUCT_TOP_FIELD         = 1, ///<  1: top field
142     SEI_PIC_STRUCT_BOTTOM_FIELD      = 2, ///<  2: bottom field
143     SEI_PIC_STRUCT_TOP_BOTTOM        = 3, ///<  3: top field, bottom field, in that order
144     SEI_PIC_STRUCT_BOTTOM_TOP        = 4, ///<  4: bottom field, top field, in that order
145     SEI_PIC_STRUCT_TOP_BOTTOM_TOP    = 5, ///<  5: top field, bottom field, top field repeated, in that order
146     SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///<  6: bottom field, top field, bottom field repeated, in that order
147     SEI_PIC_STRUCT_FRAME_DOUBLING    = 7, ///<  7: %frame doubling
148     SEI_PIC_STRUCT_FRAME_TRIPLING    = 8  ///<  8: %frame tripling
149 } SEI_PicStructType;
150
151 /**
152  * Sequence parameter set
153  */
154 typedef struct SPS{
155
156     int profile_idc;
157     int level_idc;
158     int chroma_format_idc;
159     int transform_bypass;              ///< qpprime_y_zero_transform_bypass_flag
160     int log2_max_frame_num;            ///< log2_max_frame_num_minus4 + 4
161     int poc_type;                      ///< pic_order_cnt_type
162     int log2_max_poc_lsb;              ///< log2_max_pic_order_cnt_lsb_minus4
163     int delta_pic_order_always_zero_flag;
164     int offset_for_non_ref_pic;
165     int offset_for_top_to_bottom_field;
166     int poc_cycle_length;              ///< num_ref_frames_in_pic_order_cnt_cycle
167     int ref_frame_count;               ///< num_ref_frames
168     int gaps_in_frame_num_allowed_flag;
169     int mb_width;                      ///< pic_width_in_mbs_minus1 + 1
170     int mb_height;                     ///< pic_height_in_map_units_minus1 + 1
171     int frame_mbs_only_flag;
172     int mb_aff;                        ///<mb_adaptive_frame_field_flag
173     int direct_8x8_inference_flag;
174     int crop;                   ///< frame_cropping_flag
175     unsigned int crop_left;            ///< frame_cropping_rect_left_offset
176     unsigned int crop_right;           ///< frame_cropping_rect_right_offset
177     unsigned int crop_top;             ///< frame_cropping_rect_top_offset
178     unsigned int crop_bottom;          ///< frame_cropping_rect_bottom_offset
179     int vui_parameters_present_flag;
180     AVRational sar;
181     int video_signal_type_present_flag;
182     int full_range;
183     int colour_description_present_flag;
184     enum AVColorPrimaries color_primaries;
185     enum AVColorTransferCharacteristic color_trc;
186     enum AVColorSpace colorspace;
187     int timing_info_present_flag;
188     uint32_t num_units_in_tick;
189     uint32_t time_scale;
190     int fixed_frame_rate_flag;
191     short offset_for_ref_frame[256]; //FIXME dyn aloc?
192     int bitstream_restriction_flag;
193     int num_reorder_frames;
194     int scaling_matrix_present;
195     uint8_t scaling_matrix4[6][16];
196     uint8_t scaling_matrix8[2][64];
197     int nal_hrd_parameters_present_flag;
198     int vcl_hrd_parameters_present_flag;
199     int pic_struct_present_flag;
200     int time_offset_length;
201     int cpb_cnt;                       ///< See H.264 E.1.2
202     int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1
203     int cpb_removal_delay_length;      ///< cpb_removal_delay_length_minus1 + 1
204     int dpb_output_delay_length;       ///< dpb_output_delay_length_minus1 + 1
205     int bit_depth_luma;                ///< bit_depth_luma_minus8 + 8
206     int bit_depth_chroma;              ///< bit_depth_chroma_minus8 + 8
207     int residual_color_transform_flag; ///< residual_colour_transform_flag
208 }SPS;
209
210 /**
211  * Picture parameter set
212  */
213 typedef struct PPS{
214     unsigned int sps_id;
215     int cabac;                  ///< entropy_coding_mode_flag
216     int pic_order_present;      ///< pic_order_present_flag
217     int slice_group_count;      ///< num_slice_groups_minus1 + 1
218     int mb_slice_group_map_type;
219     unsigned int ref_count[2];  ///< num_ref_idx_l0/1_active_minus1 + 1
220     int weighted_pred;          ///< weighted_pred_flag
221     int weighted_bipred_idc;
222     int init_qp;                ///< pic_init_qp_minus26 + 26
223     int init_qs;                ///< pic_init_qs_minus26 + 26
224     int chroma_qp_index_offset[2];
225     int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
226     int constrained_intra_pred; ///< constrained_intra_pred_flag
227     int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
228     int transform_8x8_mode;     ///< transform_8x8_mode_flag
229     uint8_t scaling_matrix4[6][16];
230     uint8_t scaling_matrix8[2][64];
231     uint8_t chroma_qp_table[2][64];  ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
232     int chroma_qp_diff;
233 }PPS;
234
235 /**
236  * Memory management control operation opcode.
