Branchless calculation of ref_offset.
[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 col_parity;
427     int col_fieldoff;
428     int dist_scale_factor[16];
429     int dist_scale_factor_field[2][32];
430     int map_col_to_list0[2][16+32];
431     int map_col_to_list0_field[2][2][16+32];
432
433     /**
434      * num_ref_idx_l0/1_active_minus1 + 1
435      */
436     unsigned int ref_count[2];   ///< counts frames or fields, depending on current mb mode
437     unsigned int list_count;
438     uint8_t *list_counts;            ///< Array of list_count per MB specifying the slice type
439     Picture *short_ref[32];
440     Picture *long_ref[32];
441     Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
442     Picture ref_list[2][48];         /**< 0..15: frame refs, 16..47: mbaff field refs.
443                                           Reordered version of default_ref_list
444                                           according to picture reordering in slice header */
445     int ref2frm[MAX_SLICES][2][64];  ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
446     Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
447     int outputed_poc;
448
449     /**
450      * memory management control operations buffer.
451      */
452     MMCO mmco[MAX_MMCO_COUNT];
453     int mmco_index;
454
455     int long_ref_count;  ///< number of actual long term references
456     int short_ref_count; ///< number of actual short term references
457
458     //data partitioning
459     GetBitContext intra_gb;
460     GetBitContext inter_gb;
461     GetBitContext *intra_gb_ptr;
462     GetBitContext *inter_gb_ptr;
463
464     DECLARE_ALIGNED_16(DCTELEM, mb)[16*24];
465     DCTELEM mb_padding[256];        ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either check that i is not too large or ensure that there is some unused stuff after mb
466
467     /**
468      * Cabac
469      */
470     CABACContext cabac;
471     uint8_t      cabac_state[460];
472     int          cabac_init_idc;
473
474     /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
475     uint16_t     *cbp_table;
476     int cbp;
477     int top_cbp;
478     int left_cbp;
479     /* chroma_pred_mode for i4x4 or i16x16, else 0 */
480     uint8_t     *chroma_pred_mode_table;
481     int         last_qscale_diff;
482     int16_t     (*mvd_table[2])[2];
483     DECLARE_ALIGNED_16(int16_t, mvd_cache)[2][5*8][2];
484     uint8_t     *direct_table;
485     uint8_t     direct_cache[5*8];
486
487     uint8_t zigzag_scan[16];
488     uint8_t zigzag_scan8x8[64];
489     uint8_t zigzag_scan8x8_cavlc[64];
490     uint8_t field_scan[16];
491     uint8_t field_scan8x8[64];
492     uint8_t field_scan8x8_cavlc[64];
493     const uint8_t *zigzag_scan_q0;
494     const uint8_t *zigzag_scan8x8_q0;
495     const uint8_t *zigzag_scan8x8_cavlc_q0;
496     const uint8_t *field_scan_q0;
497     const uint8_t *field_scan8x8_q0;
498     const uint8_t *field_scan8x8_cavlc_q0;
499
500     int x264_build;
501
502     /**
503      * @defgroup multithreading Members for slice based multithreading
504      * @{
505      */
506     struct H264Context *thread_context[MAX_THREADS];
507
508     /**
509      * current slice number, used to initalize slice_num of each thread/context
510      */
511     int current_slice;
512
513     /**
514      * Max number of threads / contexts.
515      * This is equal to AVCodecContext.thread_count unless
516      * multithreaded decoding is impossible, in which case it is
517      * reduced to 1.
518      */
519     int max_contexts;
520
521     /**
522      *  1 if the single thread fallback warning has already been
523      *  displayed, 0 otherwise.
524      */
525     int single_decode_warning;
526
527     int last_slice_type;
528     /** @} */
529
530     int mb_xy;
531
532     uint32_t svq3_watermark_key;
533
534     /**
535      * pic_struct in picture timing SEI message
536      */
537     SEI_PicStructType sei_pic_struct;
538
539     /**
540      * Complement sei_pic_struct
541      * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
542      * However, soft telecined frames may have these values.
543      * This is used in an attempt to flag soft telecine progressive.
