FFmpeg
 All Data Structures Namespaces 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 "error_resilience.h"
34 #include "get_bits.h"
35 #include "mpegvideo.h"
36 #include "h264chroma.h"
37 #include "h264dsp.h"
38 #include "h264pred.h"
39 #include "h264qpel.h"
40 #include "rectangle.h"
41 
42 #define MAX_SPS_COUNT 32
43 #define MAX_PPS_COUNT 256
44 
45 #define MAX_MMCO_COUNT 66
46 
47 #define MAX_DELAYED_PIC_COUNT 16
48 
49 #define MAX_MBPAIR_SIZE (256*1024) // a tighter bound could be calculated if someone cares about a few bytes
50 
51 /* Compiling in interlaced support reduces the speed
52  * of progressive decoding by about 2%. */
53 #define ALLOW_INTERLACE
54 
55 #define FMO 0
56 
57 /**
58  * The maximum number of slices supported by the decoder.
59  * must be a power of 2
60  */
61 #define MAX_SLICES 16
62 
63 #ifdef ALLOW_INTERLACE
64 #define MB_MBAFF(h) h->mb_mbaff
65 #define MB_FIELD(h) h->mb_field_decoding_flag
66 #define FRAME_MBAFF(h) h->mb_aff_frame
67 #define FIELD_PICTURE(h) (h->picture_structure != PICT_FRAME)
68 #define LEFT_MBS 2
69 #define LTOP 0
70 #define LBOT 1
71 #define LEFT(i) (i)
72 #else
73 #define MB_MBAFF(h) 0
74 #define MB_FIELD(h) 0
75 #define FRAME_MBAFF(h) 0
76 #define FIELD_PICTURE(h) 0
77 #undef IS_INTERLACED
78 #define IS_INTERLACED(mb_type) 0
79 #define LEFT_MBS 1
80 #define LTOP 0
81 #define LBOT 0
82 #define LEFT(i) 0
83 #endif
84 #define FIELD_OR_MBAFF_PICTURE(h) (FRAME_MBAFF(h) || FIELD_PICTURE(h))
85 
86 #ifndef CABAC
87 #define CABAC(h) h->pps.cabac
88 #endif
89 
90 #define CHROMA(h) (h->sps.chroma_format_idc)
91 #define CHROMA422(h) (h->sps.chroma_format_idc == 2)
92 #define CHROMA444(h) (h->sps.chroma_format_idc == 3)
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 #define QP_MAX_NUM (51 + 6*6) // The maximum supported qp
102 
103 /* NAL unit types */
104 enum {
106  NAL_DPA = 2,
107  NAL_DPB = 3,
108  NAL_DPC = 4,
110  NAL_SEI = 6,
111  NAL_SPS = 7,
112  NAL_PPS = 8,
113  NAL_AUD = 9,
119  NAL_FF_IGNORE = 0xff0f001,
120 };
121 
122 /**
123  * SEI message types
124  */
125 typedef enum {
126  SEI_TYPE_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
127  SEI_TYPE_PIC_TIMING = 1, ///< picture timing
128  SEI_TYPE_USER_DATA_ITU_T_T35 = 4, ///< user data registered by ITU-T Recommendation T.35
129  SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
130  SEI_TYPE_RECOVERY_POINT = 6, ///< recovery point (frame # to decoder sync)
131  SEI_TYPE_FRAME_PACKING = 45, ///< frame packing arrangement
132 } SEI_Type;
133 
134 /**
135  * pic_struct in picture timing SEI message
136  */
137 typedef enum {
138  SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
139  SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
140  SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
141  SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
142  SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
143  SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
144  SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
145  SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
146  SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
148 
149 /**
150  * frame_packing_arrangement types
151  */
152 typedef enum {
160 } SEI_FpaType;
161 
162 /**
163  * Sequence parameter set
164  */
165 typedef struct SPS {
166  unsigned int sps_id;
170  int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
171  int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
172  int poc_type; ///< pic_order_cnt_type
173  int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
177  int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
178  int ref_frame_count; ///< num_ref_frames
180  int mb_width; ///< pic_width_in_mbs_minus1 + 1
181  int mb_height; ///< pic_height_in_map_units_minus1 + 1
183  int mb_aff; ///< mb_adaptive_frame_field_flag
185  int crop; ///< frame_cropping_flag
186 
187  /* those 4 are already in luma samples */
188  unsigned int crop_left; ///< frame_cropping_rect_left_offset
189  unsigned int crop_right; ///< frame_cropping_rect_right_offset
190  unsigned int crop_top; ///< frame_cropping_rect_top_offset
191  unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
202  uint32_t time_scale;
204  short offset_for_ref_frame[256]; // FIXME dyn aloc?
