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 "h264chroma.h"
36 #include "h264dsp.h"
37 #include "h264pred.h"
38 #include "h264qpel.h"
39 #include "internal.h" // for avpriv_find_start_code()
40 #include "me_cmp.h"
41 #include "mpegutils.h"
42 #include "parser.h"
43 #include "qpeldsp.h"
44 #include "rectangle.h"
45 #include "videodsp.h"
46 
47 #define H264_MAX_PICTURE_COUNT 36
48 #define H264_MAX_THREADS 32
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 #define MAX_MBPAIR_SIZE (256*1024) // a tighter bound could be calculated if someone cares about a few bytes
58 
59 /* Compiling in interlaced support reduces the speed
60  * of progressive decoding by about 2%. */
61 #define ALLOW_INTERLACE
62 
63 #define FMO 0
64 
65 /**
66  * The maximum number of slices supported by the decoder.
67  * must be a power of 2
68  */
69 #define MAX_SLICES 16
70 
71 #ifdef ALLOW_INTERLACE
72 #define MB_MBAFF(h) (h)->mb_mbaff
73 #define MB_FIELD(h) (h)->mb_field_decoding_flag
74 #define FRAME_MBAFF(h) (h)->mb_aff_frame
75 #define FIELD_PICTURE(h) ((h)->picture_structure != PICT_FRAME)
76 #define LEFT_MBS 2
77 #define LTOP 0
78 #define LBOT 1
79 #define LEFT(i) (i)
80 #else
81 #define MB_MBAFF(h) 0
82 #define MB_FIELD(h) 0
83 #define FRAME_MBAFF(h) 0
84 #define FIELD_PICTURE(h) 0
85 #undef IS_INTERLACED
86 #define IS_INTERLACED(mb_type) 0
87 #define LEFT_MBS 1
88 #define LTOP 0
89 #define LBOT 0
90 #define LEFT(i) 0
91 #endif
92 #define FIELD_OR_MBAFF_PICTURE(h) (FRAME_MBAFF(h) || FIELD_PICTURE(h))
93 
94 #ifndef CABAC
95 #define CABAC(h) (h)->pps.cabac
96 #endif
97 
98 #define CHROMA(h) ((h)->sps.chroma_format_idc)
99 #define CHROMA422(h) ((h)->sps.chroma_format_idc == 2)
100 #define CHROMA444(h) ((h)->sps.chroma_format_idc == 3)
101 
102 #define EXTENDED_SAR 255
103 
104 #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit
105 #define MB_TYPE_8x8DCT 0x01000000
106 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
107 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
108 
109 #define QP_MAX_NUM (51 + 6*6) // The maximum supported qp
110 
111 /* NAL unit types */
112 enum {
114  NAL_DPA = 2,
115  NAL_DPB = 3,
116  NAL_DPC = 4,
118  NAL_SEI = 6,
119  NAL_SPS = 7,
120  NAL_PPS = 8,
121  NAL_AUD = 9,
127  NAL_FF_IGNORE = 0xff0f001,
128 };
129 
130 /**
131  * SEI message types
132  */
133 typedef enum {
134  SEI_TYPE_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
135  SEI_TYPE_PIC_TIMING = 1, ///< picture timing
136  SEI_TYPE_USER_DATA_ITU_T_T35 = 4, ///< user data registered by ITU-T Recommendation T.35
137  SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
138  SEI_TYPE_RECOVERY_POINT = 6, ///< recovery point (frame # to decoder sync)
139  SEI_TYPE_FRAME_PACKING = 45, ///< frame packing arrangement
140  SEI_TYPE_DISPLAY_ORIENTATION = 47, ///< display orientation
141 } SEI_Type;
142 
143 /**
144  * pic_struct in picture timing SEI message
145  */
146 typedef enum {
147  SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
148  SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
149  SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
150  SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
151  SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
152  SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
153  SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
154  SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
155  SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
157 
158 /**
159  * frame_packing_arrangement types
160  */
161 typedef enum {
169 } SEI_FpaType;
170 
171 /**
172  * Sequence parameter set
173  */
174 typedef struct SPS {
175  unsigned int sps_id;
179  int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
180  int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
181  int poc_type; ///< pic_order_cnt_type
182  int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
186  int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
187  int ref_frame_count; ///< num_ref_frames
189  int mb_width; ///< pic_width_in_mbs_minus1 + 1
190  int mb_height; ///< pic_height_in_map_units_minus1 + 1
192  int mb_aff; ///< mb_adaptive_frame_field_flag
194  int crop; ///< frame_cropping_flag
195 
196  /* those 4 are already in luma samples */
197  unsigned int crop_left; ///< frame_cropping_rect_left_offset
198  unsigned int crop_right; ///< frame_cropping_rect_right_offset
199  unsigned int crop_top; ///< frame_cropping_rect_top_offset
200  unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
211  uint32_t time_scale;
213  short offset_for_ref_frame[256]; // FIXME dyn aloc?