544      */
545     int prev_interlaced_frame;
546
547     /**
548      * Bit set of clock types for fields/frames in picture timing SEI message.
549      * For each found ct_type, appropriate bit is set (e.g., bit 1 for
550      * interlaced).
551      */
552     int sei_ct_type;
553
554     /**
555      * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
556      */
557     int sei_dpb_output_delay;
558
559     /**
560      * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
561      */
562     int sei_cpb_removal_delay;
563
564     /**
565      * recovery_frame_cnt from SEI message
566      *
567      * Set to -1 if no recovery point SEI message found or to number of frames
568      * before playback synchronizes. Frames having recovery point are key
569      * frames.
570      */
571     int sei_recovery_frame_cnt;
572
573     int is_complex;
574
575     int luma_weight_flag[2];   ///< 7.4.3.2 luma_weight_lX_flag
576     int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
577
578     // Timestamp stuff
579     int sei_buffering_period_present;  ///< Buffering period SEI flag
580     int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
581 }H264Context;
582
583
584 extern const uint8_t ff_h264_chroma_qp[52];
585
586 void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
587
588 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
589
590 /**
591  * Decode SEI
592  */
593 int ff_h264_decode_sei(H264Context *h);
594
595 /**
596  * Decode SPS
597  */
598 int ff_h264_decode_seq_parameter_set(H264Context *h);
599
600 /**
601  * Decode PPS
602  */
603 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
604
605 /**
606  * Decodes a network abstraction layer unit.
607  * @param consumed is the number of bytes used as input
608  * @param length is the length of the array
609  * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
610  * @returns decoded bytes, might be src+1 if no escapes
611  */
612 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
613
614 /**
615  * identifies the exact end of the bitstream
616  * @return the length of the trailing, or 0 if damaged
617  */
618 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
619
620 /**
621  * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
622  */
623 av_cold void ff_h264_free_context(H264Context *h);
624
625 /**
626  * reconstructs bitstream slice_type.
627  */
628 int ff_h264_get_slice_type(const H264Context *h);
629
630 /**
631  * allocates tables.
632  * needs width/height
633  */
634 int ff_h264_alloc_tables(H264Context *h);
635
636 /**
637  * fills the default_ref_list.
638  */
639 int ff_h264_fill_default_ref_list(H264Context *h);
640
641 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
642 void ff_h264_fill_mbaff_ref_list(H264Context *h);
643 void ff_h264_remove_all_refs(H264Context *h);
644
645 /**
646  * Executes the reference picture marking (memory management control operations).
647  */
648 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
649
650 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
651
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_intra4x4_pred_mode(H264Context *h);
657
658 /**
659  * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
660  */
661 int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
662
663 void ff_h264_write_back_intra_pred_mode(H264Context *h);
664 void ff_h264_hl_decode_mb(H264Context *h);
665 int ff_h264_frame_start(H264Context *h);
666 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
667 av_cold int ff_h264_decode_end(AVCodecContext *avctx);
668 av_cold void ff_h264_decode_init_vlc(void);
669
670 /**
671  * decodes a macroblock
672  * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
673  */
674 int ff_h264_decode_mb_cavlc(H264Context *h);
675
676 /**
677  * decodes a CABAC coded macroblock
678  * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
679  */
680 int ff_h264_decode_mb_cabac(H264Context *h);
681
682 void ff_h264_init_cabac_states(H264Context *h);
683
684 void ff_h264_direct_dist_scale_factor(H264Context * const h);
685 void ff_h264_direct_ref_list_init(H264Context * const h);
686 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
687
688 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);
689 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);
690
691 /**
692  * Reset SEI values at the beginning of the frame.
693  *
694  * @param h H.264 context.