237  */
238 typedef enum MMCOOpcode{
239     MMCO_END=0,
240     MMCO_SHORT2UNUSED,
241     MMCO_LONG2UNUSED,
242     MMCO_SHORT2LONG,
243     MMCO_SET_MAX_LONG,
244     MMCO_RESET,
245     MMCO_LONG,
246 } MMCOOpcode;
247
248 /**
249  * Memory management control operation.
250  */
251 typedef struct MMCO{
252     MMCOOpcode opcode;
253     int short_pic_num;  ///< pic_num without wrapping (pic_num & max_pic_num)
254     int long_arg;       ///< index, pic_num, or num long refs depending on opcode
255 } MMCO;
256
257 /**
258  * H264Context
259  */
260 typedef struct H264Context{
261     MpegEncContext s;
262     int nal_ref_idc;
263     int nal_unit_type;
264     uint8_t *rbsp_buffer[2];
265     unsigned int rbsp_buffer_size[2];
266
267     /**
268       * Used to parse AVC variant of h264
269       */
270     int is_avc; ///< this flag is != 0 if codec is avc1
271     int got_avcC; ///< flag used to parse avcC data only once
272     int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
273
274     int chroma_qp[2]; //QPc
275
276     int qp_thresh;      ///< QP threshold to skip loopfilter
277
278     int prev_mb_skipped;
279     int next_mb_skipped;
280
281     //prediction stuff
282     int chroma_pred_mode;
283     int intra16x16_pred_mode;
284
285     int top_mb_xy;
286     int left_mb_xy[2];
287
288     int8_t intra4x4_pred_mode_cache[5*8];
289     int8_t (*intra4x4_pred_mode)[8];
290     H264PredContext hpc;
291     unsigned int topleft_samples_available;
292     unsigned int top_samples_available;
293     unsigned int topright_samples_available;
294     unsigned int left_samples_available;
295     uint8_t (*top_borders[2])[16+2*8];
296     uint8_t left_border[2*(17+2*9)];
297
298     /**
299      * non zero coeff count cache.
300      * is 64 if not available.
301      */
302     DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache[6*8]);
303     uint8_t (*non_zero_count)[32];
304
305     /**
306      * Motion vector cache.
307      */
308     DECLARE_ALIGNED_8(int16_t, mv_cache[2][5*8][2]);
309     DECLARE_ALIGNED_8(int8_t, ref_cache[2][5*8]);
310 #define LIST_NOT_USED -1 //FIXME rename?
311 #define PART_NOT_AVAILABLE -2
312
313     /**
314      * is 1 if the specific list MV&references are set to 0,0,-2.
315      */
316     int mv_cache_clean[2];
317
318     /**
319      * number of neighbors (top and/or left) that used 8x8 dct
320      */
321     int neighbor_transform_size;
322
323     /**
324      * block_offset[ 0..23] for frame macroblocks
325      * block_offset[24..47] for field macroblocks
326      */
327     int block_offset[2*(16+8)];
328
329     uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
330     uint32_t *mb2b8_xy;
331     int b_stride; //FIXME use s->b4_stride
332     int b8_stride;
333
334     int mb_linesize;   ///< may be equal to s->linesize or s->linesize*2, for mbaff
335     int mb_uvlinesize;
336
337     int emu_edge_width;
338     int emu_edge_height;
339
340     int halfpel_flag;
341     int thirdpel_flag;
342
343     int unknown_svq3_flag;
344     int next_slice_index;
345
346     SPS *sps_buffers[MAX_SPS_COUNT];
347     SPS sps; ///< current sps
348
349     PPS *pps_buffers[MAX_PPS_COUNT];
350     /**
351      * current pps
352      */
353     PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
354
355     uint32_t dequant4_buffer[6][52][16];
356     uint32_t dequant8_buffer[2][52][64];
357     uint32_t (*dequant4_coeff[6])[16];
358     uint32_t (*dequant8_coeff[2])[64];
359     int dequant_coeff_pps;     ///< reinit tables when pps changes
360
361     int slice_num;
362     uint16_t *slice_table_base;
363     uint16_t *slice_table;     ///< slice_table_base + 2*mb_stride + 1
364     int slice_type;
365     int slice_type_nos;        ///< S free slice type (SI/SP are remapped to I/P)
366     int slice_type_fixed;
367
368     //interlacing specific flags
369     int mb_aff_frame;
370     int mb_field_decoding_flag;
371     int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag
372
373     DECLARE_ALIGNED_8(uint16_t, sub_mb_type[4]);
374
375     //POC stuff
376     int poc_lsb;
377     int poc_msb;
378     int delta_poc_bottom;
379     int delta_poc[2];
380     int frame_num;
381     int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
382     int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
383     int frame_num_offset;         ///< for POC type 2
384     int prev_frame_num_offset;    ///< for POC type 2
385     int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
386
387     /**
388      * frame_num for frames or 2*frame_num+1 for field pics.
