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 {
119  NAL_FF_IGNORE = 0xff0f001,
120 };
121 
122 /**
123  * SEI message types
124  */
125 typedef enum {
126  SEI_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 {
169  int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
170  int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
171  int poc_type; ///< pic_order_cnt_type
172  int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
176  int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
177  int ref_frame_count; ///< num_ref_frames
179  int mb_width; ///< pic_width_in_mbs_minus1 + 1
180  int mb_height; ///< pic_height_in_map_units_minus1 + 1
182  int mb_aff; ///< mb_adaptive_frame_field_flag
184  int crop; ///< frame_cropping_flag
185 
186  /* those 4 are already in luma samples */
187  unsigned int crop_left; ///< frame_cropping_rect_left_offset
188  unsigned int crop_right; ///< frame_cropping_rect_right_offset
189  unsigned int crop_top; ///< frame_cropping_rect_top_offset
190  unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
201  uint32_t time_scale;
203  short offset_for_ref_frame[256]; // FIXME dyn aloc?
213  int cpb_cnt; ///< See H.264 E.1.2
214  int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
215  int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
216  int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
217  int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
218  int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
219  int residual_color_transform_flag; ///< residual_colour_transform_flag
220  int constraint_set_flags; ///< constraint_set[0-3]_flag
221  int new; ///< flag to keep track if the decoder context needs re-init due to changed SPS
222 } SPS;
223 
224 /**
225  * Picture parameter set
226  */
227 typedef struct PPS {
228  unsigned int sps_id;
229  int cabac; ///< entropy_coding_mode_flag
230  int pic_order_present; ///< pic_order_present_flag
231  int slice_group_count; ///< num_slice_groups_minus1 + 1
233  unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
234  int weighted_pred; ///< weighted_pred_flag
236  int init_qp; ///< pic_init_qp_minus26 + 26
237  int init_qs; ///< pic_init_qs_minus26 + 26
239  int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
240  int constrained_intra_pred; ///< constrained_intra_pred_flag
241  int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
242  int transform_8x8_mode; ///< transform_8x8_mode_flag
245  uint8_t chroma_qp_table[2][QP_MAX_NUM+1]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
247 } PPS;
248 
249 /**
250  * Frame Packing Arrangement Type
251  */
252 typedef struct FPA {
254  int frame_packing_arrangement_cancel_flag; ///< is previous arrangement canceled, -1 if never received
259 } FPA;
260 
261 /**
262  * Memory management control operation opcode.
263  */
264 typedef enum MMCOOpcode {
265  MMCO_END = 0,
272 } MMCOOpcode;
273 
274 /**
275  * Memory management control operation.
276  */
277 typedef struct MMCO {
279  int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
280  int long_arg; ///< index, pic_num, or num long refs depending on opcode
281 } MMCO;
282 
283 /**
284  * H264Context
285  */
286 typedef struct H264Context {
297 
301 
302  int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
303  int chroma_qp[2]; // QPc
304 
305  int qp_thresh; ///< QP threshold to skip loopfilter
306 
307  /* coded dimensions -- 16 * mb w/h */
308  int width, height;
309  ptrdiff_t linesize, uvlinesize;
311 
312  int qscale;
317 
319  int flags;
321 
324 
325  // prediction stuff
328 
333 
335  int top_type;
338 
341 
346  unsigned int top_samples_available;
349  uint8_t (*top_borders[2])[(16 * 3) * 2];
350 
351  /**
352  * non zero coeff count cache.
353  * is 64 if not available.
354  */
356 
358 
359  /**
360  * Motion vector cache.
361  */
362  DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
363  DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
364 #define LIST_NOT_USED -1 // FIXME rename?
365 #define PART_NOT_AVAILABLE -2
366 
367  /**
368  * number of neighbors (top and/or left) that used 8x8 dct
369  */
371 
372  /**
373  * block_offset[ 0..23] for frame macroblocks
374  * block_offset[24..47] for field macroblocks
375  */
376  int block_offset[2 * (16 * 3)];
377 
378  uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
379  uint32_t *mb2br_xy;
380  int b_stride; // FIXME use s->b4_stride
381 
382  ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
383  ptrdiff_t mb_uvlinesize;
