FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
h264.h
Go to the documentation of this file.
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
24  * H.264 / AVC / MPEG4 part10 codec.
25  * @author Michael Niedermayer <michaelni@gmx.at>
26  */
27 
28 #ifndef AVCODEC_H264_H
29 #define AVCODEC_H264_H
30 
31 #include "libavutil/intreadwrite.h"
32 #include "cabac.h"
33 #include "get_bits.h"
34 #include "mpegvideo.h"
35 #include "h264chroma.h"
36 #include "h264dsp.h"
37 #include "h264pred.h"
38 #include "h264qpel.h"
39 #include "rectangle.h"
40 
41 #define MAX_SPS_COUNT 32
42 #define MAX_PPS_COUNT 256
43 
44 #define MAX_MMCO_COUNT 66
45 
46 #define MAX_DELAYED_PIC_COUNT 16
47 
48 #define MAX_MBPAIR_SIZE (256*1024) // a tighter bound could be calculated if someone cares about a few bytes
49 
50 /* Compiling in interlaced support reduces the speed
51  * of progressive decoding by about 2%. */
52 #define ALLOW_INTERLACE
53 
54 #define FMO 0
55 
56 /**
57  * The maximum number of slices supported by the decoder.
58  * must be a power of 2
59  */
60 #define MAX_SLICES 16
61 
62 #ifdef ALLOW_INTERLACE
63 #define MB_MBAFF h->mb_mbaff
64 #define MB_FIELD h->mb_field_decoding_flag
65 #define FRAME_MBAFF h->mb_aff_frame
66 #define FIELD_PICTURE (h->picture_structure != PICT_FRAME)
67 #define LEFT_MBS 2
68 #define LTOP 0
69 #define LBOT 1
70 #define LEFT(i) (i)
71 #else
72 #define MB_MBAFF 0
73 #define MB_FIELD 0
74 #define FRAME_MBAFF 0
75 #define FIELD_PICTURE 0
76 #undef IS_INTERLACED
77 #define IS_INTERLACED(mb_type) 0
78 #define LEFT_MBS 1
79 #define LTOP 0
80 #define LBOT 0
81 #define LEFT(i) 0
82 #endif
83 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
84 
85 #ifndef CABAC
86 #define CABAC h->pps.cabac
87 #endif
88 
89 #define CHROMA (h->sps.chroma_format_idc)
90 #define CHROMA422 (h->sps.chroma_format_idc == 2)
91 #define CHROMA444 (h->sps.chroma_format_idc == 3)
92 
93 #define EXTENDED_SAR 255
94 
95 #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit
96 #define MB_TYPE_8x8DCT 0x01000000
97 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
98 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
99 
100 #define QP_MAX_NUM (51 + 6*6) // The maximum supported qp
101 
102 /* NAL unit types */
103 enum {
118  NAL_FF_IGNORE = 0xff0f001,
119 };
120 
121 /**
122  * SEI message types
123  */
124 typedef enum {
125  SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
126  SEI_TYPE_PIC_TIMING = 1, ///< picture timing
127  SEI_TYPE_USER_DATA_ITU_T_T35 = 4, ///< user data registered by ITU-T Recommendation T.35
128  SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
129  SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync)
130 } SEI_Type;
131 
132 /**
133  * pic_struct in picture timing SEI message
134  */
135 typedef enum {
136  SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
137  SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
138  SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
139  SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
140  SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
141  SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
142  SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
143  SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
144  SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
146 
147 /**
148  * Sequence parameter set
149  */
150 typedef struct SPS {
154  int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
155  int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
156  int poc_type; ///< pic_order_cnt_type
157  int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
161  int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
162  int ref_frame_count; ///< num_ref_frames
164  int mb_width; ///< pic_width_in_mbs_minus1 + 1
165  int mb_height; ///< pic_height_in_map_units_minus1 + 1
167  int mb_aff; ///< mb_adaptive_frame_field_flag
169  int crop; ///< frame_cropping_flag
170  unsigned int crop_left; ///< frame_cropping_rect_left_offset
171  unsigned int crop_right; ///< frame_cropping_rect_right_offset
172  unsigned int crop_top; ///< frame_cropping_rect_top_offset
173  unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
184  uint32_t time_scale;
186  short offset_for_ref_frame[256]; // FIXME dyn aloc?