389      */
390     int curr_pic_num;
391
392     /**
393      * max_frame_num or 2*max_frame_num for field pics.
394      */
395     int max_pic_num;
396
397     //Weighted pred stuff
398     int use_weight;
399     int use_weight_chroma;
400     int luma_log2_weight_denom;
401     int chroma_log2_weight_denom;
402     int luma_weight[2][48];
403     int luma_offset[2][48];
404     int chroma_weight[2][48][2];
405     int chroma_offset[2][48][2];
406     int implicit_weight[48][48];
407
408     //deblock
409     int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
410     int slice_alpha_c0_offset;
411     int slice_beta_offset;
412
413     int redundant_pic_count;
414
415     int direct_spatial_mv_pred;
416     int dist_scale_factor[16];
417     int dist_scale_factor_field[2][32];
418     int map_col_to_list0[2][16+32];
419     int map_col_to_list0_field[2][2][16+32];
420
421     /**
422      * num_ref_idx_l0/1_active_minus1 + 1
423      */
424     unsigned int ref_count[2];   ///< counts frames or fields, depending on current mb mode
425     unsigned int list_count;
426     uint8_t *list_counts;            ///< Array of list_count per MB specifying the slice type
427     Picture *short_ref[32];
428     Picture *long_ref[32];
429     Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
430     Picture ref_list[2][48];         /**< 0..15: frame refs, 16..47: mbaff field refs.
431                                           Reordered version of default_ref_list
432                                           according to picture reordering in slice header */
433     int ref2frm[MAX_SLICES][2][64];  ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
434     Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
435     int outputed_poc;
436
437     /**
438      * memory management control operations buffer.
439      */
440     MMCO mmco[MAX_MMCO_COUNT];
441     int mmco_index;
442
443     int long_ref_count;  ///< number of actual long term references
444     int short_ref_count; ///< number of actual short term references
445
446     //data partitioning
447     GetBitContext intra_gb;
448     GetBitContext inter_gb;
449     GetBitContext *intra_gb_ptr;
450     GetBitContext *inter_gb_ptr;
451
452     DECLARE_ALIGNED_16(DCTELEM, mb[16*24]);
453     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
454
455     /**
456      * Cabac
457      */
458     CABACContext cabac;
459     uint8_t      cabac_state[460];
460     int          cabac_init_idc;
461
462     /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
463     uint16_t     *cbp_table;
464     int cbp;
465     int top_cbp;
466     int left_cbp;
467     /* chroma_pred_mode for i4x4 or i16x16, else 0 */
468     uint8_t     *chroma_pred_mode_table;
469     int         last_qscale_diff;
470     int16_t     (*mvd_table[2])[2];
471     DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]);
472     uint8_t     *direct_table;
473     uint8_t     direct_cache[5*8];
474
475     uint8_t zigzag_scan[16];
476     uint8_t zigzag_scan8x8[64];
477     uint8_t zigzag_scan8x8_cavlc[64];
478     uint8_t field_scan[16];
479     uint8_t field_scan8x8[64];
480     uint8_t field_scan8x8_cavlc[64];
481     const uint8_t *zigzag_scan_q0;
482     const uint8_t *zigzag_scan8x8_q0;
483     const uint8_t *zigzag_scan8x8_cavlc_q0;
484     const uint8_t *field_scan_q0;
485     const uint8_t *field_scan8x8_q0;
486     const uint8_t *field_scan8x8_cavlc_q0;
487
488     int x264_build;
489
490     /**
491      * @defgroup multithreading Members for slice based multithreading
492      * @{
493      */
494     struct H264Context *thread_context[MAX_THREADS];
495
496     /**
497      * current slice number, used to initalize slice_num of each thread/context
498      */
499     int current_slice;
500
501     /**
502      * Max number of threads / contexts.
503      * This is equal to AVCodecContext.thread_count unless
504      * multithreaded decoding is impossible, in which case it is
505      * reduced to 1.
506      */
507     int max_contexts;
508
509     /**
510      *  1 if the single thread fallback warning has already been
511      *  displayed, 0 otherwise.