223  int cpb_cnt; ///< See H.264 E.1.2
224  int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
225  int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
226  int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
227  int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
228  int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
229  int residual_color_transform_flag; ///< residual_colour_transform_flag
230  int constraint_set_flags; ///< constraint_set[0-3]_flag
231  int new; ///< flag to keep track if the decoder context needs re-init due to changed SPS
232 } SPS;
233 
234 /**
235  * Picture parameter set
236  */
237 typedef struct PPS {
238  unsigned int sps_id;
239  int cabac; ///< entropy_coding_mode_flag
240  int pic_order_present; ///< pic_order_present_flag
241  int slice_group_count; ///< num_slice_groups_minus1 + 1
243  unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
244  int weighted_pred; ///< weighted_pred_flag
246  int init_qp; ///< pic_init_qp_minus26 + 26
247  int init_qs; ///< pic_init_qs_minus26 + 26
249  int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
250  int constrained_intra_pred; ///< constrained_intra_pred_flag
251  int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
252  int transform_8x8_mode; ///< transform_8x8_mode_flag
255  uint8_t chroma_qp_table[2][QP_MAX_NUM+1]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
257 } PPS;
258 
259 /**
260  * Frame Packing Arrangement Type
261  */
262 typedef struct FPA {
264  int frame_packing_arrangement_cancel_flag; ///< is previous arrangement canceled, -1 if never received
269 } FPA;
270 
271 /**
272  * Memory management control operation opcode.
273  */
274 typedef enum MMCOOpcode {
275  MMCO_END = 0,
282 } MMCOOpcode;
283 
284 /**
285  * Memory management control operation.
286  */
287 typedef struct MMCO {
289  int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
290  int long_arg; ///< index, pic_num, or num long refs depending on opcode
291 } MMCO;
292 
293 typedef struct H264Picture {
294  struct AVFrame f;
295  uint8_t avframe_padding[1024]; // hack to allow linking to a avutil with larger AVFrame
297 
299  int8_t *qscale_table;
300 
302  int16_t (*motion_val[2])[2];
303 
305  uint32_t *mb_type;
306 
308  void *hwaccel_picture_private; ///< hardware accelerator private data
309 
311  int8_t *ref_index[2];
312 
313  int field_poc[2]; ///< top/bottom POC
314  int poc; ///< frame POC
315  int frame_num; ///< frame_num (raw frame_num from slice header)
316  int mmco_reset; /**< MMCO_RESET set this 1. Reordering code must
317  not mix pictures before and after MMCO_RESET. */
318  int pic_id; /**< pic_num (short -> no wrap version of pic_num,
319  pic_num & max_pic_num; long -> long_pic_num) */
320  int long_ref; ///< 1->long term reference 0->short term reference
321  int ref_poc[2][2][32]; ///< POCs of the frames/fields used as reference (FIXME need per slice)
322  int ref_count[2][2]; ///< number of entries in ref_poc (FIXME need per slice)
323  int mbaff; ///< 1 -> MBAFF frame 0-> not MBAFF
324  int field_picture; ///< whether or not picture was encoded in separate fields
325 
326  int needs_realloc; ///< picture needs to be reallocated (eg due to a frame size change)
328  int recovered; ///< picture at IDR or recovery point + recovery count
331 
332  int crop;
334  int crop_top;
335 } H264Picture;
336 
337 /**
338  * H264Context
339  */
340 typedef struct H264Context {
350 
354 
355  int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
356  int chroma_qp[2]; // QPc
357 
358  int qp_thresh; ///< QP threshold to skip loopfilter
359 
360  /* coded dimensions -- 16 * mb w/h */
361  int width, height;
362  ptrdiff_t linesize, uvlinesize;
364 
365  int qscale;
370 
372  int flags;
374 
377 
378  // prediction stuff
381 
386 
388  int top_type;
391 
394 
399  unsigned int top_samples_available;
402  uint8_t (*top_borders[2])[(16 * 3) * 2];
403 
404  /**
405  * non zero coeff count cache.