214  int cpb_cnt; ///< See H.264 E.1.2
215  int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
216  int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
217  int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
218  int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
219  int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
220  int residual_color_transform_flag; ///< residual_colour_transform_flag
221  int constraint_set_flags; ///< constraint_set[0-3]_flag
222  int new; ///< flag to keep track if the decoder context needs re-init due to changed SPS
223 } SPS;
224 
225 /**
226  * Picture parameter set
227  */
228 typedef struct PPS {
229  unsigned int sps_id;
230  int cabac; ///< entropy_coding_mode_flag
231  int pic_order_present; ///< pic_order_present_flag
232  int slice_group_count; ///< num_slice_groups_minus1 + 1
234  unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
235  int weighted_pred; ///< weighted_pred_flag
237  int init_qp; ///< pic_init_qp_minus26 + 26
238  int init_qs; ///< pic_init_qs_minus26 + 26
240  int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
241  int constrained_intra_pred; ///< constrained_intra_pred_flag
242  int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
243  int transform_8x8_mode; ///< transform_8x8_mode_flag
246  uint8_t chroma_qp_table[2][QP_MAX_NUM+1]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
248 } PPS;
249 
250 /**
251  * Frame Packing Arrangement Type
252  */
253 typedef struct FPA {
255  int frame_packing_arrangement_cancel_flag; ///< is previous arrangement canceled, -1 if never received
260 } FPA;
261 
262 /**
263  * Memory management control operation opcode.
264  */
265 typedef enum MMCOOpcode {
266  MMCO_END = 0,
273 } MMCOOpcode;
274 
275 /**
276  * Memory management control operation.
277  */
278 typedef struct MMCO {
280  int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
281  int long_arg; ///< index, pic_num, or num long refs depending on opcode
282 } MMCO;
283 
284 /**
285  * H264Context
286  */
287 typedef struct H264Context {
298 
302 
303  int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
304  int chroma_qp[2]; // QPc
305 
306  int qp_thresh; ///< QP threshold to skip loopfilter
307 
308  /* coded dimensions -- 16 * mb w/h */
309  int width, height;
310  ptrdiff_t linesize, uvlinesize;
312 
313  int qscale;
318 
320  int flags;
322 
325 
326  // prediction stuff
329 
334 
336  int top_type;
339 
342 
347  unsigned int top_samples_available;
350  uint8_t (*top_borders[2])[(16 * 3) * 2];
351 
352  /**
353  * non zero coeff count cache.
354  * is 64 if not available.
355  */
357 
359 
360  /**
361  * Motion vector cache.
362  */
363  DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
364  DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
365 #define LIST_NOT_USED -1 // FIXME rename?
366 #define PART_NOT_AVAILABLE -2
367 
368  /**
369  * number of neighbors (top and/or left) that used 8x8 dct
370  */
372 
373  /**
374  * block_offset[ 0..23] for frame macroblocks
375  * block_offset[24..47] for field macroblocks
376  */
377  int block_offset[2 * (16 * 3)];
378 
379  uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
380  uint32_t *mb2br_xy;
381  int b_stride; // FIXME use s->b4_stride
382 
383  ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
384  ptrdiff_t mb_uvlinesize;
385 
386  unsigned current_sps_id; ///< id of the current SPS
387  SPS sps; ///< current sps
388  PPS pps; ///< current pps
389 
390  int au_pps_id; ///< pps_id of current access unit
391 
392  uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
393  uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
394  uint32_t(*dequant4_coeff[6])[16];
395  uint32_t(*dequant8_coeff[6])[64];
396 
398  uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
400  int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
402 
403  // interlacing specific flags
406  int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
409 
410  DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
411 
412  // Weighted pred stuff
417  // The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
418  int luma_weight[48][2][2];
419  int chroma_weight[48][2][2][2];
420  int implicit_weight[48][48][2];
421 
427  int map_col_to_list0[2][16 + 32];
428  int map_col_to_list0_field[2][2][16 + 32];
429 
430  /**
431  * num_ref_idx_l0/1_active_minus1 + 1
432  */
433  unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
434  unsigned int list_count;
435  uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
436  Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
437  * Reordered version of default_ref_list
438  * according to picture reordering in slice header */
439  int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
440 
441  // data partitioning
446 
448  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.