512      */
513     int single_decode_warning;
514
515     int last_slice_type;
516     /** @} */
517
518     int mb_xy;
519
520     uint32_t svq3_watermark_key;
521
522     /**
523      * pic_struct in picture timing SEI message
524      */
525     SEI_PicStructType sei_pic_struct;
526
527     /**
528      * Complement sei_pic_struct
529      * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
530      * However, soft telecined frames may have these values.
531      * This is used in an attempt to flag soft telecine progressive.
532      */
533     int prev_interlaced_frame;
534
535     /**
536      * Bit set of clock types for fields/frames in picture timing SEI message.
537      * For each found ct_type, appropriate bit is set (e.g., bit 1 for
538      * interlaced).
539      */
540     int sei_ct_type;
541
542     /**
543      * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
544      */
545     int sei_dpb_output_delay;
546
547     /**
548      * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
549      */
550     int sei_cpb_removal_delay;
551
552     /**
553      * recovery_frame_cnt from SEI message
554      *
555      * Set to -1 if no recovery point SEI message found or to number of frames
556      * before playback synchronizes. Frames having recovery point are key
557      * frames.
558      */
559     int sei_recovery_frame_cnt;
560
561     int is_complex;
562
563     int luma_weight_flag[2];   ///< 7.4.3.2 luma_weight_lX_flag
564     int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
565
566     // Timestamp stuff
567     int sei_buffering_period_present;  ///< Buffering period SEI flag
568     int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
569 }H264Context;
570
571
572 extern const uint8_t ff_h264_chroma_qp[52];
573
574
575 /**
576  * Decode SEI
577  */
578 int ff_h264_decode_sei(H264Context *h);
579
580 /**
581  * Decode SPS
582  */
583 int ff_h264_decode_seq_parameter_set(H264Context *h);
584
585 /**
586  * Decode PPS
587  */
588 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
589
590 /**
591  * Decodes a network abstraction layer unit.
592  * @param consumed is the number of bytes used as input
593  * @param length is the length of the array
594  * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
595  * @returns decoded bytes, might be src+1 if no escapes
596  */
597 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
598
599 /**
600  * identifies the exact end of the bitstream
601  * @return the length of the trailing, or 0 if damaged
602  */
603 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
604
605 /**
606  * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
607  */
608 av_cold void ff_h264_free_context(H264Context *h);
609
610 /**
611  * reconstructs bitstream slice_type.
612  */
613 int ff_h264_get_slice_type(H264Context *h);
614
615 /**
616  * allocates tables.
617  * needs width/height
618  */
619 int ff_h264_alloc_tables(H264Context *h);
620
621 /**
622  * fills the default_ref_list.
623  */
624 int ff_h264_fill_default_ref_list(H264Context *h);
625
626 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
627 void ff_h264_fill_mbaff_ref_list(H264Context *h);
628 void ff_h264_remove_all_refs(H264Context *h);
629
630 /**
631  * Executes the reference picture marking (memory management control operations).
632  */
633 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
634
635 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
636
637
638 /**
639  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
640  */
641 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
642
643 /**
644  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
645  */
646 int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
647
648 void ff_h264_write_back_intra_pred_mode(H264Context *h);
649 void ff_h264_hl_decode_mb(H264Context *h);
650 int ff_h264_frame_start(H264Context *h);
651 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
652 av_cold int ff_h264_decode_end(AVCodecContext *avctx);
653 av_cold void ff_h264_decode_init_vlc(void);
654
655 /**
656  * decodes a macroblock
657  * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
658  */
659 int ff_h264_decode_mb_cavlc(H264Context *h);
660
661 /**
662  * decodes a CABAC coded macroblock
663  * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
664  */
665 int ff_h264_decode_mb_cabac(H264Context *h);
666
667 void ff_h264_init_cabac_states(H264Context *h);
668
669 void ff_h264_direct_dist_scale_factor(H264Context * const h);
670 void ff_h264_direct_ref_list_init(H264Context * const h);
671 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
672
673 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);
674 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);
675
676 /**
677  * Reset SEI values at the beginning of the frame.
678  *
679  * @param h H.264 context.
680  */
681 void ff_h264_reset_sei(H264Context *h);
682
683
684 /*
685 o-o o-o
686  / / /
687 o-o o-o
688  ,---'
689 o-o o-o
690  / / /
691 o-o o-o
692 */
693 //This table must be here because scan8[constant] must be known at compiletime
694 static const uint8_t scan8[16 + 2*4]={
695  4+1*8, 5+1*8, 4+2*8, 5+2*8,
696  6+1*8, 7+1*8, 6+2*8, 7+2*8,
697  4+3*8, 5+3*8, 4+4*8, 5+4*8,
698  6+3*8, 7+3*8, 6+4*8, 7+4*8,
699  1+1*8, 2+1*8,
700  1+2*8, 2+2*8,
701  1+4*8, 2+4*8,
702  1+5*8, 2+5*8,
703 };
704
705 static av_always_inline uint32_t pack16to32(int a, int b){
706 #if HAVE_BIGENDIAN
707    return (b&0xFFFF) + (a<<16);
708 #else
709    return (a&0xFFFF) + (b<<16);
710 #endif
711 }
712
713 /**
714  * gets the chroma qp.