695  */
696 void ff_h264_reset_sei(H264Context *h);
697
698
699 /*
700 o-o o-o
701  / / /
702 o-o o-o
703  ,---'
704 o-o o-o
705  / / /
706 o-o o-o
707 */
708 //This table must be here because scan8[constant] must be known at compiletime
709 static const uint8_t scan8[16 + 2*4]={
710  4+1*8, 5+1*8, 4+2*8, 5+2*8,
711  6+1*8, 7+1*8, 6+2*8, 7+2*8,
712  4+3*8, 5+3*8, 4+4*8, 5+4*8,
713  6+3*8, 7+3*8, 6+4*8, 7+4*8,
714  1+1*8, 2+1*8,
715  1+2*8, 2+2*8,
716  1+4*8, 2+4*8,
717  1+5*8, 2+5*8,
718 };
719
720 static av_always_inline uint32_t pack16to32(int a, int b){
721 #if HAVE_BIGENDIAN
722    return (b&0xFFFF) + (a<<16);
723 #else
724    return (a&0xFFFF) + (b<<16);
725 #endif
726 }
727
728 /**
729  * gets the chroma qp.
730  */
731 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
732     return h->pps.chroma_qp_table[t][qscale];
733 }
734
735 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
736
737 static void fill_decode_caches(H264Context *h, int mb_type){
738     MpegEncContext * const s = &h->s;
739     const int mb_xy= h->mb_xy;
740     int topleft_xy, top_xy, topright_xy, left_xy[2];
741     int topleft_type, top_type, topright_type, left_type[2];
742     const uint8_t * left_block;
743     int topleft_partition= -1;
744     int i;
745     static const uint8_t left_block_options[4][16]={
746         {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},
747         {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},
748         {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},
749         {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}
750     };
751
752     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
753
754     /* Wow, what a mess, why didn't they simplify the interlacing & intra
755      * stuff, I can't imagine that these complex rules are worth it. */
756
757     topleft_xy = top_xy - 1;
758     topright_xy= top_xy + 1;
759     left_xy[1] = left_xy[0] = mb_xy-1;
760     left_block = left_block_options[0];
761     if(FRAME_MBAFF){
762         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
763         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
764         if(s->mb_y&1){
765             if (left_mb_field_flag != curr_mb_field_flag) {
766                 left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
767                 if (curr_mb_field_flag) {
768                     left_xy[1] += s->mb_stride;
769                     left_block = left_block_options[3];
770                 } else {
771                     topleft_xy += s->mb_stride;
772                     // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
773                     topleft_partition = 0;
774                     left_block = left_block_options[1];
775                 }
776             }
777         }else{
778             if(curr_mb_field_flag){
779                 topleft_xy  += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
780                 topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
781                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
782             }
783             if (left_mb_field_flag != curr_mb_field_flag) {
784                 left_xy[1] = left_xy[0] = mb_xy - 1;
785                 if (curr_mb_field_flag) {
786                     left_xy[1] += s->mb_stride;
787                     left_block = left_block_options[3];
788                 } else {
789                     left_block = left_block_options[2];
790                 }
791             }
792         }
793     }
794
795     h->top_mb_xy = top_xy;
796     h->left_mb_xy[0] = left_xy[0];
797     h->left_mb_xy[1] = left_xy[1];
798         topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
799         top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
800         topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
801         left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
802         left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
803
804         if(IS_INTRA(mb_type)){
805             int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
806             h->topleft_samples_available=
807             h->top_samples_available=
808             h->left_samples_available= 0xFFFF;
809             h->topright_samples_available= 0xEEEA;
810
811             if(!(top_type & type_mask)){
812                 h->topleft_samples_available= 0xB3FF;
813                 h->top_samples_available= 0x33FF;
814                 h->topright_samples_available= 0x26EA;
815             }
816             if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
817                 if(IS_INTERLACED(mb_type)){
818                     if(!(left_type[0] & type_mask)){
819                         h->topleft_samples_available&= 0xDFFF;
820                         h->left_samples_available&= 0x5FFF;
821                     }
822                     if(!(left_type[1] & type_mask)){
823                         h->topleft_samples_available&= 0xFF5F;
824                         h->left_samples_available&= 0xFF5F;
825                     }
826                 }else{
827                     int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
828                                     ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
829                     assert(left_xy[0] == left_xy[1]);
830                     if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
831                         h->topleft_samples_available&= 0xDF5F;
832                         h->left_samples_available&= 0x5F5F;
833                     }
834                 }
835             }else{
836                 if(!(left_type[0] & type_mask)){
837                     h->topleft_samples_available&= 0xDF5F;
838                     h->left_samples_available&= 0x5F5F;
839                 }
840             }
841
842             if(!(topleft_type & type_mask))
843                 h->topleft_samples_available&= 0x7FFF;
844
845             if(!(topright_type & type_mask))
846                 h->topright_samples_available&= 0xFBFF;
847
848             if(IS_INTRA4x4(mb_type)){
849                 if(IS_INTRA4x4(top_type)){
850                     h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
851                     h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
852                     h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
853                     h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
854                 }else{
855                     int pred;
856                     if(!(top_type & type_mask))
857                         pred= -1;
858                     else{
859                         pred= 2;
860                     }
861                     h->intra4x4_pred_mode_cache[4+8*0]=