196  int cpb_cnt; ///< See H.264 E.1.2
197  int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
198  int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
199  int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
200  int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
201  int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
202  int residual_color_transform_flag; ///< residual_colour_transform_flag
203  int constraint_set_flags; ///< constraint_set[0-3]_flag
204  int new; ///< flag to keep track if the decoder context needs re-init due to changed SPS
205 } SPS;
206 
207 /**
208  * Picture parameter set
209  */
210 typedef struct PPS {
211  unsigned int sps_id;
212  int cabac; ///< entropy_coding_mode_flag
213  int pic_order_present; ///< pic_order_present_flag
214  int slice_group_count; ///< num_slice_groups_minus1 + 1
216  unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
217  int weighted_pred; ///< weighted_pred_flag
219  int init_qp; ///< pic_init_qp_minus26 + 26
220  int init_qs; ///< pic_init_qs_minus26 + 26
222  int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
223  int constrained_intra_pred; ///< constrained_intra_pred_flag
224  int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
225  int transform_8x8_mode; ///< transform_8x8_mode_flag
228  uint8_t chroma_qp_table[2][QP_MAX_NUM+1]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
230 } PPS;
231 
232 /**
233  * Memory management control operation opcode.
234  */
235 typedef enum MMCOOpcode {
236  MMCO_END = 0,
243 } MMCOOpcode;
244 
245 /**
246  * Memory management control operation.
247  */
248 typedef struct MMCO {
250  int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
251  int long_arg; ///< index, pic_num, or num long refs depending on opcode
252 } MMCO;
253 
254 /**
255  * H264Context
256  */
257 typedef struct H264Context {
268 
274 
275  int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
276  int chroma_qp[2]; // QPc
277 
278  int qp_thresh; ///< QP threshold to skip loopfilter
279 
280  int width, height;
283 
284  int qscale;
289 
291  int flags;
293 
296 
297  // prediction stuff
300 
305 
307  int top_type;
310 
313 
318  unsigned int top_samples_available;
321  uint8_t (*top_borders[2])[(16 * 3) * 2];
322 
323  /**
324  * non zero coeff count cache.
325  * is 64 if not available.
326  */
328 
330 
331  /**
332  * Motion vector cache.
333  */
334  DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
335  DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
336 #define LIST_NOT_USED -1 // FIXME rename?
337 #define PART_NOT_AVAILABLE -2
338 
339  /**
340  * number of neighbors (top and/or left) that used 8x8 dct
341  */
343 
344  /**
345  * block_offset[ 0..23] for frame macroblocks
346  * block_offset[24..47] for field macroblocks
347  */
348  int block_offset[2 * (16 * 3)];
349 
350  uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
351  uint32_t *mb2br_xy;
352  int b_stride; // FIXME use s->b4_stride
353 
354  int mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
356 
357  unsigned current_sps_id; ///< id of the current SPS
358  SPS sps; ///< current sps
359 
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][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
366  uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
367  uint32_t(*dequant4_coeff[6])[16];
368  uint32_t(*dequant8_coeff[6])[64];
369 
371  uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
373  int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
375 
376  // interlacing specific flags
379  int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
382 
383  DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
384 
385  // Weighted pred stuff
390  // The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
391  int luma_weight[48][2][2];
392  int chroma_weight[48][2][2][2];
393  int implicit_weight[48][48][2];
394 
400  int map_col_to_list0[2][16 + 32];
401  int map_col_to_list0_field[2][2][16 + 32];
402 
403  /**
404  * num_ref_idx_l0/1_active_minus1 + 1
405  */
406  unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
407  unsigned int list_count;
408  uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
409  Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
410  * Reordered version of default_ref_list
411  * according to picture reordering in slice header */
412  int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
413 
414  // data partitioning
419 
421  DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2]; ///< as a dct coeffecient is int32_t in high depth, we need to reserve twice the space.