384 
385  unsigned current_sps_id; ///< id of the current SPS
386  SPS sps; ///< current sps
387 
388  /**
389  * current pps
390  */
391  PPS pps; // FIXME move to Picture perhaps? (->no) do we need that?
392 
393  uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
394  uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
395  uint32_t(*dequant4_coeff[6])[16];
396  uint32_t(*dequant8_coeff[6])[64];
397 
399  uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
401  int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
403 
404  // interlacing specific flags
407  int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
410 
411  DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
412 
413  // Weighted pred stuff
418  // The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
419  int luma_weight[48][2][2];
420  int chroma_weight[48][2][2][2];
421  int implicit_weight[48][48][2];
422 
428  int map_col_to_list0[2][16 + 32];
429  int map_col_to_list0_field[2][2][16 + 32];
430 
431  /**
432  * num_ref_idx_l0/1_active_minus1 + 1
433  */
434  unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
435  unsigned int list_count;
436  uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
437  Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
438  * Reordered version of default_ref_list
439  * according to picture reordering in slice header */
440  int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
441 
442  // data partitioning
447 
449  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.
450  DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
451  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
452 
453  /**
454  * Cabac
455  */
458 
459  /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
460  uint16_t *cbp_table;
461  int cbp;
462  int top_cbp;
463  int left_cbp;
464  /* chroma_pred_mode for i4x4 or i16x16, else 0 */
467  uint8_t (*mvd_table[2])[2];
468  DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
471 
484 
486 
487  int mb_x, mb_y;
493  int mb_num;
494  int mb_xy;
495 
497 
498  // deblock
499  int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
502 
503  // =============================================================
504  // Things below are not used in the MB or more inner code
505 
509  unsigned int rbsp_buffer_size[2];
510 
511  /**
512  * Used to parse AVC variant of h264
513  */
514  int is_avc; ///< this flag is != 0 if codec is avc1
515  int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
516  int got_first; ///< this flag is != 0 if we've parsed a frame
517 
518  int bit_depth_luma; ///< luma bit depth from sps to detect changes
519  int chroma_format_idc; ///< chroma format from sps to detect changes
520 
523 
524  int dequant_coeff_pps; ///< reinit tables when pps changes
525 
526  uint16_t *slice_table_base;
527 
528  // POC stuff
529  int poc_lsb;
530  int poc_msb;
532  int delta_poc[2];
534  int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
535  int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
536  int frame_num_offset; ///< for POC type 2
537  int prev_frame_num_offset; ///< for POC type 2
538  int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
539 
540  /**
541  * frame_num for frames or 2 * frame_num + 1 for field pics.
542  */
544 
545  /**
546  * max_frame_num or 2 * max_frame_num for field pics.
547  */
549 
551 
552  Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
560 
561  /**
562  * memory management control operations buffer.
563  */
567 
568  int long_ref_count; ///< number of actual long term references
569  int short_ref_count; ///< number of actual short term references
570 
572 
573  /**
574  * @name Members for slice based multithreading
575  * @{
576  */
578 
579  /**
580  * current slice number, used to initialize slice_num of each thread/context
581  */
583 
584  /**
585  * Max number of threads / contexts.
586  * This is equal to AVCodecContext.thread_count unless
587  * multithreaded decoding is impossible, in which case it is
588  * reduced to 1.
589  */
591 
593 
594  /**
595  * 1 if the single thread fallback warning has already been
596  * displayed, 0 otherwise.
597  */
599 
601 
603  unsigned int last_ref_count[2];
604  /** @} */
605 
606  /**
607  * pic_struct in picture timing SEI message
608  */
610 
611  /**
612  * Complement sei_pic_struct
613  * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
614  * However, soft telecined frames may have these values.
615  * This is used in an attempt to flag soft telecine progressive.
616  */
618 
619  /**
620  * Bit set of clock types for fields/frames in picture timing SEI message.
621  * For each found ct_type, appropriate bit is set (e.g., bit 1 for
622  * interlaced).
623  */
625 
626  /**
627  * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
628  */
630 
631  /**
632  * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
633  */
635 
636  /**
637  * recovery_frame_cnt from SEI message
638  *
639  * Set to -1 if no recovery point SEI message found or to number of frames
640  * before playback synchronizes. Frames having recovery point are key
641  * frames.
642  */
644  /**
645  * recovery_frame is the frame_num at which the next frame should
646  * be fully constructed.
647  *
648  * Set to -1 when not expecting a recovery point.
649  */
651 
652  /**
653  * Are the SEI recovery points looking valid.