862                     h->intra4x4_pred_mode_cache[5+8*0]=
863                     h->intra4x4_pred_mode_cache[6+8*0]=
864                     h->intra4x4_pred_mode_cache[7+8*0]= pred;
865                 }
866                 for(i=0; i<2; i++){
867                     if(IS_INTRA4x4(left_type[i])){
868                         h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
869                         h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
870                     }else{
871                         int pred;
872                         if(!(left_type[i] & type_mask))
873                             pred= -1;
874                         else{
875                             pred= 2;
876                         }
877                         h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
878                         h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
879                     }
880                 }
881             }
882         }
883
884
885 /*
886 0 . T T. T T T T
887 1 L . .L . . . .
888 2 L . .L . . . .
889 3 . T TL . . . .
890 4 L . .L . . . .
891 5 L . .. . . . .
892 */
893 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
894     if(top_type){
895         *(uint32_t*)&h->non_zero_count_cache[4+8*0]= *(uint32_t*)&h->non_zero_count[top_xy][4+3*8];
896             h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
897             h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
898
899             h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
900             h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
901     }else {
902             h->non_zero_count_cache[1+8*0]=
903             h->non_zero_count_cache[2+8*0]=
904
905             h->non_zero_count_cache[1+8*3]=
906             h->non_zero_count_cache[2+8*3]=
907             *(uint32_t*)&h->non_zero_count_cache[4+8*0]= CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040;
908     }
909
910     for (i=0; i<2; i++) {
911         if(left_type[i]){
912             h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
913             h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
914                 h->non_zero_count_cache[0+8*1 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
915                 h->non_zero_count_cache[0+8*4 +   8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
916         }else{
917                 h->non_zero_count_cache[3+8*1 + 2*8*i]=
918                 h->non_zero_count_cache[3+8*2 + 2*8*i]=
919                 h->non_zero_count_cache[0+8*1 +   8*i]=
920                 h->non_zero_count_cache[0+8*4 +   8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
921         }
922     }
923
924     if( CABAC ) {
925         // top_cbp
926         if(top_type) {
927             h->top_cbp = h->cbp_table[top_xy];
928         } else if(IS_INTRA(mb_type)) {
929             h->top_cbp = 0x1C0;
930         } else {
931             h->top_cbp = 0;
932         }
933         // left_cbp
934         if (left_type[0]) {
935             h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
936         } else if(IS_INTRA(mb_type)) {
937             h->left_cbp = 0x1C0;
938         } else {
939             h->left_cbp = 0;
940         }
941         if (left_type[0]) {
942             h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
943         }
944         if (left_type[1]) {
945             h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
946         }
947     }
948
949 #if 1
950     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
951         int list;
952         for(list=0; list<h->list_count; list++){
953             if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type)){
954                 /*if(!h->mv_cache_clean[list]){
955                     memset(h->mv_cache [list],  0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
956                     memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
957                     h->mv_cache_clean[list]= 1;
958                 }*/
959                 continue;
960             }
961             h->mv_cache_clean[list]= 0;
962
963             if(USES_LIST(top_type, list)){
964                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
965                 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
966                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
967                     h->ref_cache[list][scan8[0] + 0 - 1*8]=
968                     h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
969                     h->ref_cache[list][scan8[0] + 2 - 1*8]=
970                     h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
971             }else{
972                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
973                 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
974             }
975
976             for(i=0; i<2; i++){
977                 int cache_idx = scan8[0] - 1 + i*2*8;
978                 if(USES_LIST(left_type[i], list)){
979                     const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
980                     const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
981                     *(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]];
982                     *(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]];
983                         h->ref_cache[list][cache_idx  ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
984                         h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
985                 }else{
986                     *(uint32_t*)h->mv_cache [list][cache_idx  ]=
987                     *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
988                     h->ref_cache[list][cache_idx  ]=
989                     h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
990                 }
991             }
992
993             if((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF)
994                 continue;
995
996             if(USES_LIST(topleft_type, list)){
997                 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
998                 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
999                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
1000                 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1001             }else{
1002                 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
1003                 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1004             }
1005
1006             if(USES_LIST(topright_type, list)){
1007                 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1008                 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