422  DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
423  int16_t mb_padding[256 * 2]; ///< 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
424 
425  /**
426  * Cabac
427  */
430 
431  /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
432  uint16_t *cbp_table;
433  int cbp;
434  int top_cbp;
435  int left_cbp;
436  /* chroma_pred_mode for i4x4 or i16x16, else 0 */
439  uint8_t (*mvd_table[2])[2];
440  DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
443 
456 
458 
459  int mb_x, mb_y;
465  int mb_num;
466  int mb_xy;
467 
469 
470  // deblock
471  int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
474 
475  // =============================================================
476  // Things below are not used in the MB or more inner code
477 
481  unsigned int rbsp_buffer_size[2];
482 
483  /**
484  * Used to parse AVC variant of h264
485  */
486  int is_avc; ///< this flag is != 0 if codec is avc1
487  int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
488  int got_first; ///< this flag is != 0 if we've parsed a frame
489 
490  int bit_depth_luma; ///< luma bit depth from sps to detect changes
491  int chroma_format_idc; ///< chroma format from sps to detect changes
492 
495 
496  int dequant_coeff_pps; ///< reinit tables when pps changes
497 
498  uint16_t *slice_table_base;
499 
500  // POC stuff
501  int poc_lsb;
502  int poc_msb;
504  int delta_poc[2];
506  int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
507  int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
508  int frame_num_offset; ///< for POC type 2
509  int prev_frame_num_offset; ///< for POC type 2
510  int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
511 
512  /**
513  * frame_num for frames or 2 * frame_num + 1 for field pics.
514  */
516 
517  /**
518  * max_frame_num or 2 * max_frame_num for field pics.
519  */
521 
523 
524  Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
532 
533  /**
534  * memory management control operations buffer.
535  */
539 
540  int long_ref_count; ///< number of actual long term references
541  int short_ref_count; ///< number of actual short term references
542 
544 
545  /**
546  * @name Members for slice based multithreading
547  * @{
548  */
550 
551  /**
552  * current slice number, used to initialize slice_num of each thread/context
553  */
555 
556  /**
557  * Max number of threads / contexts.
558  * This is equal to AVCodecContext.thread_count unless
559  * multithreaded decoding is impossible, in which case it is
560  * reduced to 1.
561  */
563 
565 
566  /**
567  * 1 if the single thread fallback warning has already been
568  * displayed, 0 otherwise.
569  */
571 
573 
575  unsigned int last_ref_count[2];
576  /** @} */
577 
578  /**
579  * pic_struct in picture timing SEI message
580  */
582 
583  /**
584  * Complement sei_pic_struct
585  * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
586  * However, soft telecined frames may have these values.
587  * This is used in an attempt to flag soft telecine progressive.
588  */
590 
591  /**
592  * Bit set of clock types for fields/frames in picture timing SEI message.
593  * For each found ct_type, appropriate bit is set (e.g., bit 1 for
594  * interlaced).
595  */
597 
598  /**
599  * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
600  */
602 
603  /**
604  * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
605  */
607 
608  /**
609  * recovery_frame_cnt from SEI message
610  *
611  * Set to -1 if no recovery point SEI message found or to number of frames
612  * before playback synchronizes. Frames having recovery point are key
613  * frames.
614  */
616  /**
617  * recovery_frame is the frame_num at which the next frame should
618  * be fully constructed.
619  *
620  * Set to -1 when not expecting a recovery point.
621  */
623 
624  /**
625  * Are the SEI recovery points looking valid.
626  */
628 
629  int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
630  int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
631 
632  // Timestamp stuff
633  int sei_buffering_period_present; ///< Buffering period SEI flag
634  int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
635 
638 
639  int16_t slice_row[MAX_SLICES]; ///< to detect when MAX_SLICES is too low
640 
641  int sync; ///< did we had a keyframe or recovery point
642 
647  int16_t *dc_val_base;
648 
649  uint8_t *visualization_buffer[3]; ///< temporary buffer vor MV visualization
650 } H264Context;
651 
652 extern const uint8_t ff_h264_chroma_qp[7][QP_MAX_NUM + 1]; ///< One chroma qp table for each possible bit depth (8-14).
