]> git.sesse.net Git - ffmpeg/blob - libavcodec/h264.h
Avoid wasting 4 cpu cycles per MB in redundantly calculating qp_thresh.
[ffmpeg] / 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)[16];
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     Picture *short_ref[32];
427     Picture *long_ref[32];
428     Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
429     Picture ref_list[2][48];         /**< 0..15: frame refs, 16..47: mbaff field refs.
430                                           Reordered version of default_ref_list
431                                           according to picture reordering in slice header */
432     int ref2frm[MAX_SLICES][2][64];  ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
433     Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
434     int outputed_poc;
435
436     /**
437      * memory management control operations buffer.
438      */
439     MMCO mmco[MAX_MMCO_COUNT];
440     int mmco_index;
441
442     int long_ref_count;  ///< number of actual long term references
443     int short_ref_count; ///< number of actual short term references
444
445     //data partitioning
446     GetBitContext intra_gb;
447     GetBitContext inter_gb;
448     GetBitContext *intra_gb_ptr;
449     GetBitContext *inter_gb_ptr;
450
451     DECLARE_ALIGNED_16(DCTELEM, mb[16*24]);
452     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
453
454     /**
455      * Cabac
456      */
457     CABACContext cabac;
458     uint8_t      cabac_state[460];
459     int          cabac_init_idc;
460
461     /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
462     uint16_t     *cbp_table;
463     int cbp;
464     int top_cbp;
465     int left_cbp;
466     /* chroma_pred_mode for i4x4 or i16x16, else 0 */
467     uint8_t     *chroma_pred_mode_table;
468     int         last_qscale_diff;
469     int16_t     (*mvd_table[2])[2];
470     DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]);
471     uint8_t     *direct_table;
472     uint8_t     direct_cache[5*8];
473
474     uint8_t zigzag_scan[16];
475     uint8_t zigzag_scan8x8[64];
476     uint8_t zigzag_scan8x8_cavlc[64];
477     uint8_t field_scan[16];
478     uint8_t field_scan8x8[64];
479     uint8_t field_scan8x8_cavlc[64];
480     const uint8_t *zigzag_scan_q0;
481     const uint8_t *zigzag_scan8x8_q0;
482     const uint8_t *zigzag_scan8x8_cavlc_q0;
483     const uint8_t *field_scan_q0;
484     const uint8_t *field_scan8x8_q0;
485     const uint8_t *field_scan8x8_cavlc_q0;
486
487     int x264_build;
488
489     /**
490      * @defgroup multithreading Members for slice based multithreading
491      * @{
492      */
493     struct H264Context *thread_context[MAX_THREADS];
494
495     /**
496      * current slice number, used to initalize slice_num of each thread/context
497      */
498     int current_slice;
499
500     /**
501      * Max number of threads / contexts.
502      * This is equal to AVCodecContext.thread_count unless
503      * multithreaded decoding is impossible, in which case it is
504      * reduced to 1.
505      */
506     int max_contexts;
507
508     /**
509      *  1 if the single thread fallback warning has already been
510      *  displayed, 0 otherwise.
511      */
512     int single_decode_warning;
513
514     int last_slice_type;
515     /** @} */
516
517     int mb_xy;
518
519     uint32_t svq3_watermark_key;
520
521     /**
522      * pic_struct in picture timing SEI message
523      */
524     SEI_PicStructType sei_pic_struct;
525
526     /**
527      * Complement sei_pic_struct
528      * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
529      * However, soft telecined frames may have these values.
530      * This is used in an attempt to flag soft telecine progressive.
531      */
532     int prev_interlaced_frame;
533
534     /**
535      * Bit set of clock types for fields/frames in picture timing SEI message.
536      * For each found ct_type, appropriate bit is set (e.g., bit 1 for
537      * interlaced).
538      */
539     int sei_ct_type;
540
541     /**
542      * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
543      */
544     int sei_dpb_output_delay;
545
546     /**
547      * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
548      */
549     int sei_cpb_removal_delay;
550
551     /**
552      * recovery_frame_cnt from SEI message
553      *
554      * Set to -1 if no recovery point SEI message found or to number of frames
555      * before playback synchronizes. Frames having recovery point are key
556      * frames.