1009                 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
1010                 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1011             }else{
1012                 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
1013                 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1014             }
1015
1016             if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
1017                 continue;
1018
1019             h->ref_cache[list][scan8[5 ]+1] =
1020             h->ref_cache[list][scan8[7 ]+1] =
1021             h->ref_cache[list][scan8[13]+1] =  //FIXME remove past 3 (init somewhere else)
1022             h->ref_cache[list][scan8[4 ]] =
1023             h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1024             *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
1025             *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
1026             *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
1027             *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
1028             *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
1029
1030             if( CABAC ) {
1031                 /* XXX beurk, Load mvd */
1032                 if(USES_LIST(top_type, list)){
1033                     const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1034                     AV_COPY128(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1035                 }else{
1036                     AV_ZERO128(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1037                 }
1038                 if(USES_LIST(left_type[0], list)){
1039                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1040                     *(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]];
1041                     *(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]];
1042                 }else{
1043                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
1044                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
1045                 }
1046                 if(USES_LIST(left_type[1], list)){
1047                     const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
1048                     *(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]];
1049                     *(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]];
1050                 }else{
1051                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
1052                     *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
1053                 }
1054                 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
1055                 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
1056                 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
1057                 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
1058                 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
1059
1060                 if(h->slice_type_nos == FF_B_TYPE){
1061                     fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
1062
1063                     if(IS_DIRECT(top_type)){
1064                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
1065                     }else if(IS_8X8(top_type)){
1066                         int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
1067                         h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
1068                         h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
1069                     }else{
1070                         *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
1071                     }
1072
1073                     if(IS_DIRECT(left_type[0]))
1074                         h->direct_cache[scan8[0] - 1 + 0*8]= 1;
1075                     else if(IS_8X8(left_type[0]))
1076                         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)];
1077                     else
1078                         h->direct_cache[scan8[0] - 1 + 0*8]= 0;
1079
1080                     if(IS_DIRECT(left_type[1]))
1081                         h->direct_cache[scan8[0] - 1 + 2*8]= 1;
1082                     else if(IS_8X8(left_type[1]))
1083                         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)];
1084                     else
1085                         h->direct_cache[scan8[0] - 1 + 2*8]= 0;
1086                 }
1087             }
1088
1089             if(FRAME_MBAFF){
1090 #define MAP_MVS\
1091                     MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1092                     MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1093                     MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1094                     MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1095                     MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1096                     MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1097                     MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1098                     MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1099                     MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1100                     MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1101                 if(MB_FIELD){
1102 #define MAP_F2F(idx, mb_type)\
1103                     if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1104                         h->ref_cache[list][idx] <<= 1;\
1105                         h->mv_cache[list][idx][1] /= 2;\
1106                         h->mvd_cache[list][idx][1] /= 2;\
1107                     }
1108                     MAP_MVS
1109 #undef MAP_F2F
1110                 }else{
1111 #define MAP_F2F(idx, mb_type)\
1112                     if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1113                         h->ref_cache[list][idx] >>= 1;\
1114                         h->mv_cache[list][idx][1] <<= 1;\
1115                         h->mvd_cache[list][idx][1] <<= 1;\
1116                     }
1117                     MAP_MVS
1118 #undef MAP_F2F
1119                 }
1120             }
1121         }
1122     }
1123 #endif
1124
1125         h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1126 }
1127
1128 /**
1129  *
1130  * @returns non zero if the loop filter can be skiped
1131  */
1132 static int fill_filter_caches(H264Context *h, int mb_type){
1133     MpegEncContext * const s = &h->s;
1134     const int mb_xy= h->mb_xy;
1135     int top_xy, left_xy[2];
1136     int top_type, left_type[2];
1137     int i;
1138
1139     top_xy     = mb_xy  - (s->mb_stride << MB_FIELD);
1140
1141     //FIXME deblocking could skip the intra and nnz parts.
