e03e0e439c750a27ca0caff9c8ace9aeb4db2cf6
[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     int top_type;
289     int left_type[2];
290
291     int8_t intra4x4_pred_mode_cache[5*8];
292     int8_t (*intra4x4_pred_mode)[8];
293     H264PredContext hpc;
294     unsigned int topleft_samples_available;
295     unsigned int top_samples_available;
296     unsigned int topright_samples_available;
297     unsigned int left_samples_available;
298     uint8_t (*top_borders[2])[16+2*8];
299     uint8_t left_border[2*(17+2*9)];
300
301     /**
302      * non zero coeff count cache.
303      * is 64 if not available.
304      */
305     DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache)[6*8];
306
307     /*
308     .UU.YYYY
309     .UU.YYYY
310     .vv.YYYY
311     .VV.YYYY
312     */
313     uint8_t (*non_zero_count)[32];
314
315     /**
316      * Motion vector cache.
317      */
318     DECLARE_ALIGNED_16(int16_t, mv_cache)[2][5*8][2];
319     DECLARE_ALIGNED_8(int8_t, ref_cache)[2][5*8];
320 #define LIST_NOT_USED -1 //FIXME rename?
321 #define PART_NOT_AVAILABLE -2
322
323     /**
324      * is 1 if the specific list MV&references are set to 0,0,-2.
325      */
326     int mv_cache_clean[2];
327
328     /**
329      * number of neighbors (top and/or left) that used 8x8 dct
330      */
331     int neighbor_transform_size;
332
333     /**
334      * block_offset[ 0..23] for frame macroblocks
335      * block_offset[24..47] for field macroblocks
336      */
337     int block_offset[2*(16+8)];
338
339     uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
340     uint32_t *mb2b8_xy;
341     int b_stride; //FIXME use s->b4_stride
342     int b8_stride;
343
344     int mb_linesize;   ///< may be equal to s->linesize or s->linesize*2, for mbaff
345     int mb_uvlinesize;
346
347     int emu_edge_width;
348     int emu_edge_height;
349
350     int halfpel_flag;
351     int thirdpel_flag;
352
353     int unknown_svq3_flag;
354     int next_slice_index;
355
356     SPS *sps_buffers[MAX_SPS_COUNT];
357     SPS sps; ///< current sps
358
359     PPS *pps_buffers[MAX_PPS_COUNT];
360     /**
361      * current pps
362      */
363     PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
364
365     uint32_t dequant4_buffer[6][52][16];
366     uint32_t dequant8_buffer[2][52][64];
367     uint32_t (*dequant4_coeff[6])[16];
368     uint32_t (*dequant8_coeff[2])[64];
369     int dequant_coeff_pps;     ///< reinit tables when pps changes
370
371     int slice_num;
372     uint16_t *slice_table_base;
373     uint16_t *slice_table;     ///< slice_table_base + 2*mb_stride + 1
374     int slice_type;
375     int slice_type_nos;        ///< S free slice type (SI/SP are remapped to I/P)
376     int slice_type_fixed;
377
378     //interlacing specific flags
379     int mb_aff_frame;
380     int mb_field_decoding_flag;
381     int mb_mbaff;              ///< mb_aff_frame && mb_field_decoding_flag
382
383     DECLARE_ALIGNED_8(uint16_t, sub_mb_type)[4];
384
385     //POC stuff
386     int poc_lsb;
387     int poc_msb;
388     int delta_poc_bottom;
389     int delta_poc[2];
390     int frame_num;
391     int prev_poc_msb;             ///< poc_msb of the last reference pic for POC type 0
392     int prev_poc_lsb;             ///< poc_lsb of the last reference pic for POC type 0
393     int frame_num_offset;         ///< for POC type 2
394     int prev_frame_num_offset;    ///< for POC type 2
395     int prev_frame_num;           ///< frame_num of the last pic for POC type 1/2
396
397     /**
398      * frame_num for frames or 2*frame_num+1 for field pics.
399      */
400     int curr_pic_num;
401
402     /**
403      * max_frame_num or 2*max_frame_num for field pics.