653 extern const uint16_t ff_h264_mb_sizes[4];
654 
655 /**
656  * Decode SEI
657  */
659 
660 /**
661  * Decode SPS
662  */
664 
665 /**
666  * compute profile from sps
667  */
669 
670 /**
671  * Decode PPS
672  */
673 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
674 
675 /**
676  * Decode a network abstraction layer unit.
677  * @param consumed is the number of bytes used as input
678  * @param length is the length of the array
679  * @param dst_length is the number of decoded bytes FIXME here
680  * or a decode rbsp tailing?
681  * @return decoded bytes, might be src+1 if no escapes
682  */
683 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
684  int *dst_length, int *consumed, int length);
685 
686 /**
687  * Free any data that may have been allocated in the H264 context
688  * like SPS, PPS etc.
689  */
691 
692 /**
693  * Reconstruct bitstream slice_type.
694  */
695 int ff_h264_get_slice_type(const H264Context *h);
696 
697 /**
698  * Allocate tables.
699  * needs width/height
700  */
702 
703 /**
704  * Fill the default_ref_list.
705  */
707 
711 
712 /**
713  * Execute the reference picture marking (memory management control operations).
714  */
715 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
716 
718  int first_slice);
719 
720 int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
721 
722 /**
723  * Check if the top & left blocks are available if needed & change the
724  * dc mode so it only uses the available blocks.
725  */
727 
728 /**
729  * Check if the top & left blocks are available if needed & change the
730  * dc mode so it only uses the available blocks.
731  */
732 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma);
733 
736 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size);
738 void ff_h264_decode_init_vlc(void);
739 
740 /**
741  * Decode a macroblock
742  * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
743  */
745 
746 /**
747  * Decode a CABAC coded macroblock
748  * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
749  */
751 
753 
756 void ff_h264_pred_direct_motion(H264Context *const h, int *mb_type);
757 
758 void ff_h264_filter_mb_fast(H264Context *h, int mb_x, int mb_y,
759  uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
760  unsigned int linesize, unsigned int uvlinesize);
761 void ff_h264_filter_mb(H264Context *h, int mb_x, int mb_y,
762  uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
763  unsigned int linesize, unsigned int uvlinesize);
764 
765 /**
766  * Reset SEI values at the beginning of the frame.
767  *
768  * @param h H.264 context.
769  */
771 
772 /*
773  * o-o o-o
774  * / / /
775  * o-o o-o
776  * ,---'
777  * o-o o-o
778  * / / /
779  * o-o o-o
780  */
781 
782 /* Scan8 organization:
783  * 0 1 2 3 4 5 6 7
784  * 0 DY y y y y y
785  * 1 y Y Y Y Y
786  * 2 y Y Y Y Y
787  * 3 y Y Y Y Y
788  * 4 y Y Y Y Y
789  * 5 DU u u u u u
790  * 6 u U U U U
791  * 7 u U U U U
792  * 8 u U U U U
793  * 9 u U U U U
794  * 10 DV v v v v v
795  * 11 v V V V V
796  * 12 v V V V V
797  * 13 v V V V V
798  * 14 v V V V V
799  * DY/DU/DV are for luma/chroma DC.
800  */
801 
802 #define LUMA_DC_BLOCK_INDEX 48
803 #define CHROMA_DC_BLOCK_INDEX 49
804 
805 // This table must be here because scan8[constant] must be known at compiletime
806 static const uint8_t scan8[16 * 3 + 3] = {
807  4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
808  6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
809  4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
810  6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
811  4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
812  6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
813  4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
814  6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
815  4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
816  6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
817  4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
818  6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
819  0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
820 };
821 
822 static av_always_inline uint32_t pack16to32(int a, int b)
823 {
824 #if HAVE_BIGENDIAN
825  return (b & 0xFFFF) + (a << 16);
826 #else
827  return (a & 0xFFFF) + (b << 16);
828 #endif
829 }
830 
831 static av_always_inline uint16_t pack8to16(int a, int b)
832 {
833 #if HAVE_BIGENDIAN
834  return (b & 0xFF) + (a << 8);
835 #else
836  return (a & 0xFF) + (b << 8);
837 #endif
838 }
839 
840 /**
841  * Get the chroma qp.