557      */
558     int sei_recovery_frame_cnt;
559
560     int is_complex;
561
562     int luma_weight_flag[2];   ///< 7.4.3.2 luma_weight_lX_flag
563     int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
564
565     // Timestamp stuff
566     int sei_buffering_period_present;  ///< Buffering period SEI flag
567     int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
568 }H264Context;
569
570
571 extern const uint8_t ff_h264_chroma_qp[52];
572
573
574 /**
575  * Decode SEI
576  */
577 int ff_h264_decode_sei(H264Context *h);
578
579 /**
580  * Decode SPS
581  */
582 int ff_h264_decode_seq_parameter_set(H264Context *h);
583
584 /**
585  * Decode PPS
586  */
587 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
588
589 /**
590  * Decodes a network abstraction layer unit.
591  * @param consumed is the number of bytes used as input
592  * @param length is the length of the array
593  * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
594  * @returns decoded bytes, might be src+1 if no escapes
595  */
596 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
597
598 /**
599  * identifies the exact end of the bitstream
600  * @return the length of the trailing, or 0 if damaged
601  */
602 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
603
604 /**
605  * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
606  */
607 av_cold void ff_h264_free_context(H264Context *h);
608
609 /**
610  * reconstructs bitstream slice_type.
611  */
612 int ff_h264_get_slice_type(H264Context *h);
613
614 /**
615  * allocates tables.
616  * needs width/height
617  */
618 int ff_h264_alloc_tables(H264Context *h);
619
620 /**
621  * fills the default_ref_list.
622  */
623 int ff_h264_fill_default_ref_list(H264Context *h);
624
625 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
626 void ff_h264_fill_mbaff_ref_list(H264Context *h);
627 void ff_h264_remove_all_refs(H264Context *h);
628
629 /**
630  * Executes the reference picture marking (memory management control operations).
631  */
632 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
633
634 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
635
636
637 /**
638  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
639  */
640 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
641
642 /**
643  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
644  */
645 int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
646
647 void ff_h264_write_back_intra_pred_mode(H264Context *h);
648 void ff_h264_hl_decode_mb(H264Context *h);
649 int ff_h264_frame_start(H264Context *h);
650 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
651 av_cold int ff_h264_decode_end(AVCodecContext *avctx);
652 av_cold void ff_h264_decode_init_vlc(void);
653
654 /**
655  * decodes a macroblock
656  * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
657  */
658 int ff_h264_decode_mb_cavlc(H264Context *h);
659
660 /**
661  * decodes a CABAC coded macroblock
662  * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
663  */
664 int ff_h264_decode_mb_cabac(H264Context *h);
665
666 void ff_h264_init_cabac_states(H264Context *h);
667
668 void ff_h264_direct_dist_scale_factor(H264Context * const h);
669 void ff_h264_direct_ref_list_init(H264Context * const h);
670 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
671
672 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);
673 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);
674
675 /**
676  * Reset SEI values at the beginning of the frame.
677  *
678  * @param h H.264 context.
679  */
680 void ff_h264_reset_sei(H264Context *h);
681
682
683 /*
684 o-o o-o
685  / / /
686 o-o o-o
687  ,---'
688 o-o o-o
689  / / /
690 o-o o-o
691 */
692 //This table must be here because scan8[constant] must be known at compiletime
693 static const uint8_t scan8[16 + 2*4]={
694  4+1*8, 5+1*8, 4+2*8, 5+2*8,
695  6+1*8, 7+1*8, 6+2*8, 7+2*8,
696  4+3*8, 5+3*8, 4+4*8, 5+4*8,
697  6+3*8, 7+3*8, 6+4*8, 7+4*8,
698  1+1*8, 2+1*8,
699  1+2*8, 2+2*8,
700  1+4*8, 2+4*8,
701  1+5*8, 2+5*8,
702 };
703
704 static av_always_inline uint32_t pack16to32(int a, int b){
705 #if HAVE_BIGENDIAN
706    return (b&0xFFFF) + (a<<16);
707 #else
708    return (a&0xFFFF) + (b<<16);
709 #endif
710 }
711
712 /**
713  * gets the chroma qp.