404      */
405     int max_pic_num;
406
407     //Weighted pred stuff
408     int use_weight;
409     int use_weight_chroma;
410     int luma_log2_weight_denom;
411     int chroma_log2_weight_denom;
412     int luma_weight[2][48];
413     int luma_offset[2][48];
414     int chroma_weight[2][48][2];
415     int chroma_offset[2][48][2];
416     int implicit_weight[48][48];
417
418     //deblock
419     int deblocking_filter;         ///< disable_deblocking_filter_idc with 1<->0
420     int slice_alpha_c0_offset;
421     int slice_beta_offset;
422
423     int redundant_pic_count;
424
425     int direct_spatial_mv_pred;
426     int dist_scale_factor[16];
427     int dist_scale_factor_field[2][32];
428     int map_col_to_list0[2][16+32];
429     int map_col_to_list0_field[2][2][16+32];
430
431     /**
432      * num_ref_idx_l0/1_active_minus1 + 1
433      */
434     unsigned int ref_count[2];   ///< counts frames or fields, depending on current mb mode
435     unsigned int list_count;
436     uint8_t *list_counts;            ///< Array of list_count per MB specifying the slice type
437     Picture *short_ref[32];
438     Picture *long_ref[32];
439     Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
440     Picture ref_list[2][48];         /**< 0..15: frame refs, 16..47: mbaff field refs.
441                                           Reordered version of default_ref_list
442                                           according to picture reordering in slice header */
443     int ref2frm[MAX_SLICES][2][64];  ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
444     Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
445     int outputed_poc;
446
447     /**
448      * memory management control operations buffer.
449      */
450     MMCO mmco[MAX_MMCO_COUNT];
451     int mmco_index;
452
453     int long_ref_count;  ///< number of actual long term references
454     int short_ref_count; ///< number of actual short term references
455
456     //data partitioning
457     GetBitContext intra_gb;
458     GetBitContext inter_gb;
459     GetBitContext *intra_gb_ptr;
460     GetBitContext *inter_gb_ptr;
461
462     DECLARE_ALIGNED_16(DCTELEM, mb)[16*24];
463     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
464
465     /**
466      * Cabac
467      */
468     CABACContext cabac;
469     uint8_t      cabac_state[460];
470     int          cabac_init_idc;
471
472     /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
473     uint16_t     *cbp_table;
474     int cbp;
475     int top_cbp;
476     int left_cbp;
477     /* chroma_pred_mode for i4x4 or i16x16, else 0 */
478     uint8_t     *chroma_pred_mode_table;
479     int         last_qscale_diff;
480     int16_t     (*mvd_table[2])[2];
481     DECLARE_ALIGNED_16(int16_t, mvd_cache)[2][5*8][2];
482     uint8_t     *direct_table;
483     uint8_t     direct_cache[5*8];
484
485     uint8_t zigzag_scan[16];
486     uint8_t zigzag_scan8x8[64];
487     uint8_t zigzag_scan8x8_cavlc[64];
488     uint8_t field_scan[16];
489     uint8_t field_scan8x8[64];
490     uint8_t field_scan8x8_cavlc[64];
491     const uint8_t *zigzag_scan_q0;
492     const uint8_t *zigzag_scan8x8_q0;
493     const uint8_t *zigzag_scan8x8_cavlc_q0;
494     const uint8_t *field_scan_q0;
495     const uint8_t *field_scan8x8_q0;
496     const uint8_t *field_scan8x8_cavlc_q0;
497
498     int x264_build;
499
500     /**
501      * @defgroup multithreading Members for slice based multithreading
502      * @{
503      */
504     struct H264Context *thread_context[MAX_THREADS];
505
506     /**
507      * current slice number, used to initalize slice_num of each thread/context
508      */
509     int current_slice;
510
511     /**
512      * Max number of threads / contexts.
513      * This is equal to AVCodecContext.thread_count unless
514      * multithreaded decoding is impossible, in which case it is
515      * reduced to 1.
516      */
517     int max_contexts;
518
519     /**
520      *  1 if the single thread fallback warning has already been
521      *  displayed, 0 otherwise.
522      */
523     int single_decode_warning;
524
525     int last_slice_type;
526     /** @} */
527
528     int mb_xy;
529
530     uint32_t svq3_watermark_key;
531
532     /**
533      * pic_struct in picture timing SEI message
534      */
535     SEI_PicStructType sei_pic_struct;
536
537     /**
538      * Complement sei_pic_struct
539      * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
540      * However, soft telecined frames may have these values.
541      * This is used in an attempt to flag soft telecine progressive.
542      */
543     int prev_interlaced_frame;
544
545     /**
546      * Bit set of clock types for fields/frames in picture timing SEI message.
547      * For each found ct_type, appropriate bit is set (e.g., bit 1 for
548      * interlaced).