1142
1143     /* Wow, what a mess, why didn't they simplify the interlacing & intra
1144      * stuff, I can't imagine that these complex rules are worth it. */
1145
1146     left_xy[1] = left_xy[0] = mb_xy-1;
1147     if(FRAME_MBAFF){
1148         const int left_mb_field_flag     = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
1149         const int curr_mb_field_flag     = IS_INTERLACED(mb_type);
1150         if(s->mb_y&1){
1151             if (left_mb_field_flag != curr_mb_field_flag) {
1152                 left_xy[0] -= s->mb_stride;
1153             }
1154         }else{
1155             if(curr_mb_field_flag){
1156                 top_xy      += s->mb_stride & (((s->current_picture.mb_type[top_xy    ]>>7)&1)-1);
1157             }
1158             if (left_mb_field_flag != curr_mb_field_flag) {
1159                 left_xy[1] += s->mb_stride;
1160             }
1161         }
1162     }
1163
1164     h->top_mb_xy = top_xy;
1165     h->left_mb_xy[0] = left_xy[0];
1166     h->left_mb_xy[1] = left_xy[1];
1167     {
1168         //for sufficiently low qp, filtering wouldn't do anything
1169         //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
1170         int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
1171         int qp = s->current_picture.qscale_table[mb_xy];
1172         if(qp <= qp_thresh
1173            && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
1174            && (top_xy   < 0 || ((qp + s->current_picture.qscale_table[top_xy    ] + 1)>>1) <= qp_thresh)){
1175             if(!FRAME_MBAFF)
1176                 return 1;
1177             if(   (left_xy[0]< 0            || ((qp + s->current_picture.qscale_table[left_xy[1]             ] + 1)>>1) <= qp_thresh)
1178                && (top_xy    < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy    -s->mb_stride] + 1)>>1) <= qp_thresh))
1179                 return 1;
1180         }
1181     }
1182
1183     if(h->deblocking_filter == 2){
1184         h->top_type    = top_type     = h->slice_table[top_xy     ] == h->slice_num ? s->current_picture.mb_type[top_xy]     : 0;
1185         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;
1186         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;
1187     }else{
1188         h->top_type    = top_type     = h->slice_table[top_xy     ] < 0xFFFF ? s->current_picture.mb_type[top_xy]     : 0;
1189         h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
1190         h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
1191     }
1192     if(IS_INTRA(mb_type))
1193         return 0;
1194
1195     AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
1196     AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
1197     *((uint32_t*)&h->non_zero_count_cache[0+8*5])= *((uint32_t*)&h->non_zero_count[mb_xy][16]);
1198     *((uint32_t*)&h->non_zero_count_cache[4+8*3])= *((uint32_t*)&h->non_zero_count[mb_xy][20]);
1199     AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
1200
1201     h->cbp= h->cbp_table[mb_xy];
1202
1203     {
1204         int list;
1205         for(list=0; list<h->list_count; list++){
1206             int8_t *ref;
1207             int y, b_stride;
1208             int16_t (*mv_dst)[2];
1209             int16_t (*mv_src)[2];
1210
1211             if(!USES_LIST(mb_type, list)){
1212                 fill_rectangle(  h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
1213                 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
1214                 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] =
1215                 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
1216                 *(uint32_t*)&h->ref_cache[list][scan8[10]] = ((LIST_NOT_USED)&0xFF)*0x01010101;
1217                 continue;
1218             }
1219
1220             ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
1221             {
1222                 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1223                 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
1224                 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
1225                 ref += h->b8_stride;
1226                 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
1227                 *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
1228             }
1229
1230             b_stride = h->b_stride;
1231             mv_dst   = &h->mv_cache[list][scan8[0]];
1232             mv_src   = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
1233             for(y=0; y<4; y++){
1234                 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
1235             }
1236
1237         }
1238     }
1239
1240
1241 /*
1242 0 . T T. T T T T
1243 1 L . .L . . . .