714  */
715 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
716     return h->pps.chroma_qp_table[t][qscale];
717 }
718
719 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
720
721 static void fill_caches(H264Context *h, int mb_type, int for_deblock){
722     MpegEncContext * const s = &h->s;
723     const int mb_xy= h->mb_xy;
724     int topleft_xy, top_xy, topright_xy, left_xy[2];
725     int topleft_type, top_type, topright_type, left_type[2];
726     const uint8_t * left_block;
727     int topleft_partition= -1;
728     int i;
729     static const uint8_t left_block_options[4][8]={
730         {0,1,2,3,7,10,8,11},
731         {2,2,3,3,8,11,8,11},
732         {0,0,1,1,7,10,7,10},
733         {0,2,0,2,7,10,7,10}
734     };
735
736     top_xy     = mb_xy  - (s->mb_stride << FIELD_PICTURE);
737
738     //FIXME deblocking could skip the intra and nnz parts.
739     if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
740         return;
741
742     /* Wow, what a mess, why didn't they simplify the interlacing & intra
743      * stuff, I can't imagine that these complex rules are worth it. */
744
745     topleft_xy = top_xy - 1;
746     topright_xy= top_xy + 1;
747     left_xy[1] = left_xy[0] = mb_xy-1;
748     left_block = left_block_options[0];
749     if(FRAME_MBAFF){
750         const int pair_xy          = s->mb_x     + (s->mb_y & ~1)*s->mb_stride;
751         const int top_pair_xy      = pair_xy     - s->mb_stride;
752         const int topleft_pair_xy  = top_pair_xy - 1;
753         const int topright_pair_xy = top_pair_xy + 1;
754         const int topleft_mb_field_flag  = IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
755         const int top_mb_field_flag      = IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
756         const int topright_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
757         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
758         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
759         const int bottom = (s->mb_y & 1);
760         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);
761
762         if (curr_mb_field_flag && (bottom || top_mb_field_flag)){
763             top_xy -= s->mb_stride;
764         }
765         if (curr_mb_field_flag && (bottom || topleft_mb_field_flag)){
766             topleft_xy -= s->mb_stride;
767         } else if(bottom && !curr_mb_field_flag && left_mb_field_flag) {
768             topleft_xy += s->mb_stride;
769             // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
770             topleft_partition = 0;
771         }
772         if (curr_mb_field_flag && (bottom || topright_mb_field_flag)){
773             topright_xy -= s->mb_stride;
774         }
775         if (left_mb_field_flag != curr_mb_field_flag) {
776             left_xy[1] = left_xy[0] = pair_xy - 1;
777             if (curr_mb_field_flag) {
778                 left_xy[1] += s->mb_stride;
779                 left_block = left_block_options[3];
780             } else {
781                 left_block= left_block_options[2 - bottom];
782             }
783         }
784     }
785
786     h->top_mb_xy = top_xy;
787     h->left_mb_xy[0] = left_xy[0];
788     h->left_mb_xy[1] = left_xy[1];
789     if(for_deblock){
790         topleft_type = 0;
791         topright_type = 0;
792         top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
793         left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
794         left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
795
796         if(MB_MBAFF && !IS_INTRA(mb_type)){
797             int list;
798             for(list=0; list<h->list_count; list++){
799                 //These values where changed for ease of performing MC, we need to change them back
800                 //FIXME maybe we can make MC and loop filter use the same values or prevent
801                 //the MC code from changing ref_cache and rather use a temporary array.