842  */
844 {
845  return h->pps.chroma_qp_table[t][qscale];
846 }
847 
848 /**
849  * Get the predicted intra4x4 prediction mode.
850  */
852 {
853  const int index8 = scan8[n];
854  const int left = h->intra4x4_pred_mode_cache[index8 - 1];
855  const int top = h->intra4x4_pred_mode_cache[index8 - 8];
856  const int min = FFMIN(left, top);
857 
858  tprintf(h->avctx, "mode:%d %d min:%d\n", left, top, min);
859 
860  if (min < 0)
861  return DC_PRED;
862  else
863  return min;
864 }
865 
867 {
868  int8_t *i4x4 = h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
869  int8_t *i4x4_cache = h->intra4x4_pred_mode_cache;
870 
871  AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
872  i4x4[4] = i4x4_cache[7 + 8 * 3];
873  i4x4[5] = i4x4_cache[7 + 8 * 2];
874  i4x4[6] = i4x4_cache[7 + 8 * 1];
875 }
876 
878 {
879  const int mb_xy = h->mb_xy;
880  uint8_t *nnz = h->non_zero_count[mb_xy];
881  uint8_t *nnz_cache = h->non_zero_count_cache;
882 
883  AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
884  AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
885  AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
886  AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
887  AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
888  AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
889  AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
890  AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
891 
892  if (!h->chroma_y_shift) {
893  AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
894  AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
895  AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
896  AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
897  }
898 }
899 
901  int b_stride,
902  int b_xy, int b8_xy,
903  int mb_type, int list)
904 {
905  int16_t(*mv_dst)[2] = &h->cur_pic.f.motion_val[list][b_xy];
906  int16_t(*mv_src)[2] = &h->mv_cache[list][scan8[0]];
907  AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
908  AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
909  AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
910  AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
911  if (CABAC) {
912  uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8 * h->mb_xy
913  : h->mb2br_xy[h->mb_xy]];
914  uint8_t(*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
915  if (IS_SKIP(mb_type)) {
916  AV_ZERO128(mvd_dst);
917  } else {
918  AV_COPY64(mvd_dst, mvd_src + 8 * 3);
919  AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
920  AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
921  AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
922  }
923  }
924 
925  {
926  int8_t *ref_index = &h->cur_pic.f.ref_index[list][b8_xy];
927  int8_t *ref_cache = h->ref_cache[list];
928  ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
929  ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
930  ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
931  ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
932  }
933 }
934 
935 static av_always_inline void write_back_motion(H264Context *h, int mb_type)
936 {
937  const int b_stride = h->b_stride;
938  const int b_xy = 4 * h->mb_x + 4 * h->mb_y * h->b_stride; // try mb2b(8)_xy
939  const int b8_xy = 4 * h->mb_xy;
940 
941  if (USES_LIST(mb_type, 0)) {
942  write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 0);
943  } else {
944  fill_rectangle(&h->cur_pic.f.ref_index[0][b8_xy],
945  2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
946  }
947  if (USES_LIST(mb_type, 1))
948  write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 1);
949 
950  if (h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC) {
951  if (IS_8X8(mb_type)) {
952  uint8_t *direct_table = &h->direct_table[4 * h->mb_xy];
953  direct_table[1] = h->sub_mb_type[1] >> 1;
954  direct_table[2] = h->sub_mb_type[2] >> 1;
955  direct_table[3] = h->sub_mb_type[3] >> 1;
956  }
957  }
958 }
959 
961 {
963  return !(AV_RN64A(h->sub_mb_type) &
965  0x0001000100010001ULL));
966  else
967  return !(AV_RN64A(h->sub_mb_type) &
969  0x0001000100010001ULL));
970 }
971 
972 void ff_h264_draw_horiz_band(H264Context *h, int y, int height);
973 
974 #endif /* AVCODEC_H264_H */