715  */
716 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
717     return h->pps.chroma_qp_table[t][qscale];
718 }
719
720 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
721
722 static void fill_caches(H264Context *h, int mb_type, int for_deblock){
723     MpegEncContext * const s = &h->s;
724     const int mb_xy= h->mb_xy;
725     int topleft_xy, top_xy, topright_xy, left_xy[2];
726     int topleft_type, top_type, topright_type, left_type[2];
727     const uint8_t * left_block;
728     int topleft_partition= -1;
729     int i;
730     static const uint8_t left_block_options[4][8]={
731         {0,1,2,3,7,10,8,11},
732         {2,2,3,3,8,11,8,11},
733         {0,0,1,1,7,10,7,10},
734         {0,2,0,2,7,10,7,10}
735     };
736
737     top_xy     = mb_xy  - (s->mb_stride << FIELD_PICTURE);
738
739     //FIXME deblocking could skip the intra and nnz parts.
740 //     if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
741 //         return;
742
743     /* Wow, what a mess, why didn't they simplify the interlacing & intra
744      * stuff, I can't imagine that these complex rules are worth it. */
745
746     topleft_xy = top_xy - 1;
747     topright_xy= top_xy + 1;
748     left_xy[1] = left_xy[0] = mb_xy-1;
749     left_block = left_block_options[0];
750     if(FRAME_MBAFF){
751         const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
752         const int top_pair_xy      = pair_xy     - s->mb_stride;
753         const int topleft_pair_xy  = top_pair_xy - 1;
754         const int topright_pair_xy = top_pair_xy + 1;
755         const int topleft_mb_field_flag  = IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
756         const int top_mb_field_flag      = IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
757         const int topright_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
758         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
759         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
760         const int bottom = (s->mb_y & 1);
761         tprintf(s->avctx, "fill_caches: curr_mb_field_flag:%d, left_mb_field_flag:%d, topleft_mb_field_flag:%d, top_mb_field_flag:%d, topright_mb_field_flag:%d\n", curr_mb_field_flag, left_mb_field_flag, topleft_mb_field_flag, top_mb_field_flag, topright_mb_field_flag);
762
763         if (curr_mb_field_flag && (bottom || top_mb_field_flag)){
764             top_xy -= s->mb_stride;
765         }
766         if (curr_mb_field_flag && (bottom || topleft_mb_field_flag)){
767             topleft_xy -= s->mb_stride;
768         } else if(bottom && !curr_mb_field_flag && left_mb_field_flag) {
769             topleft_xy += s->mb_stride;
770             // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
771             topleft_partition = 0;
772         }
773         if (curr_mb_field_flag && (bottom || topright_mb_field_flag)){
774             topright_xy -= s->mb_stride;
775         }
776         if (left_mb_field_flag != curr_mb_field_flag) {
777             left_xy[1] = left_xy[0] = pair_xy - 1;
778             if (curr_mb_field_flag) {
779                 left_xy[1] += s->mb_stride;
780                 left_block = left_block_options[3];
781             } else {
782                 left_block= left_block_options[2 - bottom];
783             }
784         }
785     }
786
787     h->top_mb_xy = top_xy;
788     h->left_mb_xy[0] = left_xy[0];
789     h->left_mb_xy[1] = left_xy[1];
790     if(for_deblock){
791         topleft_type = 0;
792         topright_type = 0;
793         top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
794         left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
795         left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
796
797         if(!IS_INTRA(mb_type)){
798             int list;
799             for(list=0; list<h->list_count; list++){
800                 int8_t *ref;
801                 int y, b_xy;
802                 if(!USES_LIST(mb_type, list)){
803                     fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
804                     *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
805                     *(uint32_t*)&h->ref_cache[list][scan8[ 2]] =
806                     *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
807                     *(uint32_t*)&h->ref_cache[list][scan8[10]] = ((LIST_NOT_USED)&0xFF)*0x01010101;
808                     continue;
809                 }
810
811                 ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
812                 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
813                 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
814                 ref += h->b8_stride;
815                 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
816                 *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
817
818                 b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
819                 for(y=0; y<4; y++){
820                     *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y]= *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride];
821                     *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y]= *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride];
822                 }
823
824             }
825         }
826     }else{
827         topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
828         top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
829         topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
830         left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
831         left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
832
833     if(IS_INTRA(mb_type)){
834         int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
835         h->topleft_samples_available=
836         h->top_samples_available=
837         h->left_samples_available= 0xFFFF;
838         h->topright_samples_available= 0xEEEA;
839
840         if(!(top_type & type_mask)){
841             h->topleft_samples_available= 0xB3FF;
842             h->top_samples_available= 0x33FF;
843             h->topright_samples_available= 0x26EA;
844         }
845         if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
846             if(IS_INTERLACED(mb_type)){
847                 if(!(left_type[0] & type_mask)){
848                     h->topleft_samples_available&= 0xDFFF;
849                     h->left_samples_available&= 0x5FFF;
850                 }
851                 if(!(left_type[1] & type_mask)){
852                     h->topleft_samples_available&= 0xFF5F;
853                     h->left_samples_available&= 0xFF5F;
854                 }
855             }else{
856                 int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
857                                 ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
858                 assert(left_xy[0] == left_xy[1]);
859                 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
860                     h->topleft_samples_available&= 0xDF5F;
861                     h->left_samples_available&= 0x5F5F;
862                 }
863             }
864         }else{
865             if(!(left_type[0] & type_mask)){
866                 h->topleft_samples_available&= 0xDF5F;
867                 h->left_samples_available&= 0x5F5F;