802                 if(USES_LIST(mb_type,list)){
803                     int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
804                     *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
805                     *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
806                     ref += h->b8_stride;
807                     *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
808                     *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
809                 }
810             }
811         }
812     }else{
813         topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
814         top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
815         topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
816         left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
817         left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
818
819     if(IS_INTRA(mb_type)){
820         int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
821         h->topleft_samples_available=
822         h->top_samples_available=
823         h->left_samples_available= 0xFFFF;
824         h->topright_samples_available= 0xEEEA;
825
826         if(!(top_type & type_mask)){
827             h->topleft_samples_available= 0xB3FF;
828             h->top_samples_available= 0x33FF;
829             h->topright_samples_available= 0x26EA;
830         }
831         if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
832             if(IS_INTERLACED(mb_type)){
833                 if(!(left_type[0] & type_mask)){
834                     h->topleft_samples_available&= 0xDFFF;
835                     h->left_samples_available&= 0x5FFF;
836                 }
837                 if(!(left_type[1] & type_mask)){
838                     h->topleft_samples_available&= 0xFF5F;
839                     h->left_samples_available&= 0xFF5F;
840                 }
841             }else{
842                 int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
843                                 ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
844                 assert(left_xy[0] == left_xy[1]);
845                 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
846                     h->topleft_samples_available&= 0xDF5F;
847                     h->left_samples_available&= 0x5F5F;
848                 }
849             }
850         }else{
851             if(!(left_type[0] & type_mask)){
852                 h->topleft_samples_available&= 0xDF5F;
853                 h->left_samples_available&= 0x5F5F;
854             }
855         }
856
857         if(!(topleft_type & type_mask))
858             h->topleft_samples_available&= 0x7FFF;
859
860         if(!(topright_type & type_mask))
861             h->topright_samples_available&= 0xFBFF;
862
863         if(IS_INTRA4x4(mb_type)){
864             if(IS_INTRA4x4(top_type)){
865                 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
866                 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
867                 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
868                 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
869             }else{
870                 int pred;
871                 if(!(top_type & type_mask))
872                     pred= -1;
873                 else{
874                     pred= 2;
875                 }
876                 h->intra4x4_pred_mode_cache[4+8*0]=
877                 h->intra4x4_pred_mode_cache[5+8*0]=
878                 h->intra4x4_pred_mode_cache[6+8*0]=
879                 h->intra4x4_pred_mode_cache[7+8*0]= pred;
880             }
881             for(i=0; i<2; i++){
882                 if(IS_INTRA4x4(left_type[i])){
883                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
884                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
885                 }else{
886                     int pred;
887                     if(!(left_type[i] & type_mask))
888                         pred= -1;
889                     else{
890                         pred= 2;
891                     }
892                     h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
893                     h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
894                 }
895             }
896         }
897     }
898     }
899
900
901 /*
902 0 . T T. T T T T
903 1 L . .L . . . .
904 2 L . .L . . . .
905 3 . T TL . . . .
906 4 L . .L . . . .
907 5 L . .. . . . .
908 */
909 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
910     if(top_type){
911         h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
912         h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
913         h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
914         h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
915
916         h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
917         h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
918
919         h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
920         h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
921
922     }else{
923         h->non_zero_count_cache[4+8*0]=
924         h->non_zero_count_cache[5+8*0]=
925         h->non_zero_count_cache[6+8*0]=
926         h->non_zero_count_cache[7+8*0]=
927
928         h->non_zero_count_cache[1+8*0]=
929         h->non_zero_count_cache[2+8*0]=
930
931         h->non_zero_count_cache[1+8*3]=
932         h->non_zero_count_cache[2+8*3]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
933
934     }
935
936     for (i=0; i<2; i++) {
937         if(left_type[i]){
938             h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
939             h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
940             h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
941             h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
942         }else{
943             h->non_zero_count_cache[3+8*1 + 2*8*i]=
944             h->non_zero_count_cache[3+8*2 + 2*8*i]=
945             h->non_zero_count_cache[0+8*1 +   8*i]=
946             h->non_zero_count_cache[0+8*4 +   8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
947         }
948     }
949
950     if( CABAC ) {
951         // top_cbp
952         if(top_type) {
953             h->top_cbp = h->cbp_table[top_xy];
954         } else if(IS_INTRA(mb_type)) {
955             h->top_cbp = 0x1C0;
956         } else {
957             h->top_cbp = 0;
958         }
959         // left_cbp
960         if (left_type[0]) {
961             h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
962         } else if(IS_INTRA(mb_type)) {
963             h->left_cbp = 0x1C0;
964         } else {
965             h->left_cbp = 0;
966         }
967         if (left_type[0]) {
968             h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
969         }
970         if (left_type[1]) {
971             h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
972         }
973     }
974
975 #if 1
976     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
977         int list;
978         for(list=0; list<h->list_count; list++){
979             if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
980                 /*if(!h->mv_cache_clean[list]){
981                     memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
982                     memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
983                     h->mv_cache_clean[list]= 1;
984                 }*/
985                 continue;
986             }
987             h->mv_cache_clean[list]= 0;
988
989             if(USES_LIST(top_type, list)){
990                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
991                 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
992                 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
993                 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
994                 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
995                 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
996                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
997                 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
998                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
999                 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
1000             }else{
1001                 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
1002                 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
1003                 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
1004                 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