654  */
656 
658 
659  int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
660  int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
661 
662  // Timestamp stuff
663  int sei_buffering_period_present; ///< Buffering period SEI flag
664  int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
665 
668 
669  int16_t slice_row[MAX_SLICES]; ///< to detect when MAX_SLICES is too low
670 
671  int sync; ///< did we had a keyframe or recovery point
672 
677  int16_t *dc_val_base;
678 
679  uint8_t *visualization_buffer[3]; ///< temporary buffer vor MV visualization
680 
685 } H264Context;
686 
687 extern const uint8_t ff_h264_chroma_qp[7][QP_MAX_NUM + 1]; ///< One chroma qp table for each possible bit depth (8-14).
688 extern const uint16_t ff_h264_mb_sizes[4];
689 
690 /**
691  * Decode SEI
692  */
694 
695 /**
696  * Decode SPS
697  */
699 
700 /**
701  * compute profile from sps
702  */
704 
705 /**
706  * Decode PPS
707  */
708 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
709 
710 /**
711  * Decode a network abstraction layer unit.
712  * @param consumed is the number of bytes used as input
713  * @param length is the length of the array
714  * @param dst_length is the number of decoded bytes FIXME here
715  * or a decode rbsp tailing?
716  * @return decoded bytes, might be src+1 if no escapes
717  */
719  int *dst_length, int *consumed, int length);
720 
721 /**
722  * Free any data that may have been allocated in the H264 context
723  * like SPS, PPS etc.
724  */
726 
727 /**
728  * Reconstruct bitstream slice_type.
729  */
730 int ff_h264_get_slice_type(const H264Context *h);
731 
732 /**
733  * Allocate tables.
734  * needs width/height
735  */
737 
738 /**
739  * Fill the default_ref_list.
740  */
742 
746 
747 /**
748  * Execute the reference picture marking (memory management control operations).
749  */
750 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
751 
753  int first_slice);
754 
755 int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
756 
757 /**
758  * Check if the top & left blocks are available if needed & change the
759  * dc mode so it only uses the available blocks.
760  */
762 
763 /**
764  * Check if the top & left blocks are available if needed & change the
765  * dc mode so it only uses the available blocks.
766  */
767 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma);
768 
772 void ff_h264_decode_init_vlc(void);
773 
774 /**
775  * Decode a macroblock
776  * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
777  */
779 
780 /**
781  * Decode a CABAC coded macroblock
782  * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
783  */
785 
787 
790 void ff_h264_pred_direct_motion(H264Context *const h, int *mb_type);
791 
792 void ff_h264_filter_mb_fast(H264Context *h, int mb_x, int mb_y,
793  uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
794  unsigned int linesize, unsigned int uvlinesize);
795 void ff_h264_filter_mb(H264Context *h, int mb_x, int mb_y,
796  uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
797  unsigned int linesize, unsigned int uvlinesize);
798 
799 /**
800  * Reset SEI values at the beginning of the frame.
801  *
802  * @param h H.264 context.
803  */
805 
806 /**
807  * Get stereo_mode string from the h264 frame_packing_arrangement
808  * @param h H.264 context.
809  */
810 const char* ff_h264_sei_stereo_mode(H264Context *h);
811 
812 /*
813  * o-o o-o
814  * / / /
815  * o-o o-o
816  * ,---'
817  * o-o o-o
818  * / / /
819  * o-o o-o
820  */
821 
822 /* Scan8 organization:
823  * 0 1 2 3 4 5 6 7
824  * 0 DY y y y y y
825  * 1 y Y Y Y Y
826  * 2 y Y Y Y Y
827  * 3 y Y Y Y Y
828  * 4 y Y Y Y Y
829  * 5 DU u u u u u
830  * 6 u U U U U
831  * 7 u U U U U
832  * 8 u U U U U
833  * 9 u U U U U
834  * 10 DV v v v v v
835  * 11 v V V V V
836  * 12 v V V V V
837  * 13 v V V V V
838  * 14 v V V V V
839  * DY/DU/DV are for luma/chroma DC.