449  DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
450  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
451 
452  /**
453  * Cabac
454  */
457 
458  /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
459  uint16_t *cbp_table;
460  int cbp;
461  int top_cbp;
462  int left_cbp;
463  /* chroma_pred_mode for i4x4 or i16x16, else 0 */
466  uint8_t (*mvd_table[2])[2];
467  DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
470 
483 
485 
486  int mb_x, mb_y;
492  int mb_num;
493  int mb_xy;
494 
496 
497  // deblock
498  int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
501 
502  // =============================================================
503  // Things below are not used in the MB or more inner code
504 
508  unsigned int rbsp_buffer_size[2];
509 
510  /**
511  * Used to parse AVC variant of h264
512  */
513  int is_avc; ///< this flag is != 0 if codec is avc1
514  int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
515  int got_first; ///< this flag is != 0 if we've parsed a frame
516 
517  int bit_depth_luma; ///< luma bit depth from sps to detect changes
518  int chroma_format_idc; ///< chroma format from sps to detect changes
519 
522 
523  int dequant_coeff_pps; ///< reinit tables when pps changes
524 
525  uint16_t *slice_table_base;
526 
527  // POC stuff
528  int poc_lsb;
529  int poc_msb;
531  int delta_poc[2];
533  int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
534  int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
535  int frame_num_offset; ///< for POC type 2
536  int prev_frame_num_offset; ///< for POC type 2
537  int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
538 
539  /**
540  * frame_num for frames or 2 * frame_num + 1 for field pics.
541  */
543 
544  /**
545  * max_frame_num or 2 * max_frame_num for field pics.
546  */
548 
550 
551  Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
559 
560  /**
561  * memory management control operations buffer.
562  */
566 
567  int long_ref_count; ///< number of actual long term references
568  int short_ref_count; ///< number of actual short term references
569 
571 
572  /**
573  * @name Members for slice based multithreading
574  * @{
575  */
577 
578  /**
579  * current slice number, used to initialize slice_num of each thread/context
580  */
582 
583  /**
584  * Max number of threads / contexts.
585  * This is equal to AVCodecContext.thread_count unless
586  * multithreaded decoding is impossible, in which case it is
587  * reduced to 1.
588  */
590 
592 
593  /**
594  * 1 if the single thread fallback warning has already been
595  * displayed, 0 otherwise.
596  */
598 
600 
602  unsigned int last_ref_count[2];
603  /** @} */
604 
605  /**
606  * pic_struct in picture timing SEI message
607  */
609 
610  /**
611  * Complement sei_pic_struct
612  * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
613  * However, soft telecined frames may have these values.
614  * This is used in an attempt to flag soft telecine progressive.
615  */
617 
618  /**
619  * frame_packing_arrangment SEI message
620  */
625 
626  /**
627  * Bit set of clock types for fields/frames in picture timing SEI message.
628  * For each found ct_type, appropriate bit is set (e.g., bit 1 for
629  * interlaced).
630  */
632 
633  /**
634  * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
635  */
637 
638  /**
639  * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
640  */
642 
643  /**
644  * recovery_frame_cnt from SEI message
645  *
646  * Set to -1 if no recovery point SEI message found or to number of frames
647  * before playback synchronizes. Frames having recovery point are key
648  * frames.
649  */
651 
652  /**
653  * Are the SEI recovery points looking valid.
654  */
656 
658 
659  /**
660  * recovery_frame is the frame_num at which the next frame should
661  * be fully constructed.
662  *
663  * Set to -1 when not expecting a recovery point.
664  */
666 
667 /**
668  * We have seen an IDR, so all the following frames in coded order are correctly
669  * decodable.
670  */
671 #define FRAME_RECOVERED_IDR (1 << 0)
672 /**
673  * Sufficient number of frames have been decoded since a SEI recovery point,
674  * so all the following frames in presentation order are correct.