549      */
550     int sei_ct_type;
551
552     /**
553      * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
554      */
555     int sei_dpb_output_delay;
556
557     /**
558      * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
559      */
560     int sei_cpb_removal_delay;
561
562     /**
563      * recovery_frame_cnt from SEI message
564      *
565      * Set to -1 if no recovery point SEI message found or to number of frames
566      * before playback synchronizes. Frames having recovery point are key
567      * frames.
568      */
569     int sei_recovery_frame_cnt;
570
571     int is_complex;
572
573     int luma_weight_flag[2];   ///< 7.4.3.2 luma_weight_lX_flag
574     int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
575
576     // Timestamp stuff
577     int sei_buffering_period_present;  ///< Buffering period SEI flag
578     int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
579 }H264Context;
580
581
582 extern const uint8_t ff_h264_chroma_qp[52];
583
584
585 /**
586  * Decode SEI
587  */
588 int ff_h264_decode_sei(H264Context *h);
589
590 /**
591  * Decode SPS
592  */
593 int ff_h264_decode_seq_parameter_set(H264Context *h);
594
595 /**
596  * Decode PPS
597  */
598 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
599
600 /**
601  * Decodes a network abstraction layer unit.
602  * @param consumed is the number of bytes used as input
603  * @param length is the length of the array
604  * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
605  * @returns decoded bytes, might be src+1 if no escapes
606  */
607 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
608
609 /**
610  * identifies the exact end of the bitstream
611  * @return the length of the trailing, or 0 if damaged
612  */
613 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
614
615 /**
616  * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
617  */
618 av_cold void ff_h264_free_context(H264Context *h);
619
620 /**
621  * reconstructs bitstream slice_type.
622  */
623 int ff_h264_get_slice_type(const H264Context *h);
624
625 /**
626  * allocates tables.
627  * needs width/height
628  */
629 int ff_h264_alloc_tables(H264Context *h);
630
631 /**
632  * fills the default_ref_list.
633  */
634 int ff_h264_fill_default_ref_list(H264Context *h);
635
636 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
637 void ff_h264_fill_mbaff_ref_list(H264Context *h);
638 void ff_h264_remove_all_refs(H264Context *h);
639
640 /**
641  * Executes the reference picture marking (memory management control operations).
642  */
643 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
644
645 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
646
647
648 /**
649  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
650  */
651 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
652
653 /**
654  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
655  */
656 int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
657
658 void ff_h264_write_back_intra_pred_mode(H264Context *h);
659 void ff_h264_hl_decode_mb(H264Context *h);
660 int ff_h264_frame_start(H264Context *h);
661 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
662 av_cold int ff_h264_decode_end(AVCodecContext *avctx);
663 av_cold void ff_h264_decode_init_vlc(void);
664
665 /**
666  * decodes a macroblock
667  * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
668  */
669 int ff_h264_decode_mb_cavlc(H264Context *h);
670
671 /**
672  * decodes a CABAC coded macroblock
673  * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
674  */
675 int ff_h264_decode_mb_cabac(H264Context *h);
676
677 void ff_h264_init_cabac_states(H264Context *h);
678
679 void ff_h264_direct_dist_scale_factor(H264Context * const h);
680 void ff_h264_direct_ref_list_init(H264Context * const h);
681 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
682
683 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);
684 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);
685
686 /**
687  * Reset SEI values at the beginning of the frame.
688  *
689  * @param h H.264 context.
690  */
691 void ff_h264_reset_sei(H264Context *h);
692
693
694 /*
695 o-o o-o
696  / / /
697 o-o o-o
698  ,---'
699 o-o o-o
700  / / /
701 o-o o-o
702 */
703 //This table must be here because scan8[constant] must be known at compiletime
704 static const uint8_t scan8[16 + 2*4]={
705  4+1*8, 5+1*8, 4+2*8, 5+2*8,
706  6+1*8, 7+1*8, 6+2*8, 7+2*8,
707  4+3*8, 5+3*8, 4+4*8, 5+4*8,
708  6+3*8, 7+3*8, 6+4*8, 7+4*8,
709  1+1*8, 2+1*8,
710  1+2*8, 2+2*8,
711  1+4*8, 2+4*8,
712  1+5*8, 2+5*8,
713 };
714
715 static av_always_inline uint32_t pack16to32(int a, int b){
716 #if HAVE_BIGENDIAN
717    return (b&0xFFFF) + (a<<16);
718 #else
719    return (a&0xFFFF) + (b<<16);
720 #endif
721 }
722
723 /**
724  * gets the chroma qp.