1244 2 L . .L . . . .
1245 3 . T TL . . . .
1246 4 L . .L . . . .
1247 5 L . .. . . . .
1248 */
1249 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
1250     if(top_type){
1251         *(uint32_t*)&h->non_zero_count_cache[4+8*0]= *(uint32_t*)&h->non_zero_count[top_xy][4+3*8];
1252     }
1253
1254     if(left_type[0]){
1255         h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
1256         h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
1257         h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
1258         h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
1259     }
1260
1261     // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
1262     if(!CABAC && h->pps.transform_8x8_mode){
1263         if(IS_8x8DCT(top_type)){
1264             h->non_zero_count_cache[4+8*0]=
1265             h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
1266             h->non_zero_count_cache[6+8*0]=
1267             h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
1268         }
1269         if(IS_8x8DCT(left_type[0])){
1270             h->non_zero_count_cache[3+8*1]=
1271             h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
1272         }
1273         if(IS_8x8DCT(left_type[1])){
1274             h->non_zero_count_cache[3+8*3]=
1275             h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
1276         }
1277
1278         if(IS_8x8DCT(mb_type)){
1279             h->non_zero_count_cache[scan8[0   ]]= h->non_zero_count_cache[scan8[1   ]]=
1280             h->non_zero_count_cache[scan8[2   ]]= h->non_zero_count_cache[scan8[3   ]]= h->cbp & 1;
1281
1282             h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
1283             h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
1284
1285             h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
1286             h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
1287
1288             h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
1289             h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
1290         }
1291     }
1292
1293     if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
1294         int list;
1295         for(list=0; list<h->list_count; list++){
1296             if(USES_LIST(top_type, list)){
1297                 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1298                 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
1299                 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1300                 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1301                 h->ref_cache[list][scan8[0] + 0 - 1*8]=
1302                 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
1303                 h->ref_cache[list][scan8[0] + 2 - 1*8]=
1304                 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
1305             }else{
1306                 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1307                 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((LIST_NOT_USED)&0xFF)*0x01010101;
1308             }
1309
1310             if(!IS_INTERLACED(mb_type^left_type[0])){
1311                 if(USES_LIST(left_type[0], list)){
1312                     const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1313                     const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
1314                     int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1315                     *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 0 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*0];
1316                     *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 8 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*1];
1317                     *(uint32_t*)h->mv_cache[list][scan8[0] - 1 +16 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*2];
1318                     *(uint32_t*)h->mv_cache[list][scan8[0] - 1 +24 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*3];
1319                     h->ref_cache[list][scan8[0] - 1 + 0 ]=
1320                     h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + h->b8_stride*0]];
1321                     h->ref_cache[list][scan8[0] - 1 +16 ]=
1322                     h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + h->b8_stride*1]];
1323                 }else{
1324                     *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0 ]=
1325                     *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 8 ]=
1326                     *(uint32_t*)h->mv_cache [list][scan8[0] - 1 +16 ]=
1327                     *(uint32_t*)h->mv_cache [list][scan8[0] - 1 +24 ]= 0;
1328                     h->ref_cache[list][scan8[0] - 1 + 0  ]=
1329                     h->ref_cache[list][scan8[0] - 1 + 8  ]=
1330                     h->ref_cache[list][scan8[0] - 1 + 16 ]=
1331                     h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
1332                 }
1333             }
1334         }
1335     }
1336
1337     return 0;
1338 }
1339
1340 /**
1341  * gets the predicted intra4x4 prediction mode.
