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;
132 
133 /**
134  * pic_struct in picture timing SEI message
135  */
136 typedef enum {
137  SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
138  SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
139  SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
140  SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
141  SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
142  SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
143  SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
144  SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
145  SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
147 
148 /**
149  * Sequence parameter set
150  */
151 typedef struct SPS {
155  int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
156  int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
157  int poc_type; ///< pic_order_cnt_type
158  int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
162  int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
163  int ref_frame_count; ///< num_ref_frames
165  int mb_width; ///< pic_width_in_mbs_minus1 + 1
166  int mb_height; ///< pic_height_in_map_units_minus1 + 1
168  int mb_aff; ///< mb_adaptive_frame_field_flag
170  int crop; ///< frame_cropping_flag
171 
172  /* those 4 are already in luma samples */
173  unsigned int crop_left; ///< frame_cropping_rect_left_offset
174  unsigned int crop_right; ///< frame_cropping_rect_right_offset
175  unsigned int crop_top; ///< frame_cropping_rect_top_offset
176  unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
187  uint32_t time_scale;
189  short offset_for_ref_frame[256]; // FIXME dyn aloc?
199  int cpb_cnt; ///< See H.264 E.1.2
200  int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
201  int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
202  int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
203  int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
204  int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
205  int residual_color_transform_flag; ///< residual_colour_transform_flag
206  int constraint_set_flags; ///< constraint_set[0-3]_flag
207  int new; ///< flag to keep track if the decoder context needs re-init due to changed SPS
208 } SPS;
209 
210 /**
211  * Picture parameter set
212  */
213 typedef struct PPS {
214  unsigned int sps_id;
215  int cabac; ///< entropy_coding_mode_flag
216  int pic_order_present; ///< pic_order_present_flag
217  int slice_group_count; ///< num_slice_groups_minus1 + 1
219  unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
220  int weighted_pred; ///< weighted_pred_flag
222  int init_qp; ///< pic_init_qp_minus26 + 26
223  int init_qs; ///< pic_init_qs_minus26 + 26
225  int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
226  int constrained_intra_pred; ///< constrained_intra_pred_flag
227  int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
228  int transform_8x8_mode; ///< transform_8x8_mode_flag
231  uint8_t chroma_qp_table[2][QP_MAX_NUM+1]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
233 } PPS;
234 
235 /**
236  * Memory management control operation opcode.
237  */
238 typedef enum MMCOOpcode {
239  MMCO_END = 0,
246 } MMCOOpcode;
247 
248 /**
249  * Memory management control operation.
250  */
251 typedef struct MMCO {
253  int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
254  int long_arg; ///< index, pic_num, or num long refs depending on opcode
255 } MMCO;
256 
257 /**
258  * H264Context
259  */
260 typedef struct H264Context {
271 
275 
276  int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
277  int chroma_qp[2]; // QPc
278 
279  int qp_thresh; ///< QP threshold to skip loopfilter
280 
281  /* coded dimensions -- 16 * mb w/h */
282  int width, height;
285 
286  int qscale;
291 
293  int flags;
295 
298 
299  // prediction stuff
302 
307 
309  int top_type;
312 
315 
320  unsigned int top_samples_available;
323  uint8_t (*top_borders[2])[(16 * 3) * 2];
324 
325  /**
326  * non zero coeff count cache.
327  * is 64 if not available.
328  */
330 
332 
333  /**
334  * Motion vector cache.
335  */
336  DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
337  DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
338 #define LIST_NOT_USED -1 // FIXME rename?
339 #define PART_NOT_AVAILABLE -2
340 
341  /**
342  * number of neighbors (top and/or left) that used 8x8 dct
343  */
345 
346  /**
347  * block_offset[ 0..23] for frame macroblocks
348  * block_offset[24..47] for field macroblocks
349  */
350  int block_offset[2 * (16 * 3)];
351 
352  uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
353  uint32_t *mb2br_xy;
354  int b_stride; // FIXME use s->b4_stride
355 
356  int mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
358 
359  unsigned current_sps_id; ///< id of the current SPS
360  SPS sps; ///< current sps
361 
362  /**
363  * current pps
364  */
365  PPS pps; // FIXME move to Picture perhaps? (->no) do we need that?
366 
367  uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
368  uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
369  uint32_t(*dequant4_coeff[6])[16];
370  uint32_t(*dequant8_coeff[6])[64];
371 
373  uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
375  int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
377 
378  // interlacing specific flags
381  int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
384 
385  DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
386 
387  // Weighted pred stuff
392  // The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
393  int luma_weight[48][2][2];
394  int chroma_weight[48][2][2][2];
395  int implicit_weight[48][48][2];
396 
402  int map_col_to_list0[2][16 + 32];
403  int map_col_to_list0_field[2][2][16 + 32];
404 
405  /**
406  * num_ref_idx_l0/1_active_minus1 + 1
407  */
408  unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
409  unsigned int list_count;
410  uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
411  Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
412  * Reordered version of default_ref_list
413  * according to picture reordering in slice header */
414  int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
415 
416  // data partitioning
421 
423  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.