406  * is 64 if not available.
407  */
409 
411 
412  /**
413  * Motion vector cache.
414  */
415  DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
416  DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
417 #define LIST_NOT_USED -1 // FIXME rename?
418 #define PART_NOT_AVAILABLE -2
419 
420  /**
421  * number of neighbors (top and/or left) that used 8x8 dct
422  */
424 
425  /**
426  * block_offset[ 0..23] for frame macroblocks
427  * block_offset[24..47] for field macroblocks
428  */
429  int block_offset[2 * (16 * 3)];
430 
431  uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
432  uint32_t *mb2br_xy;
433  int b_stride; // FIXME use s->b4_stride
434 
435  ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
436  ptrdiff_t mb_uvlinesize;
437 
438  unsigned current_sps_id; ///< id of the current SPS
439  SPS sps; ///< current sps
440  PPS pps; ///< current pps
441 
442  int au_pps_id; ///< pps_id of current access unit
443 
444  uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
445  uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
446  uint32_t(*dequant4_coeff[6])[16];
447  uint32_t(*dequant8_coeff[6])[64];
448 
450  uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
452  int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
454 
455  // interlacing specific flags
458  int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
461 
462  DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
463 
464  // Weighted pred stuff
469  // The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
470  int luma_weight[48][2][2];
471  int chroma_weight[48][2][2][2];
472  int implicit_weight[48][48][2];
473 
479  int map_col_to_list0[2][16 + 32];
480  int map_col_to_list0_field[2][2][16 + 32];
481 
482  /**
483  * num_ref_idx_l0/1_active_minus1 + 1
484  */
485  unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
486  unsigned int list_count;
487  uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
488  H264Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
489  * Reordered version of default_ref_list
490  * according to picture reordering in slice header */
491  int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
492 
493  // data partitioning
498 
500  DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2]; ///< as a dct coefficient is int32_t in high depth, we need to reserve twice the space.
501  DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
502  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
503 
504  /**
505  * Cabac
506  */
509 
510  /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
511  uint16_t *cbp_table;
512  int cbp;
513  int top_cbp;
514  int left_cbp;
515  /* chroma_pred_mode for i4x4 or i16x16, else 0 */
518  uint8_t (*mvd_table[2])[2];
519  DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
522 
535 
537 
538  int mb_x, mb_y;
544  int mb_num;
545  int mb_xy;
546 
548 
549  // deblock
550  int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
553 
554  // =============================================================
555  // Things below are not used in the MB or more inner code
556 
560  unsigned int rbsp_buffer_size[2];
561 
562  /**
563  * Used to parse AVC variant of h264
564  */
565  int is_avc; ///< this flag is != 0 if codec is avc1
566  int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
567  int got_first; ///< this flag is != 0 if we've parsed a frame
568 
569  int bit_depth_luma; ///< luma bit depth from sps to detect changes
570  int chroma_format_idc; ///< chroma format from sps to detect changes
571 
574 
575  int dequant_coeff_pps; ///< reinit tables when pps changes
576 
577  uint16_t *slice_table_base;
578 
579  // POC stuff
580  int poc_lsb;
581  int poc_msb;
583  int delta_poc[2];
585  int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
586  int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
587  int frame_num_offset; ///< for POC type 2
588  int prev_frame_num_offset; ///< for POC type 2
589  int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
590 
591  /**
592  * frame_num for frames or 2 * frame_num + 1 for field pics.
593  */
595 
596  /**
597  * max_frame_num or 2 * max_frame_num for field pics.
598  */
600 
602 
603  H264Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
611 
612  /**
613  * memory management control operations buffer.