675  */
676 #define FRAME_RECOVERED_SEI (1 << 1)
677 
678  int frame_recovered; ///< Initial frame has been completely recovered
679 
680  int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
681  int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
682 
683  // Timestamp stuff
684  int sei_buffering_period_present; ///< Buffering period SEI flag
685  int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
686 
689 
690  int16_t slice_row[MAX_SLICES]; ///< to detect when MAX_SLICES is too low
691 
696  int16_t *dc_val_base;
697 
702 } H264Context;
703 
704 extern const uint8_t ff_h264_chroma_qp[7][QP_MAX_NUM + 1]; ///< One chroma qp table for each possible bit depth (8-14).
705 extern const uint16_t ff_h264_mb_sizes[4];
706 
707 /**
708  * Decode SEI
709  */
711 
712 /**
713  * Decode SPS
714  */
716 
717 /**
718  * compute profile from sps
719  */
721 
722 /**
723  * Decode PPS
724  */
725 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
726 
727 /**
728  * Decode a network abstraction layer unit.
729  * @param consumed is the number of bytes used as input
730  * @param length is the length of the array
731  * @param dst_length is the number of decoded bytes FIXME here
732  * or a decode rbsp tailing?
733  * @return decoded bytes, might be src+1 if no escapes
734  */
736  int *dst_length, int *consumed, int length);
737 
738 /**
739  * Free any data that may have been allocated in the H264 context
740  * like SPS, PPS etc.
741  */
743 
744 /**
745  * Reconstruct bitstream slice_type.
746  */
747 int ff_h264_get_slice_type(const H264Context *h);
748 
749 /**
750  * Allocate tables.
751  * needs width/height
752  */
754 
755 /**
756  * Fill the default_ref_list.
757  */
759 
763 
764 /**
765  * Execute the reference picture marking (memory management control operations).
766  */
767 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
768 
770  int first_slice);
771 
772 int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
773 
774 /**
775  * Check if the top & left blocks are available if needed & change the
776  * dc mode so it only uses the available blocks.
777  */
779 
780 /**
781  * Check if the top & left blocks are available if needed & change the
782  * dc mode so it only uses the available blocks.
783  */
784 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma);
785 
789 void ff_h264_decode_init_vlc(void);
790 
791 /**
792  * Decode a macroblock
793  * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
794  */
796 
797 /**
798  * Decode a CABAC coded macroblock
799  * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
800  */
802 
804 
807 void ff_h264_pred_direct_motion(H264Context *const h, int *mb_type);
808 
809 void ff_h264_filter_mb_fast(H264Context *h, int mb_x, int mb_y,
810  uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
811  unsigned int linesize, unsigned int uvlinesize);
812 void ff_h264_filter_mb(H264Context *h, int mb_x, int mb_y,
813  uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
814  unsigned int linesize, unsigned int uvlinesize);
815 
816 /**
817  * Reset SEI values at the beginning of the frame.
818  *
819  * @param h H.264 context.
820  */
822 
823 /**
824  * Get stereo_mode string from the h264 frame_packing_arrangement
825  * @param h H.264 context.
826  */
827 const char* ff_h264_sei_stereo_mode(H264Context *h);
828 
829 /*
830  * o-o o-o
831  * / / /
832  * o-o o-o
833  * ,---'
834  * o-o o-o
835  * / / /
836  * o-o o-o
837  */
838 
839 /* Scan8 organization:
840  * 0 1 2 3 4 5 6 7
841  * 0 DY y y y y y
842  * 1 y Y Y Y Y
843  * 2 y Y Y Y Y
844  * 3 y Y Y Y Y
845  * 4 y Y Y Y Y
846  * 5 DU u u u u u
847  * 6 u U U U U
848  * 7 u U U U U
849  * 8 u U U U U
850  * 9 u U U U U
851  * 10 DV v v v v v
852  * 11 v V V V V
853  * 12 v V V V V
854  * 13 v V V V V
855  * 14 v V V V V
856  * DY/DU/DV are for luma/chroma DC.