725  */
726 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
727     return h->pps.chroma_qp_table[t][qscale];
728 }
729
730 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
731
732 static void fill_decode_caches(H264Context *h, int mb_type){
733     MpegEncContext * const s = &h->s;
734     const int mb_xy= h->mb_xy;
735     int topleft_xy, top_xy, topright_xy, left_xy[2];
736     int topleft_type, top_type, topright_type, left_type[2];
737     const uint8_t * left_block;
738     int topleft_partition= -1;
739     int i;
740     static const uint8_t left_block_options[4][16]={
741         {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},
742         {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},
743         {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},
744         {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}
745     };
746
747     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
748
749     /* Wow, what a mess, why didn't they simplify the interlacing & intra
750      * stuff, I can't imagine that these complex rules are worth it. */
751
752     topleft_xy = top_xy - 1;
753     topright_xy= top_xy + 1;
754     left_xy[1] = left_xy[0] = mb_xy-1;
755     left_block = left_block_options[0];
756     if(FRAME_MBAFF){
757         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
758         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
759         if(s->mb_y&1){
760             if (left_mb_field_flag != curr_mb_field_flag) {
761                 left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
762                 if (curr_mb_field_flag) {
763                     left_xy[1] += s->mb_stride;
764                     left_block = left_block_options[3];
765                 } else {
766                     topleft_xy += s->mb_stride;
767                     // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
768                     topleft_partition = 0;
769                     left_block = left_block_options[1];
770                 }
771             }
772         }else{
773             if(curr_mb_field_flag){
774                 topleft_xy  += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
775                 topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
776                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
777             }
778             if (left_mb_field_flag != curr_mb_field_flag) {
779                 left_xy[1] = left_xy[0] = mb_xy - 1;
780                 if (curr_mb_field_flag) {
781                     left_xy[1] += s->mb_stride;
782                     left_block = left_block_options[3];
783                 } else {
784                     left_block = left_block_options[2];
785                 }
786             }
787         }
788     }
789
790     h->top_mb_xy = top_xy;
791     h->left_mb_xy[0] = left_xy[0];
792     h->left_mb_xy[1] = left_xy[1];
793         topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
794         top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
795         topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
796         left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
797         left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
798
799         if(IS_INTRA(mb_type)){
800             int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
801             h->topleft_samples_available=
802             h->top_samples_available=
803             h->left_samples_available= 0xFFFF;
804             h->topright_samples_available= 0xEEEA;
805
806             if(!(top_type & type_mask)){
807                 h->topleft_samples_available= 0xB3FF;
808                 h->top_samples_available= 0x33FF;
809                 h->topright_samples_available= 0x26EA;
810             }
811             if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
812                 if(IS_INTERLACED(mb_type)){
813                     if(!(left_type[0] & type_mask)){
814                         h->topleft_samples_available&= 0xDFFF;
815                         h->left_samples_available&= 0x5FFF;
816                     }
817                     if(!(left_type[1] & type_mask)){
818                         h->topleft_samples_available&= 0xFF5F;
819                         h->left_samples_available&= 0xFF5F;
820                     }
821                 }else{
822                     int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
823                                     ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
824                     assert(left_xy[0] == left_xy[1]);
825                     if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
826                         h->topleft_samples_available&= 0xDF5F;
827                         h->left_samples_available&= 0x5F5F;
828                     }
829                 }
830             }else{
831                 if(!(left_type[0] & type_mask)){
832                     h->topleft_samples_available&= 0xDF5F;
833                     h->left_samples_available&= 0x5F5F;
834                 }
835             }
836
837             if(!(topleft_type & type_mask))
838                 h->topleft_samples_available&= 0x7FFF;
839
840             if(!(topright_type & type_mask))
841                 h->topright_samples_available&= 0xFBFF;
842
843             if(IS_INTRA4x4(mb_type)){
844                 if(IS_INTRA4x4(top_type)){
845                     h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
846                     h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
847                     h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
848                     h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
849                 }else{
850                     int pred;
851                     if(!(top_type & type_mask))
852                         pred= -1;
853                     else{
854                         pred= 2;
855                     }
856                     h->intra4x4_pred_mode_cache[4+8*0]=
857                     h->intra4x4_pred_mode_cache[5+8*0]=
858                     h->intra4x4_pred_mode_cache[6+8*0]=
859                     h->intra4x4_pred_mode_cache[7+8*0]= pred;
860                 }
861                 for(i=0; i<2; i++){
862                     if(IS_INTRA4x4(left_type[i])){
863                         h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
864                         h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
865                     }else{
866                         int pred;
867                         if(!(left_type[i] & type_mask))
868                             pred= -1;
869                         else{
870                             pred= 2;
871                         }
872                         h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
873                         h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
874                     }
875                 }
876             }
877         }
878
879
880 /*
881 0 . T T. T T T T
882 1 L . .L . . . .