1005                 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
1006             }
1007
1008             for(i=0; i<2; i++){
1009                 int cache_idx = scan8[0] - 1 + i*2*8;
1010                 if(USES_LIST(left_type[i], list)){
1011                     const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1012                     const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
1013                     *(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]];
1014                     *(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]];
1015                     h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
1016                     h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
1017                 }else{
1018                     *(uint32_t*)h->mv_cache [list][cache_idx  ]=
1019                     *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
1020                     h->ref_cache[list][cache_idx  ]=
1021                     h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1022                 }
1023             }
1024
1025             if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
1026                 continue;
1027
1028             if(USES_LIST(topleft_type, list)){
1029                 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
1030                 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
1031                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
1032                 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1033             }else{
1034                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
1035                 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1036             }
1037
1038             if(USES_LIST(topright_type, list)){
1039                 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1040                 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
1041                 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
1042                 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1043             }else{
1044                 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
1045                 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1046             }
1047
1048             if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
1049                 continue;
1050
1051             h->ref_cache[list][scan8[5 ]+1] =
1052             h->ref_cache[list][scan8[7 ]+1] =
1053             h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
1054             h->ref_cache[list][scan8[4 ]] =
1055             h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1056             *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
1057             *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
1058             *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
1059             *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
1060             *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
1061
1062             if( CABAC ) {
1063                 /* XXX beurk, Load mvd */
1064                 if(USES_LIST(top_type, list)){
1065                     const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1066                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
1067                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
1068                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
1069                     *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
1070                 }else{
1071                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
1072                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
1073                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
1074                     *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
1075                 }
1076                 if(USES_LIST(left_type[0], list)){
1077                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1078                     *(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]];
1079                     *(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]];
1080                 }else{
1081                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
1082                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
1083                 }
1084                 if(USES_LIST(left_type[1], list)){
1085                     const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
1086                     *(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]];
1087                     *(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]];
1088                 }else{
1089                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
1090                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
1091                 }
1092                 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
1093                 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
1094                 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
1095                 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
1096                 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
1097
1098                 if(h->slice_type_nos == FF_B_TYPE){
1099                     fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
1100
1101                     if(IS_DIRECT(top_type)){
1102                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
1103                     }else if(IS_8X8(top_type)){
1104                         int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
1105                         h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
1106                         h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
1107                     }else{
1108                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
1109                     }
1110
1111                     if(IS_DIRECT(left_type[0]))
1112                         h->direct_cache[scan8[0] - 1 + 0*8]= 1;
1113                     else if(IS_8X8(left_type[0]))
1114                         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)];
1115                     else
1116                         h->direct_cache[scan8[0] - 1 + 0*8]= 0;
1117
1118                     if(IS_DIRECT(left_type[1]))
1119                         h->direct_cache[scan8[0] - 1 + 2*8]= 1;
1120                     else if(IS_8X8(left_type[1]))
1121                         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)];
1122                     else
1123                         h->direct_cache[scan8[0] - 1 + 2*8]= 0;
1124                 }
1125             }
1126
1127             if(FRAME_MBAFF){
1128 #define MAP_MVS\
1129                     MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1130                     MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1131                     MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1132                     MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1133                     MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1134                     MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1135                     MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1136                     MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1137                     MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1138                     MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1139                 if(MB_FIELD){
1140 #define MAP_F2F(idx, mb_type)\
1141                     if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1142                         h->ref_cache[list][idx] <<= 1;\
1143                         h->mv_cache[list][idx][1] /= 2;\
1144                         h->mvd_cache[list][idx][1] /= 2;\
1145                     }
1146                     MAP_MVS
1147 #undef MAP_F2F
1148                 }else{
1149 #define MAP_F2F(idx, mb_type)\
1150                     if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1151                         h->ref_cache[list][idx] >>= 1;\
1152                         h->mv_cache[list][idx][1] <<= 1;\
1153                         h->mvd_cache[list][idx][1] <<= 1;\
1154                     }
1155                     MAP_MVS
1156 #undef MAP_F2F
1157                 }
1158             }
1159         }
1160     }
1161 #endif
1162
1163     h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1164 }
1165
1166 /**
1167  * gets the predicted intra4x4 prediction mode.