840  */
841 
842 #define LUMA_DC_BLOCK_INDEX 48
843 #define CHROMA_DC_BLOCK_INDEX 49
844 
845 // This table must be here because scan8[constant] must be known at compiletime
846 static const uint8_t scan8[16 * 3 + 3] = {
847  4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
848  6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
849  4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
850  6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
851  4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
852  6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
853  4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
854  6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
855  4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
856  6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
857  4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
858  6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
859  0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
860 };
861 
862 static av_always_inline uint32_t pack16to32(int a, int b)
863 {
864 #if HAVE_BIGENDIAN
865  return (b & 0xFFFF) + (a << 16);
866 #else
867  return (a & 0xFFFF) + (b << 16);
868 #endif
869 }
870 
871 static av_always_inline uint16_t pack8to16(int a, int b)
872 {
873 #if HAVE_BIGENDIAN
874  return (b & 0xFF) + (a << 8);
875 #else
876  return (a & 0xFF) + (b << 8);
877 #endif
878 }
879 
880 /**
881  * Get the chroma qp.
882  */
884 {
885  return h->pps.chroma_qp_table[t][qscale];
886 }
887 
888 /**
889  * Get the predicted intra4x4 prediction mode.
890  */
892 {
893  const int index8 = scan8[n];
894  const int left = h->intra4x4_pred_mode_cache[index8 - 1];
895  const int top = h->intra4x4_pred_mode_cache[index8 - 8];
896  const int min = FFMIN(left, top);
897 
898  tprintf(h->avctx, "mode:%d %d min:%d\n", left, top, min);
899 
900  if (min < 0)
901  return DC_PRED;
902  else
903  return min;
904 }
905 
907 {
908  int8_t *i4x4 = h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
909  int8_t *i4x4_cache = h->intra4x4_pred_mode_cache;
910 
911  AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
912  i4x4[4] = i4x4_cache[7 + 8 * 3];
913  i4x4[5] = i4x4_cache[7 + 8 * 2];
914  i4x4[6] = i4x4_cache[7 + 8 * 1];
915 }
916 
918 {
919  const int mb_xy = h->mb_xy;
920  uint8_t *nnz = h->non_zero_count[mb_xy];
921  uint8_t *nnz_cache = h->non_zero_count_cache;
922 
923  AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
924  AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
925  AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
926  AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
927  AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
928  AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
929  AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
930  AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
931 
932  if (!h->chroma_y_shift) {
933  AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
934  AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
935  AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
936  AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
937  }
938 }
939 
941  int b_stride,
942  int b_xy, int b8_xy,
943  int mb_type, int list)
944 {
945  int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
946  int16_t(*mv_src)[2] = &h->mv_cache[list][scan8[0]];
947  AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
948  AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
949  AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
950  AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
951  if (CABAC(h)) {
952  uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8 * h->mb_xy
953  : h->mb2br_xy[h->mb_xy]];
954  uint8_t(*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
955  if (IS_SKIP(mb_type)) {
956  AV_ZERO128(mvd_dst);
957  } else {
958  AV_COPY64(mvd_dst, mvd_src + 8 * 3);
959  AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
960  AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
961  AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
962  }
963  }
964 
965  {
966  int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
967  int8_t *ref_cache = h->ref_cache[list];
968  ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
969  ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
970  ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
971  ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
972  }
973 }
974 
975 static av_always_inline void write_back_motion(H264Context *h, int mb_type)
976 {
977  const int b_stride = h->b_stride;
978  const int b_xy = 4 * h->mb_x + 4 * h->mb_y * h->b_stride; // try mb2b(8)_xy
979  const int b8_xy = 4 * h->mb_xy;
980 
981  if (USES_LIST(mb_type, 0)) {
982  write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 0);
983  } else {
984  fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
985  2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
986  }
987  if (USES_LIST(mb_type, 1))
988  write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 1);
989 
990  if (h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) {
991  if (IS_8X8(mb_type)) {
992  uint8_t *direct_table = &h->direct_table[4 * h->mb_xy];
993  direct_table[1] = h->sub_mb_type[1] >> 1;
994  direct_table[2] = h->sub_mb_type[2] >> 1;
995  direct_table[3] = h->sub_mb_type[3] >> 1;
996  }
997  }
998 }
999 
1001 {
1003  return !(AV_RN64A(h->sub_mb_type) &
1005  0x0001000100010001ULL));
1006  else
1007  return !(AV_RN64A(h->sub_mb_type) &
1009  0x0001000100010001ULL));
1010 }
1011 
1012 void ff_h264_draw_horiz_band(H264Context *h, int y, int height);
1013 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc);
1016 
1017 #endif /* AVCODEC_H264_H */