883 2 L . .L . . . .
884 3 . T TL . . . .
885 4 L . .L . . . .
886 5 L . .. . . . .
887 */
888 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
889     if(top_type){
890         *(uint32_t*)&h->non_zero_count_cache[4+8*0]= *(uint32_t*)&h->non_zero_count[top_xy][4+3*8];
891             h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
892             h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
893
894             h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
895             h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
896     }else {
897             h->non_zero_count_cache[1+8*0]=
898             h->non_zero_count_cache[2+8*0]=
899
900             h->non_zero_count_cache[1+8*3]=
901             h->non_zero_count_cache[2+8*3]=
902             *(uint32_t*)&h->non_zero_count_cache[4+8*0]= CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040;
903     }
904
905     for (i=0; i<2; i++) {
906         if(left_type[i]){
907             h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
908             h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
909                 h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
910                 h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
911         }else{
912                 h->non_zero_count_cache[3+8*1 + 2*8*i]=
913                 h->non_zero_count_cache[3+8*2 + 2*8*i]=
914                 h->non_zero_count_cache[0+8*1 +   8*i]=
915                 h->non_zero_count_cache[0+8*4 +   8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
916         }
917     }
918
919     if( CABAC ) {
920         // top_cbp
921         if(top_type) {
922             h->top_cbp = h->cbp_table[top_xy];
923         } else if(IS_INTRA(mb_type)) {
924             h->top_cbp = 0x1C0;
925         } else {
926             h->top_cbp = 0;
927         }
928         // left_cbp
929         if (left_type[0]) {
930             h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
931         } else if(IS_INTRA(mb_type)) {
932             h->left_cbp = 0x1C0;
933         } else {
934             h->left_cbp = 0;
935         }
936         if (left_type[0]) {
937             h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
938         }
939         if (left_type[1]) {
940             h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
941         }
942     }
943
944 #if 1
945     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
946         int list;
947         for(list=0; list<h->list_count; list++){
948             if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type)){
949                 /*if(!h->mv_cache_clean[list]){
950                     memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
951                     memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
952                     h->mv_cache_clean[list]= 1;
953                 }*/
954                 continue;
955             }
956             h->mv_cache_clean[list]= 0;
957
958             if(USES_LIST(top_type, list)){
959                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
960                 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
961                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
962                     h->ref_cache[list][scan8[0] + 0 - 1*8]=
963                     h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
964                     h->ref_cache[list][scan8[0] + 2 - 1*8]=
965                     h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
966             }else{
967                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
968                 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
969             }
970
971             for(i=0; i<2; i++){
972                 int cache_idx = scan8[0] - 1 + i*2*8;
973                 if(USES_LIST(left_type[i], list)){
974                     const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
975                     const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
976                     *(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]];
977                     *(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]];
978                         h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
979                         h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
980                 }else{
981                     *(uint32_t*)h->mv_cache [list][cache_idx  ]=
982                     *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
983                     h->ref_cache[list][cache_idx  ]=
984                     h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
985                 }
986             }
987
988             if((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF)
989                 continue;
990
991             if(USES_LIST(topleft_type, list)){
992                 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
993                 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
994                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
995                 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
996             }else{
997                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
998                 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
999             }
1000
1001             if(USES_LIST(topright_type, list)){
1002                 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1003                 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
1004                 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
1005                 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1006             }else{
1007                 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
1008                 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1009             }
1010
1011             if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
1012                 continue;
1013
1014             h->ref_cache[list][scan8[5 ]+1] =
1015             h->ref_cache[list][scan8[7 ]+1] =
1016             h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
1017             h->ref_cache[list][scan8[4 ]] =
1018             h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1019             *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
1020             *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
1021             *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
1022             *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
1023             *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
1024
1025             if( CABAC ) {
1026                 /* XXX beurk, Load mvd */
1027                 if(USES_LIST(top_type, list)){
1028                     const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1029                     AV_COPY128(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1030                 }else{
1031                     AV_ZERO128(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1032                 }
1033                 if(USES_LIST(left_type[0], list)){