868             }
869         }
870
871         if(!(topleft_type & type_mask))
872             h->topleft_samples_available&= 0x7FFF;
873
874         if(!(topright_type & type_mask))
875             h->topright_samples_available&= 0xFBFF;
876
877         if(IS_INTRA4x4(mb_type)){
878             if(IS_INTRA4x4(top_type)){
879                 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
880                 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
881                 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
882                 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
883             }else{
884                 int pred;
885                 if(!(top_type & type_mask))
886                     pred= -1;
887                 else{
888                     pred= 2;
889                 }
890                 h->intra4x4_pred_mode_cache[4+8*0]=
891                 h->intra4x4_pred_mode_cache[5+8*0]=
892                 h->intra4x4_pred_mode_cache[6+8*0]=
893                 h->intra4x4_pred_mode_cache[7+8*0]= pred;
894             }
895             for(i=0; i<2; i++){
896                 if(IS_INTRA4x4(left_type[i])){
897                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
898                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
899                 }else{
900                     int pred;
901                     if(!(left_type[i] & type_mask))
902                         pred= -1;
903                     else{
904                         pred= 2;
905                     }
906                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
907                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
908                 }
909             }
910         }
911     }
912     }
913
914
915 /*
916 0 . T T. T T T T
917 1 L . .L . . . .
918 2 L . .L . . . .
919 3 . T TL . . . .
920 4 L . .L . . . .
921 5 L . .. . . . .
922 */
923 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
924     if(top_type){
925         h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
926         h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
927         h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
928         h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
929
930         h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
931         h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
932
933         h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
934         h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
935
936     }else{
937         h->non_zero_count_cache[4+8*0]=
938         h->non_zero_count_cache[5+8*0]=
939         h->non_zero_count_cache[6+8*0]=
940         h->non_zero_count_cache[7+8*0]=
941
942         h->non_zero_count_cache[1+8*0]=
943         h->non_zero_count_cache[2+8*0]=
944
945         h->non_zero_count_cache[1+8*3]=
946         h->non_zero_count_cache[2+8*3]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
947
948     }
949
950     for (i=0; i<2; i++) {
951         if(left_type[i]){
952             h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
953             h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
954             h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
955             h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
956         }else{
957             h->non_zero_count_cache[3+8*1 + 2*8*i]=
958             h->non_zero_count_cache[3+8*2 + 2*8*i]=
959             h->non_zero_count_cache[0+8*1 +   8*i]=
960             h->non_zero_count_cache[0+8*4 +   8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
961         }
962     }
963
964     if( CABAC ) {
965         // top_cbp
966         if(top_type) {
967             h->top_cbp = h->cbp_table[top_xy];
968         } else if(IS_INTRA(mb_type)) {
969             h->top_cbp = 0x1C0;
970         } else {
971             h->top_cbp = 0;
972         }
973         // left_cbp
974         if (left_type[0]) {
975             h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
976         } else if(IS_INTRA(mb_type)) {
977             h->left_cbp = 0x1C0;
978         } else {
979             h->left_cbp = 0;
980         }
981         if (left_type[0]) {
982             h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
983         }
984         if (left_type[1]) {
985             h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
986         }
987     }
988
989 #if 1
990     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
991         int list;
992         for(list=0; list<h->list_count; list++){
993             if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
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             h->mv_cache_clean[list]= 0;
1002
1003             if(USES_LIST(top_type, list)){
1004                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1005                 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
1006                 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
1007                 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
1008                 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
1009                 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
1010                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
1011                 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
1012                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
1013                 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
1014             }else{
1015                 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
1016                 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
1017                 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
1018                 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
1019                 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
1020             }
1021
1022             for(i=0; i<2; i++){
1023                 int cache_idx = scan8[0] - 1 + i*2*8;
1024                 if(USES_LIST(left_type[i], list)){
1025                     const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1026                     const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
1027                     *(uint32_t*)h->mv_cache[list][cache_idx  ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]];
1028                     *(uint32_t*)h->mv_cache[list][cache_idx+8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]];
1029                     h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
1030                     h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
1031                 }else{
1032                     *(uint32_t*)h->mv_cache [list][cache_idx  ]=
1033                     *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
1034                     h->ref_cache[list][cache_idx  ]=
1035                     h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1036                 }
1037             }
1038
1039             if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
1040                 continue;
1041
1042             if(USES_LIST(topleft_type, list)){
1043                 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