1168  */
1169 static inline int pred_intra_mode(H264Context *h, int n){
1170     const int index8= scan8[n];
1171     const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1172     const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1173     const int min= FFMIN(left, top);
1174
1175     tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1176
1177     if(min<0) return DC_PRED;
1178     else      return min;
1179 }
1180
1181 static inline void write_back_non_zero_count(H264Context *h){
1182     const int mb_xy= h->mb_xy;
1183
1184     h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
1185     h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
1186     h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
1187     h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
1188     h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
1189     h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
1190     h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
1191
1192     h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
1193     h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
1194     h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
1195
1196     h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
1197     h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
1198     h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
1199 }
1200
1201 static inline void write_back_motion(H264Context *h, int mb_type){
1202     MpegEncContext * const s = &h->s;
1203     const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1204     const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1205     int list;
1206
1207     if(!USES_LIST(mb_type, 0))
1208         fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1209
1210     for(list=0; list<h->list_count; list++){
1211         int y;
1212         if(!USES_LIST(mb_type, list))
1213             continue;
1214
1215         for(y=0; y<4; y++){
1216             *(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];
1217             *(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];
1218         }
1219         if( CABAC ) {
1220             if(IS_SKIP(mb_type))
1221                 fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
1222             else
1223             for(y=0; y<4; y++){
1224                 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1225                 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1226             }
1227         }
1228
1229         {
1230             int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1231             ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1232             ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1233             ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1234             ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1235         }
1236     }
1237
1238     if(h->slice_type_nos == FF_B_TYPE && CABAC){
1239         if(IS_8X8(mb_type)){
1240             uint8_t *direct_table = &h->direct_table[b8_xy];
1241             direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1242             direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1243             direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1244         }
1245     }
1246 }
1247
1248 static inline int get_dct8x8_allowed(H264Context *h){
1249     if(h->sps.direct_8x8_inference_flag)
1250         return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
1251     else
1252         return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1253 }
1254
1255 static void predict_field_decoding_flag(H264Context *h){
1256     MpegEncContext * const s = &h->s;
1257     const int mb_xy= h->mb_xy;
1258     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
1259                 ? s->current_picture.mb_type[mb_xy-1]
1260                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
1261                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
1262                 : 0;
1263     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
1264 }
1265
1266 /**
1267  * decodes a P_SKIP or B_SKIP macroblock
1268  */
1269 static void decode_mb_skip(H264Context *h){
1270     MpegEncContext * const s = &h->s;
1271     const int mb_xy= h->mb_xy;
1272     int mb_type=0;
1273
1274     memset(h->non_zero_count[mb_xy], 0, 16);
1275     memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1276
1277     if(MB_FIELD)
1278         mb_type|= MB_TYPE_INTERLACED;
1279
1280     if( h->slice_type_nos == FF_B_TYPE )
1281     {
1282         // just for fill_caches. pred_direct_motion will set the real mb_type
1283         mb_type|= MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1284
1285         fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
1286         ff_h264_pred_direct_motion(h, &mb_type);
1287         mb_type|= MB_TYPE_SKIP;
1288     }
1289     else
1290     {
1291         int mx, my;
1292         mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1293
1294         fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
1295         pred_pskip_motion(h, &mx, &my);
1296         fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1297         fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1298     }
1299
1300     write_back_motion(h, mb_type);
1301     s->current_picture.mb_type[mb_xy]= mb_type;
1302     s->current_picture.qscale_table[mb_xy]= s->qscale;
1303     h->slice_table[ mb_xy ]= h->slice_num;
1304     h->prev_mb_skipped= 1;
1305 }
1306
1307 #include "h264_mvpred.h" //For pred_pskip_motion()
1308
1309 #endif /* AVCODEC_H264_H */