614  */
618 
619  int long_ref_count; ///< number of actual long term references
620  int short_ref_count; ///< number of actual short term references
621 
623 
624  /**
625  * @name Members for slice based multithreading
626  * @{
627  */
629 
630  /**
631  * current slice number, used to initialize slice_num of each thread/context
632  */
634 
635  /**
636  * Max number of threads / contexts.
637  * This is equal to AVCodecContext.thread_count unless
638  * multithreaded decoding is impossible, in which case it is
639  * reduced to 1.
640  */
642 
644 
645  /**
646  * 1 if the single thread fallback warning has already been
647  * displayed, 0 otherwise.
648  */
650 
652 
654  unsigned int last_ref_count[2];
655  /** @} */
656 
657  /**
658  * pic_struct in picture timing SEI message
659  */
661 
662  /**
663  * Complement sei_pic_struct
664  * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
665  * However, soft telecined frames may have these values.
666  * This is used in an attempt to flag soft telecine progressive.
667  */
669 
670  /**
671  * frame_packing_arrangment SEI message
672  */
677 
678  /**
679  * display orientation SEI message
680  */
684 
685  /**
686  * Bit set of clock types for fields/frames in picture timing SEI message.
687  * For each found ct_type, appropriate bit is set (e.g., bit 1 for
688  * interlaced).
689  */
691 
692  /**
693  * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
694  */
696 
697  /**
698  * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
699  */
701 
702  /**
703  * recovery_frame_cnt from SEI message
704  *
705  * Set to -1 if no recovery point SEI message found or to number of frames
706  * before playback synchronizes. Frames having recovery point are key
707  * frames.
708  */
710 
711  /**
712  * Are the SEI recovery points looking valid.
713  */
715 
717 
718  /**
719  * recovery_frame is the frame_num at which the next frame should
720  * be fully constructed.
721  *
722  * Set to -1 when not expecting a recovery point.
723  */
725 
726 /**
727  * We have seen an IDR, so all the following frames in coded order are correctly
728  * decodable.
729  */
730 #define FRAME_RECOVERED_IDR (1 << 0)
731 /**
732  * Sufficient number of frames have been decoded since a SEI recovery point,
733  * so all the following frames in presentation order are correct.
734  */
735 #define FRAME_RECOVERED_SEI (1 << 1)
736 
737  int frame_recovered; ///< Initial frame has been completely recovered
738 
740 
741  int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
742  int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
743 
744  // Timestamp stuff
745  int sei_buffering_period_present; ///< Buffering period SEI flag
746  int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
747 
750 
751  int16_t slice_row[MAX_SLICES]; ///< to detect when MAX_SLICES is too low
752 
757  int16_t *dc_val_base;
758 
763 
764  /* Motion Estimation */
767 } H264Context;
768 
769 extern const uint8_t ff_h264_chroma_qp[7][QP_MAX_NUM + 1]; ///< One chroma qp table for each possible bit depth (8-14).
770 extern const uint16_t ff_h264_mb_sizes[4];
771 
772 /**
773  * Decode SEI
774  */
776 
777 /**
778  * Decode SPS
779  */
781 
782 /**
783  * compute profile from sps
784  */
786 
787 /**
788  * Decode PPS
789  */
790 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
791 
792 /**
793  * Decode a network abstraction layer unit.
794  * @param consumed is the number of bytes used as input
795  * @param length is the length of the array
796  * @param dst_length is the number of decoded bytes FIXME here
797  * or a decode rbsp tailing?
798  * @return decoded bytes, might be src+1 if no escapes
799  */
801  int *dst_length, int *consumed, int length);
802 
803 /**
804  * Free any data that may have been allocated in the H264 context
805  * like SPS, PPS etc.
806  */
808 
809 /**
810  * Reconstruct bitstream slice_type.
811  */
812 int ff_h264_get_slice_type(const H264Context *h);
813 
814 /**
815  * Allocate tables.
816  * needs width/height
817  */
819 
820 /**
821  * Fill the default_ref_list.
822  */
824 
828 
829 /**
830  * Execute the reference picture marking (memory management control operations).
831  */
832 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
833 
835  int first_slice);
836 
837 int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
838 
839 /**
840  * Check if the top & left blocks are available if needed & change the
841  * dc mode so it only uses the available blocks.