1034                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1035                     *(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]];
1036                     *(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]];
1037                 }else{
1038                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
1039                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
1040                 }
1041                 if(USES_LIST(left_type[1], list)){
1042                     const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
1043                     *(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]];
1044                     *(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]];
1045                 }else{
1046                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
1047                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
1048                 }
1049                 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
1050                 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
1051                 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
1052                 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
1053                 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
1054
1055                 if(h->slice_type_nos == FF_B_TYPE){
1056                     fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
1057
1058                     if(IS_DIRECT(top_type)){
1059                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
1060                     }else if(IS_8X8(top_type)){
1061                         int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
1062                         h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
1063                         h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
1064                     }else{
1065                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
1066                     }
1067
1068                     if(IS_DIRECT(left_type[0]))
1069                         h->direct_cache[scan8[0] - 1 + 0*8]= 1;
1070                     else if(IS_8X8(left_type[0]))
1071                         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)];
1072                     else
1073                         h->direct_cache[scan8[0] - 1 + 0*8]= 0;
1074
1075                     if(IS_DIRECT(left_type[1]))
1076                         h->direct_cache[scan8[0] - 1 + 2*8]= 1;
1077                     else if(IS_8X8(left_type[1]))
1078                         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)];
1079                     else
1080                         h->direct_cache[scan8[0] - 1 + 2*8]= 0;
1081                 }
1082             }
1083
1084             if(FRAME_MBAFF){
1085 #define MAP_MVS\
1086                     MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1087                     MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1088                     MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1089                     MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1090                     MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1091                     MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1092                     MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1093                     MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1094                     MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1095                     MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1096                 if(MB_FIELD){
1097 #define MAP_F2F(idx, mb_type)\
1098                     if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1099                         h->ref_cache[list][idx] <<= 1;\
1100                         h->mv_cache[list][idx][1] /= 2;\
1101                         h->mvd_cache[list][idx][1] /= 2;\
1102                     }
1103                     MAP_MVS
1104 #undef MAP_F2F
1105                 }else{
1106 #define MAP_F2F(idx, mb_type)\
1107                     if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1108                         h->ref_cache[list][idx] >>= 1;\
1109                         h->mv_cache[list][idx][1] <<= 1;\
1110                         h->mvd_cache[list][idx][1] <<= 1;\
1111                     }
1112                     MAP_MVS
1113 #undef MAP_F2F
1114                 }
1115             }
1116         }
1117     }
1118 #endif
1119
1120         h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1121 }
1122
1123 /**
1124  *
1125  * @returns non zero if the loop filter can be skiped
1126  */
1127 static int fill_filter_caches(H264Context *h, int mb_type){
1128     MpegEncContext * const s = &h->s;
1129     const int mb_xy= h->mb_xy;
1130     int top_xy, left_xy[2];
1131     int top_type, left_type[2];
1132     int i;
1133
1134     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
1135
1136     //FIXME deblocking could skip the intra and nnz parts.
1137
1138     /* Wow, what a mess, why didn't they simplify the interlacing & intra
1139      * stuff, I can't imagine that these complex rules are worth it. */
1140
1141     left_xy[1] = left_xy[0] = mb_xy-1;
1142     if(FRAME_MBAFF){
1143         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
1144         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
1145         if(s->mb_y&1){
1146             if (left_mb_field_flag != curr_mb_field_flag) {
1147                 left_xy[0] -= s->mb_stride;
1148             }
1149         }else{
1150             if(curr_mb_field_flag){
1151                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
1152             }
1153             if (left_mb_field_flag != curr_mb_field_flag) {
1154                 left_xy[1] += s->mb_stride;
1155             }
1156         }
1157     }
1158
1159     h->top_mb_xy = top_xy;
1160     h->left_mb_xy[0] = left_xy[0];
1161     h->left_mb_xy[1] = left_xy[1];
1162     {
1163         //for sufficiently low qp, filtering wouldn't do anything
1164         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
1165         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
1166         int qp = s->current_picture.qscale_table[mb_xy];
1167         if(qp <= qp_thresh
1168            && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
1169            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
1170             if(!FRAME_MBAFF)
1171                 return 1;
1172             if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
1173                && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
1174                 return 1;
1175         }
1176     }
1177
1178     if(h->deblocking_filter == 2){
1179         h->top_type    = top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
1180         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;
1181         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;
1182     }else{
1183         h->top_type    = top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
1184         h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
1185         h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