1342  */
1343 static inline int pred_intra_mode(H264Context *h, int n){
1344     const int index8= scan8[n];
1345     const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1346     const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1347     const int min= FFMIN(left, top);
1348
1349     tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1350
1351     if(min<0) return DC_PRED;
1352     else      return min;
1353 }
1354
1355 static inline void write_back_non_zero_count(H264Context *h){
1356     const int mb_xy= h->mb_xy;
1357
1358     AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1359     AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1360     *((uint32_t*)&h->non_zero_count[mb_xy][16]) = *((uint32_t*)&h->non_zero_count_cache[0+8*5]);
1361     *((uint32_t*)&h->non_zero_count[mb_xy][20]) = *((uint32_t*)&h->non_zero_count_cache[4+8*3]);
1362     AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1363 }
1364
1365 static inline void write_back_motion(H264Context *h, int mb_type){
1366     MpegEncContext * const s = &h->s;
1367     const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1368     const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1369     int list;
1370
1371     if(!USES_LIST(mb_type, 0))
1372         fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1373
1374     for(list=0; list<h->list_count; list++){
1375         int y, b_stride;
1376         int16_t (*mv_dst)[2];
1377         int16_t (*mv_src)[2];
1378
1379         if(!USES_LIST(mb_type, list))
1380             continue;
1381
1382         b_stride = h->b_stride;
1383         mv_dst   = &s->current_picture.motion_val[list][b_xy];
1384         mv_src   = &h->mv_cache[list][scan8[0]];
1385         for(y=0; y<4; y++){
1386             AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1387         }
1388         if( CABAC ) {
1389             int16_t (*mvd_dst)[2] = &h->mvd_table[list][b_xy];
1390             int16_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1391             if(IS_SKIP(mb_type))
1392                 fill_rectangle(mvd_dst, 4, 4, h->b_stride, 0, 4);
1393             else
1394             for(y=0; y<4; y++){
1395                 AV_COPY128(mvd_dst + y*b_stride, mvd_src + 8*y);
1396             }
1397         }
1398
1399         {
1400             int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1401             ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1402             ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1403             ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1404             ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1405         }
1406     }
1407
1408     if(h->slice_type_nos == FF_B_TYPE && CABAC){
1409         if(IS_8X8(mb_type)){
1410             uint8_t *direct_table = &h->direct_table[b8_xy];
1411             direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1412             direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1413             direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1414         }
1415     }
1416 }
1417
1418 static inline int get_dct8x8_allowed(H264Context *h){
1419     if(h->sps.direct_8x8_inference_flag)
1420         return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8                )*0x0001000100010001ULL));
1421     else
1422         return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1423 }
1424
1425 static void predict_field_decoding_flag(H264Context *h){
1426     MpegEncContext * const s = &h->s;
1427     const int mb_xy= h->mb_xy;
1428     int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
1429                 ? s->current_picture.mb_type[mb_xy-1]
1430                 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
1431                 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
1432                 : 0;
1433     h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
1434 }
1435
1436 /**
1437  * decodes a P_SKIP or B_SKIP macroblock
1438  */
1439 static void decode_mb_skip(H264Context *h){
1440     MpegEncContext * const s = &h->s;
1441     const int mb_xy= h->mb_xy;
1442     int mb_type=0;
1443
1444     memset(h->non_zero_count[mb_xy], 0, 32);
1445     memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1446
1447     if(MB_FIELD)
1448         mb_type|= MB_TYPE_INTERLACED;
1449
1450     if( h->slice_type_nos == FF_B_TYPE )
1451     {
1452         // just for fill_caches. pred_direct_motion will set the real mb_type
1453         mb_type|= MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1454
1455         fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1456         ff_h264_pred_direct_motion(h, &mb_type);
1457         mb_type|= MB_TYPE_SKIP;
1458     }
1459     else
1460     {
1461         int mx, my;
1462         mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1463
1464         fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1465         pred_pskip_motion(h, &mx, &my);
1466         fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1467         fill_rectangle(  h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1468     }
1469
1470     write_back_motion(h, mb_type);
1471     s->current_picture.mb_type[mb_xy]= mb_type;
1472     s->current_picture.qscale_table[mb_xy]= s->qscale;
1473     h->slice_table[ mb_xy ]= h->slice_num;
1474     h->prev_mb_skipped= 1;
1475 }
1476
1477 #include "h264_mvpred.h" //For pred_pskip_motion()
1478
1479 #endif /* AVCODEC_H264_H */