1044                 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
1045                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
1046                 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1047             }else{
1048                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
1049                 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1050             }
1051
1052             if(USES_LIST(topright_type, list)){
1053                 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1054                 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
1055                 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
1056                 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1057             }else{
1058                 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
1059                 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1060             }
1061
1062             if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
1063                 continue;
1064
1065             h->ref_cache[list][scan8[5 ]+1] =
1066             h->ref_cache[list][scan8[7 ]+1] =
1067             h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
1068             h->ref_cache[list][scan8[4 ]] =
1069             h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1070             *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
1071             *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
1072             *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
1073             *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
1074             *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
1075
1076             if( CABAC ) {
1077                 /* XXX beurk, Load mvd */
1078                 if(USES_LIST(top_type, list)){
1079                     const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1080                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
1081                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
1082                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
1083                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
1084                 }else{
1085                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
1086                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
1087                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
1088                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
1089                 }
1090                 if(USES_LIST(left_type[0], list)){
1091                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1092                     *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
1093                     *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
1094                 }else{
1095                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
1096                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
1097                 }
1098                 if(USES_LIST(left_type[1], list)){
1099                     const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
1100                     *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
1101                     *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
1102                 }else{
1103                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
1104                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
1105                 }
1106                 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
1107                 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
1108                 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
1109                 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
1110                 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
1111
1112                 if(h->slice_type_nos == FF_B_TYPE){
1113                     fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
1114
1115                     if(IS_DIRECT(top_type)){
1116                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
1117                     }else if(IS_8X8(top_type)){
1118                         int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
1119                         h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
1120                         h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
1121                     }else{
1122                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
1123                     }
1124
1125                     if(IS_DIRECT(left_type[0]))
1126                         h->direct_cache[scan8[0] - 1 + 0*8]= 1;
1127                     else if(IS_8X8(left_type[0]))
1128                         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)];
1129                     else
1130                         h->direct_cache[scan8[0] - 1 + 0*8]= 0;
1131
1132                     if(IS_DIRECT(left_type[1]))
1133                         h->direct_cache[scan8[0] - 1 + 2*8]= 1;
1134                     else if(IS_8X8(left_type[1]))
1135                         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)];
1136                     else
1137                         h->direct_cache[scan8[0] - 1 + 2*8]= 0;
1138                 }
1139             }
1140
1141             if(FRAME_MBAFF){
1142 #define MAP_MVS\
1143                     MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1144                     MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1145                     MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1146                     MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1147                     MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1148                     MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1149                     MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1150                     MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1151                     MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1152                     MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1153                 if(MB_FIELD){
1154 #define MAP_F2F(idx, mb_type)\
1155                     if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1156                         h->ref_cache[list][idx] <<= 1;\
1157                         h->mv_cache[list][idx][1] /= 2;\
1158                         h->mvd_cache[list][idx][1] /= 2;\
1159                     }
1160                     MAP_MVS
1161 #undef MAP_F2F
1162                 }else{
1163 #define MAP_F2F(idx, mb_type)\
1164                     if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1165                         h->ref_cache[list][idx] >>= 1;\
1166                         h->mv_cache[list][idx][1] <<= 1;\
1167                         h->mvd_cache[list][idx][1] <<= 1;\
1168                     }
1169                     MAP_MVS
1170 #undef MAP_F2F
1171                 }
1172             }
1173         }
1174     }
1175 #endif
1176
1177     h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1178 }
1179
1180 /**
1181  * gets the predicted intra4x4 prediction mode.