1186     }
1187     if(IS_INTRA(mb_type))
1188         return 0;
1189
1190     AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
1191     AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
1192     *((uint32_t*)&h->non_zero_count_cache[0+8*5])= *((uint32_t*)&h->non_zero_count[mb_xy][16]);
1193     *((uint32_t*)&h->non_zero_count_cache[4+8*3])= *((uint32_t*)&h->non_zero_count[mb_xy][20]);
1194     AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
1195
1196     h->cbp= h->cbp_table[mb_xy];
1197
1198     {
1199         int list;
1200         for(list=0; list<h->list_count; list++){
1201             int8_t *ref;
1202             int y, b_stride;
1203             int16_t (*mv_dst)[2];
1204             int16_t (*mv_src)[2];
1205
1206             if(!USES_LIST(mb_type, list)){
1207                 fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
1208                 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
1209                 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] =
1210                 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
1211                 *(uint32_t*)&h->ref_cache[list][scan8[10]] = ((LIST_NOT_USED)&0xFF)*0x01010101;
1212                 continue;
1213             }
1214
1215             ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
1216             {
1217                 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1218                 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
1219                 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
1220                 ref += h->b8_stride;
1221                 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
1222                 *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
1223             }
1224
1225             b_stride = h->b_stride;
1226             mv_dst   = &h->mv_cache[list][scan8[0]];
1227             mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
1228             for(y=0; y<4; y++){
1229                 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
1230             }
1231
1232         }
1233     }
1234
1235
1236 /*
1237 0 . T T. T T T T
1238 1 L . .L . . . .
1239 2 L . .L . . . .
1240 3 . T TL . . . .
1241 4 L . .L . . . .
1242 5 L . .. . . . .
1243 */
1244 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
1245     if(top_type){
1246         *(uint32_t*)&h->non_zero_count_cache[4+8*0]= *(uint32_t*)&h->non_zero_count[top_xy][4+3*8];
1247     }
1248
1249     if(left_type[0]){
1250         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
1251         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
1252         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
1253         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
1254     }
1255
1256     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
1257     if(!CABAC && h->pps.transform_8x8_mode){
1258         if(IS_8x8DCT(top_type)){
1259             h->non_zero_count_cache[4+8*0]=
1260             h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
1261             h->non_zero_count_cache[6+8*0]=
1262             h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
1263         }
1264         if(IS_8x8DCT(left_type[0])){
1265             h->non_zero_count_cache[3+8*1]=
1266             h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
1267         }
1268         if(IS_8x8DCT(left_type[1])){
1269             h->non_zero_count_cache[3+8*3]=
1270             h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
1271         }
1272
1273         if(IS_8x8DCT(mb_type)){
1274             h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
1275             h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
1276
1277             h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
1278             h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
1279
1280             h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
1281             h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
1282
1283             h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
1284             h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
1285         }
1286     }
1287
1288     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
1289         int list;
1290         for(list=0; list<h->list_count; list++){
1291             if(USES_LIST(top_type, list)){
1292                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1293                 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
1294                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1295                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1296                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
1297                 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
1298                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
1299                 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
1300             }else{
1301                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1302                 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((LIST_NOT_USED)&0xFF)*0x01010101;
1303             }
1304
1305             if(!IS_INTERLACED(mb_type^left_type[0])){
1306                 if(USES_LIST(left_type[0], list)){
1307                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1308                     const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
1309                     int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1310                     *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 0 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*0];
1311                     *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 8 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*1];
1312                     *(uint32_t*)h->mv_cache[list][scan8[0] - 1 +16 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*2];
1313                     *(uint32_t*)h->mv_cache[list][scan8[0] - 1 +24 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*3];
1314                     h->ref_cache[list][scan8[0] - 1 + 0 ]=
1315                     h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + h->b8_stride*0]];
1316                     h->ref_cache[list][scan8[0] - 1 +16 ]=
1317                     h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + h->b8_stride*1]];
1318                 }else{
1319                     *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0 ]=
1320                     *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 8 ]=
1321                     *(uint32_t*)h->mv_cache [list][scan8[0] - 1 +16 ]=
1322                     *(uint32_t*)h->mv_cache [list][scan8[0] - 1 +24 ]= 0;
1323                     h->ref_cache[list][scan8[0] - 1 + 0  ]=
1324                     h->ref_cache[list][scan8[0] - 1 + 8  ]=
1325                     h->ref_cache[list][scan8[0] - 1 + 16 ]=
1326                     h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
1327                 }
1328             }
1329         }
1330     }
1331
1332     return 0;
1333 }
1334
1335 /**
1336  * gets the predicted intra4x4 prediction mode.