842  */
844 
845 /**
846  * Check if the top & left blocks are available if needed & change the
847  * dc mode so it only uses the available blocks.
848  */
849 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma);
850 
854 void ff_h264_decode_init_vlc(void);
855 
856 /**
857  * Decode a macroblock
858  * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
859  */
861 
862 /**
863  * Decode a CABAC coded macroblock
864  * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
865  */
867 
869 
871 
874 void ff_h264_pred_direct_motion(H264Context *const h, int *mb_type);
875 
876 void ff_h264_filter_mb_fast(H264Context *h, int mb_x, int mb_y,
877  uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
878  unsigned int linesize, unsigned int uvlinesize);
879 void ff_h264_filter_mb(H264Context *h, int mb_x, int mb_y,
880  uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
881  unsigned int linesize, unsigned int uvlinesize);
882 
883 /**
884  * Reset SEI values at the beginning of the frame.
885  *
886  * @param h H.264 context.
887  */
889 
890 /**
891  * Get stereo_mode string from the h264 frame_packing_arrangement
892  * @param h H.264 context.
893  */
894 const char* ff_h264_sei_stereo_mode(H264Context *h);
895 
896 /*
897  * o-o o-o
898  * / / /
899  * o-o o-o
900  * ,---'
901  * o-o o-o
902  * / / /
903  * o-o o-o
904  */
905 
906 /* Scan8 organization:
907  * 0 1 2 3 4 5 6 7
908  * 0 DY y y y y y
909  * 1 y Y Y Y Y
910  * 2 y Y Y Y Y
911  * 3 y Y Y Y Y
912  * 4 y Y Y Y Y
913  * 5 DU u u u u u
914  * 6 u U U U U
915  * 7 u U U U U
916  * 8 u U U U U
917  * 9 u U U U U
918  * 10 DV v v v v v
919  * 11 v V V V V
920  * 12 v V V V V
921  * 13 v V V V V
922  * 14 v V V V V
923  * DY/DU/DV are for luma/chroma DC.
924  */
925 
926 #define LUMA_DC_BLOCK_INDEX 48
927 #define CHROMA_DC_BLOCK_INDEX 49
928 
929 // This table must be here because scan8[constant] must be known at compiletime
930 static const uint8_t scan8[16 * 3 + 3] = {
931  4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
932  6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
933  4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
934  6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
935  4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
936  6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
937  4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
938  6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
939  4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
940  6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
941  4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
942  6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
943  0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
944 };
945 
946 static av_always_inline uint32_t pack16to32(int a, int b)
947 {
948 #if HAVE_BIGENDIAN
949  return (b & 0xFFFF) + (a << 16);
950 #else
951  return (a & 0xFFFF) + (b << 16);
952 #endif
953 }
954 
955 static av_always_inline uint16_t pack8to16(int a, int b)
956 {
957 #if HAVE_BIGENDIAN
958  return (b & 0xFF) + (a << 8);
959 #else
960  return (a & 0xFF) + (b << 8);
961 #endif
962 }
963 
964 /**
965  * Get the chroma qp.
966  */
968 {
969  return h->pps.chroma_qp_table[t][qscale];
970 }
971 
972 /**
973  * Get the predicted intra4x4 prediction mode.