1182  */
1183 static inline int pred_intra_mode(H264Context *h, int n){
1184     const int index8= scan8[n];
1185     const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1186     const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1187     const int min= FFMIN(left, top);
1188
1189     tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1190
1191     if(min<0) return DC_PRED;
1192     else      return min;
1193 }
1194
1195 static inline void write_back_non_zero_count(H264Context *h){
1196     const int mb_xy= h->mb_xy;
1197
1198     h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
1199     h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
1200     h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
1201     h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
1202     h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
1203     h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
1204     h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
1205
1206     h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
1207     h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
1208     h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
1209
1210     h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
1211     h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
1212     h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
1213
1214     //FIXME sort better how things are stored in non_zero_count
1215
1216
1217     h->non_zero_count[mb_xy][13]= h->non_zero_count_cache[6+8*1];
1218     h->non_zero_count[mb_xy][14]= h->non_zero_count_cache[6+8*2];
1219     h->non_zero_count[mb_xy][15]= h->non_zero_count_cache[6+8*3];
1220     h->non_zero_count[mb_xy][16]= h->non_zero_count_cache[5+8*1];
1221     h->non_zero_count[mb_xy][17]= h->non_zero_count_cache[5+8*2];
1222     h->non_zero_count[mb_xy][18]= h->non_zero_count_cache[5+8*3];
1223     h->non_zero_count[mb_xy][19]= h->non_zero_count_cache[4+8*1];
1224     h->non_zero_count[mb_xy][20]= h->non_zero_count_cache[4+8*2];
1225     h->non_zero_count[mb_xy][21]= h->non_zero_count_cache[4+8*3];
1226
1227     h->non_zero_count[mb_xy][22]= h->non_zero_count_cache[1+8*1];
1228     h->non_zero_count[mb_xy][23]= h->non_zero_count_cache[1+8*4];
1229
1230 }
1231
1232 static inline void write_back_motion(H264Context *h, int mb_type){
1233     MpegEncContext * const s = &h->s;
1234     const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1235     const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1236     int list;
1237
1238     if(!USES_LIST(mb_type, 0))
1239         fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1240
1241     for(list=0; list<h->list_count; list++){
1242         int y;
1243         if(!USES_LIST(mb_type, list))
1244             continue;
1245
1246         for(y=0; y<4; y++){
1247             *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y];
1248             *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];
1249         }
1250         if( CABAC ) {
1251             if(IS_SKIP(mb_type))
1252                 fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1253             else
1254             for(y=0; y<4; y++){
1255                 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1256                 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1257             }
1258         }
1259
1260         {
1261             int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1262             ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1263             ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1264             ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1265             ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1266         }
1267     }
1268
1269     if(h->slice_type_nos == FF_B_TYPE && CABAC){
1270         if(IS_8X8(mb_type)){
1271             uint8_t *direct_table = &h->direct_table[b8_xy];
1272             direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1273             direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1274             direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1275         }
1276     }
1277 }
1278
1279 static inline int get_dct8x8_allowed(H264Context *h){
1280     if(h->sps.direct_8x8_inference_flag)
1281         return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
1282     else
1283         return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1284 }
1285
1286 static void predict_field_decoding_flag(H264Context *h){
1287     MpegEncContext * const s = &h->s;
1288     const int mb_xy= h->mb_xy;
1289     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
1290                 ? s->current_picture.mb_type[mb_xy-1]
1291                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
1292                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
1293                 : 0;
1294     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
1295 }
1296
1297 /**
1298  * decodes a P_SKIP or B_SKIP macroblock
1299  */
1300 static void decode_mb_skip(H264Context *h){
1301     MpegEncContext * const s = &h->s;
1302     const int mb_xy= h->mb_xy;
1303     int mb_type=0;
1304
1305     memset(h->non_zero_count[mb_xy], 0, 32);
1306     memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1307
1308     if(MB_FIELD)
1309         mb_type|= MB_TYPE_INTERLACED;
1310
1311     if( h->slice_type_nos == FF_B_TYPE )
1312     {
1313         // just for fill_caches. pred_direct_motion will set the real mb_type
1314         mb_type|= MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1315
1316         fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
1317         ff_h264_pred_direct_motion(h, &mb_type);
1318         mb_type|= MB_TYPE_SKIP;
1319     }
1320     else
1321     {
1322         int mx, my;
1323         mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1324
1325         fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
1326         pred_pskip_motion(h, &mx, &my);
1327         fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1328         fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1329     }
1330
1331     write_back_motion(h, mb_type);
1332     s->current_picture.mb_type[mb_xy]= mb_type;
1333     s->current_picture.qscale_table[mb_xy]= s->qscale;
1334     h->slice_table[ mb_xy ]= h->slice_num;
1335     h->prev_mb_skipped= 1;
1336 }
1337
1338 #include "h264_mvpred.h" //For pred_pskip_motion()
1339
1340 #endif /* AVCODEC_H264_H */