1337  */
1338 static inline int pred_intra_mode(H264Context *h, int n){
1339     const int index8= scan8[n];
1340     const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1341     const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1342     const int min= FFMIN(left, top);
1343
1344     tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1345
1346     if(min<0) return DC_PRED;
1347     else      return min;
1348 }
1349
1350 static inline void write_back_non_zero_count(H264Context *h){
1351     const int mb_xy= h->mb_xy;
1352
1353     AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1354     AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1355     *((uint32_t*)&h->non_zero_count[mb_xy][16]) = *((uint32_t*)&h->non_zero_count_cache[0+8*5]);
1356     *((uint32_t*)&h->non_zero_count[mb_xy][20]) = *((uint32_t*)&h->non_zero_count_cache[4+8*3]);
1357     AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1358 }
1359
1360 static inline void write_back_motion(H264Context *h, int mb_type){
1361     MpegEncContext * const s = &h->s;
1362     const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1363     const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1364     int list;
1365
1366     if(!USES_LIST(mb_type, 0))
1367         fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1368
1369     for(list=0; list<h->list_count; list++){
1370         int y, b_stride;
1371         int16_t (*mv_dst)[2];
1372         int16_t (*mv_src)[2];
1373
1374         if(!USES_LIST(mb_type, list))
1375             continue;
1376
1377         b_stride = h->b_stride;
1378         mv_dst   = &s->current_picture.motion_val[list][b_xy];
1379         mv_src   = &h->mv_cache[list][scan8[0]];
1380         for(y=0; y<4; y++){
1381             AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1382         }
1383         if( CABAC ) {
1384             int16_t (*mvd_dst)[2] = &h->mvd_table[list][b_xy];
1385             int16_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1386             if(IS_SKIP(mb_type))
1387                 fill_rectangle(mvd_dst, 4, 4, h->b_stride, 0, 4);
1388             else
1389             for(y=0; y<4; y++){
1390                 AV_COPY128(mvd_dst + y*b_stride, mvd_src + 8*y);
1391             }
1392         }
1393
1394         {
1395             int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1396             ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1397             ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1398             ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1399             ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1400         }
1401     }
1402
1403     if(h->slice_type_nos == FF_B_TYPE && CABAC){
1404         if(IS_8X8(mb_type)){
1405             uint8_t *direct_table = &h->direct_table[b8_xy];
1406             direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1407             direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1408             direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1409         }
1410     }
1411 }
1412
1413 static inline int get_dct8x8_allowed(H264Context *h){
1414     if(h->sps.direct_8x8_inference_flag)
1415         return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
1416     else
1417         return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1418 }
1419
1420 static void predict_field_decoding_flag(H264Context *h){
1421     MpegEncContext * const s = &h->s;
1422     const int mb_xy= h->mb_xy;
1423     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
1424                 ? s->current_picture.mb_type[mb_xy-1]
1425                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
1426                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
1427                 : 0;
1428     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
1429 }
1430
1431 /**
1432  * decodes a P_SKIP or B_SKIP macroblock
1433  */
1434 static void decode_mb_skip(H264Context *h){
1435     MpegEncContext * const s = &h->s;
1436     const int mb_xy= h->mb_xy;
1437     int mb_type=0;
1438
1439     memset(h->non_zero_count[mb_xy], 0, 32);
1440     memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1441
1442     if(MB_FIELD)
1443         mb_type|= MB_TYPE_INTERLACED;
1444
1445     if( h->slice_type_nos == FF_B_TYPE )
1446     {
1447         // just for fill_caches. pred_direct_motion will set the real mb_type
1448         mb_type|= MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1449
1450         fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1451         ff_h264_pred_direct_motion(h, &mb_type);
1452         mb_type|= MB_TYPE_SKIP;
1453     }
1454     else
1455     {
1456         int mx, my;
1457         mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1458
1459         fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1460         pred_pskip_motion(h, &mx, &my);
1461         fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1462         fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1463     }
1464
1465     write_back_motion(h, mb_type);
1466     s->current_picture.mb_type[mb_xy]= mb_type;
1467     s->current_picture.qscale_table[mb_xy]= s->qscale;
1468     h->slice_table[ mb_xy ]= h->slice_num;
1469     h->prev_mb_skipped= 1;
1470 }
1471
1472 #include "h264_mvpred.h" //For pred_pskip_motion()
1473
1474 #endif /* AVCODEC_H264_H */