974  */
976 {
977  const int index8 = scan8[n];
978  const int left = h->intra4x4_pred_mode_cache[index8 - 1];
979  const int top = h->intra4x4_pred_mode_cache[index8 - 8];
980  const int min = FFMIN(left, top);
981 
982  tprintf(h->avctx, "mode:%d %d min:%d\n", left, top, min);
983 
984  if (min < 0)
985  return DC_PRED;
986  else
987  return min;
988 }
989 
991 {
992  int8_t *i4x4 = h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
993  int8_t *i4x4_cache = h->intra4x4_pred_mode_cache;
994 
995  AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
996  i4x4[4] = i4x4_cache[7 + 8 * 3];
997  i4x4[5] = i4x4_cache[7 + 8 * 2];
998  i4x4[6] = i4x4_cache[7 + 8 * 1];
999 }
1000 
1002 {
1003  const int mb_xy = h->mb_xy;
1004  uint8_t *nnz = h->non_zero_count[mb_xy];
1005  uint8_t *nnz_cache = h->non_zero_count_cache;
1006 
1007  AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
1008  AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
1009  AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
1010  AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
1011  AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
1012  AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
1013  AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
1014  AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
1015 
1016  if (!h->chroma_y_shift) {
1017  AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
1018  AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
1019  AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
1020  AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
1021  }
1022 }
1023 
1025  int b_stride,
1026  int b_xy, int b8_xy,
1027  int mb_type, int list)
1028 {
1029  int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
1030  int16_t(*mv_src)[2] = &h->mv_cache[list][scan8[0]];
1031  AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
1032  AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
1033  AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
1034  AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
1035  if (CABAC(h)) {
1036  uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8 * h->mb_xy
1037  : h->mb2br_xy[h->mb_xy]];
1038  uint8_t(*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1039  if (IS_SKIP(mb_type)) {
1040  AV_ZERO128(mvd_dst);
1041  } else {
1042  AV_COPY64(mvd_dst, mvd_src + 8 * 3);
1043  AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
1044  AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
1045  AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
1046  }
1047  }
1048 
1049  {
1050  int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
1051  int8_t *ref_cache = h->ref_cache[list];
1052  ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
1053  ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
1054  ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
1055  ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
1056  }
1057 }
1058 
1059 static av_always_inline void write_back_motion(H264Context *h, int mb_type)
1060 {
1061  const int b_stride = h->b_stride;
1062  const int b_xy = 4 * h->mb_x + 4 * h->mb_y * h->b_stride; // try mb2b(8)_xy
1063  const int b8_xy = 4 * h->mb_xy;
1064 
1065  if (USES_LIST(mb_type, 0)) {
1066  write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 0);
1067  } else {
1068  fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
1069  2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1070  }
1071  if (USES_LIST(mb_type, 1))
1072  write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 1);
1073 
1074  if (h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) {
1075  if (IS_8X8(mb_type)) {
1076  uint8_t *direct_table = &h->direct_table[4 * h->mb_xy];
1077  direct_table[1] = h->sub_mb_type[1] >> 1;
1078  direct_table[2] = h->sub_mb_type[2] >> 1;
1079  direct_table[3] = h->sub_mb_type[3] >> 1;
1080  }
1081  }
1082 }
1083 
1085 {
1087  return !(AV_RN64A(h->sub_mb_type) &
1089  0x0001000100010001ULL));
1090  else
1091  return !(AV_RN64A(h->sub_mb_type) &
1093  0x0001000100010001ULL));
1094 }
1095 
1096 static inline int find_start_code(const uint8_t *buf, int buf_size,
1097  int buf_index, int next_avc)
1098 {
1099  uint32_t state = -1;
1100 
1101  buf_index = avpriv_find_start_code(buf + buf_index, buf + next_avc + 1, &state) - buf - 1;
1102 
1103  return FFMIN(buf_index, buf_size);
1104 }
1105 
1106 static inline int get_avc_nalsize(H264Context *h, const uint8_t *buf,
1107  int buf_size, int *buf_index)
1108 {
1109  int i, nalsize = 0;
1110 
1111  if (*buf_index >= buf_size - h->nal_length_size)
1112  return -1;
1113 
1114  for (i = 0; i < h->nal_length_size; i++)
1115  nalsize = ((unsigned)nalsize << 8) | buf[(*buf_index)++];
1116  if (nalsize <= 0 || nalsize > buf_size - *buf_index) {
1118  "AVC: nal size %d\n", nalsize);
1119  return -1;
1120  }
1121  return nalsize;
1122 }
1123 
1124 int ff_h264_field_end(H264Context *h, int in_setup);
1125 
1128 
1131 
1132 void ff_h264_draw_horiz_band(H264Context *h, int y, int height);
1133 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc);
1136 
1138 #define SLICE_SINGLETHREAD 1
1139 #define SLICE_SKIPED 2
1140 
1141 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count);
1143  const AVCodecContext *src);
1144 
1146 
1147 void ff_h264_free_tables(H264Context *h, int free_rbsp);
1148 
1150 
1151 #endif /* AVCODEC_H264_H */