857  */
858 
859 #define LUMA_DC_BLOCK_INDEX 48
860 #define CHROMA_DC_BLOCK_INDEX 49
861 
862 // This table must be here because scan8[constant] must be known at compiletime
863 static const uint8_t scan8[16 * 3 + 3] = {
864  4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
865  6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
866  4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
867  6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
868  4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
869  6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
870  4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
871  6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
872  4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
873  6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
874  4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
875  6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
876  0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
877 };
878 
879 static av_always_inline uint32_t pack16to32(int a, int b)
880 {
881 #if HAVE_BIGENDIAN
882  return (b & 0xFFFF) + (a << 16);
883 #else
884  return (a & 0xFFFF) + (b << 16);
885 #endif
886 }
887 
888 static av_always_inline uint16_t pack8to16(int a, int b)
889 {
890 #if HAVE_BIGENDIAN
891  return (b & 0xFF) + (a << 8);
892 #else
893  return (a & 0xFF) + (b << 8);
894 #endif
895 }
896 
897 /**
898  * Get the chroma qp.
899  */
901 {
902  return h->pps.chroma_qp_table[t][qscale];
903 }
904 
905 /**
906  * Get the predicted intra4x4 prediction mode.
907  */
909 {
910  const int index8 = scan8[n];
911  const int left = h->intra4x4_pred_mode_cache[index8 - 1];
912  const int top = h->intra4x4_pred_mode_cache[index8 - 8];
913  const int min = FFMIN(left, top);
914 
915  tprintf(h->avctx, "mode:%d %d min:%d\n", left, top, min);
916 
917  if (min < 0)
918  return DC_PRED;
919  else
920  return min;
921 }
922 
924 {
925  int8_t *i4x4 = h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
926  int8_t *i4x4_cache = h->intra4x4_pred_mode_cache;
927 
928  AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
929  i4x4[4] = i4x4_cache[7 + 8 * 3];
930  i4x4[5] = i4x4_cache[7 + 8 * 2];
931  i4x4[6] = i4x4_cache[7 + 8 * 1];
932 }
933 
935 {
936  const int mb_xy = h->mb_xy;
937  uint8_t *nnz = h->non_zero_count[mb_xy];
938  uint8_t *nnz_cache = h->non_zero_count_cache;
939 
940  AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
941  AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
942  AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
943  AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
944  AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
945  AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
946  AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
947  AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
948 
949  if (!h->chroma_y_shift) {
950  AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
951  AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
952  AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
953  AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
954  }
955 }
956 
958  int b_stride,
959  int b_xy, int b8_xy,
960  int mb_type, int list)
961 {
962  int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
963  int16_t(*mv_src)[2] = &h->mv_cache[list][scan8[0]];
964  AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
965  AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
966  AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
967  AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
968  if (CABAC(h)) {
969  uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8 * h->mb_xy
970  : h->mb2br_xy[h->mb_xy]];
971  uint8_t(*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
972  if (IS_SKIP(mb_type)) {
973  AV_ZERO128(mvd_dst);
974  } else {
975  AV_COPY64(mvd_dst, mvd_src + 8 * 3);
976  AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
977  AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
978  AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
979  }
980  }
981 
982  {
983  int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
984  int8_t *ref_cache = h->ref_cache[list];
985  ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
986  ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
987  ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
988  ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
989  }
990 }
991 
992 static av_always_inline void write_back_motion(H264Context *h, int mb_type)
993 {
994  const int b_stride = h->b_stride;
995  const int b_xy = 4 * h->mb_x + 4 * h->mb_y * h->b_stride; // try mb2b(8)_xy
996  const int b8_xy = 4 * h->mb_xy;
997 
998  if (USES_LIST(mb_type, 0)) {
999  write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 0);
1000  } else {
1001  fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
1002  2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1003  }
1004  if (USES_LIST(mb_type, 1))
1005  write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 1);
1006 
1007  if (h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) {
1008  if (IS_8X8(mb_type)) {
1009  uint8_t *direct_table = &h->direct_table[4 * h->mb_xy];
1010  direct_table[1] = h->sub_mb_type[1] >> 1;
1011  direct_table[2] = h->sub_mb_type[2] >> 1;
1012  direct_table[3] = h->sub_mb_type[3] >> 1;
1013  }
1014  }
1015 }
1016 
1018 {
1020  return !(AV_RN64A(h->sub_mb_type) &
1022  0x0001000100010001ULL));
1023  else
1024  return !(AV_RN64A(h->sub_mb_type) &
1026  0x0001000100010001ULL));
1027 }
1028 
1029 void ff_h264_draw_horiz_band(H264Context *h, int y, int height);
1030 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc);
1033 
1034 #endif /* AVCODEC_H264_H */