424  DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
425  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
426 
427  /**
428  * Cabac
429  */
432 
433  /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
434  uint16_t *cbp_table;
435  int cbp;
436  int top_cbp;
437  int left_cbp;
438  /* chroma_pred_mode for i4x4 or i16x16, else 0 */
441  uint8_t (*mvd_table[2])[2];
442  DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
445 
458 
460 
461  int mb_x, mb_y;
467  int mb_num;
468  int mb_xy;
469 
471 
472  // deblock
473  int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
476 
477  // =============================================================
478  // Things below are not used in the MB or more inner code
479 
483  unsigned int rbsp_buffer_size[2];
484 
485  /**
486  * Used to parse AVC variant of h264
487  */
488  int is_avc; ///< this flag is != 0 if codec is avc1
489  int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
490  int got_first; ///< this flag is != 0 if we've parsed a frame
491 
492  int bit_depth_luma; ///< luma bit depth from sps to detect changes
493  int chroma_format_idc; ///< chroma format from sps to detect changes
494 
497 
498  int dequant_coeff_pps; ///< reinit tables when pps changes
499 
500  uint16_t *slice_table_base;
501 
502  // POC stuff
503  int poc_lsb;
504  int poc_msb;
506  int delta_poc[2];
508  int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
509  int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
510  int frame_num_offset; ///< for POC type 2
511  int prev_frame_num_offset; ///< for POC type 2
512  int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
513 
514  /**
515  * frame_num for frames or 2 * frame_num + 1 for field pics.
516  */
518 
519  /**
520  * max_frame_num or 2 * max_frame_num for field pics.
521  */
523 
525 
526  Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
534 
535  /**
536  * memory management control operations buffer.
537  */
541 
542  int long_ref_count; ///< number of actual long term references
543  int short_ref_count; ///< number of actual short term references
544 
546 
547  /**
548  * @name Members for slice based multithreading
549  * @{
550  */
552 
553  /**
554  * current slice number, used to initialize slice_num of each thread/context
555  */
557 
558  /**
559  * Max number of threads / contexts.
560  * This is equal to AVCodecContext.thread_count unless
561  * multithreaded decoding is impossible, in which case it is
562  * reduced to 1.
563  */
565 
567 
568  /**
569  * 1 if the single thread fallback warning has already been
570  * displayed, 0 otherwise.
571  */
573 
575 
577  unsigned int last_ref_count[2];
578  /** @} */
579 
580  /**
581  * pic_struct in picture timing SEI message
582  */
584 
585  /**
586  * Complement sei_pic_struct
587  * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
588  * However, soft telecined frames may have these values.
589  * This is used in an attempt to flag soft telecine progressive.
590  */
592 
593  /**
594  * Bit set of clock types for fields/frames in picture timing SEI message.
595  * For each found ct_type, appropriate bit is set (e.g., bit 1 for
596  * interlaced).
597  */
599 
600  /**
601  * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
602  */
604 
605  /**
606  * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
607  */
609 
610  /**
611  * recovery_frame_cnt from SEI message
612  *
613  * Set to -1 if no recovery point SEI message found or to number of frames
614  * before playback synchronizes. Frames having recovery point are key
615  * frames.
616  */
618  /**
619  * recovery_frame is the frame_num at which the next frame should
620  * be fully constructed.
621  *
622  * Set to -1 when not expecting a recovery point.
623  */
625 
626  /**
627  * Are the SEI recovery points looking valid.
628  */
630 
631  int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
632  int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
633 
634  // Timestamp stuff
635  int sei_buffering_period_present; ///< Buffering period SEI flag
636  int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
637 
640 
641  int16_t slice_row[MAX_SLICES]; ///< to detect when MAX_SLICES is too low
642 
643  int sync; ///< did we had a keyframe or recovery point
644 
649  int16_t *dc_val_base;
650 
651  uint8_t *visualization_buffer[3]; ///< temporary buffer vor MV visualization
652 
657 } H264Context;
658 
659 extern const uint8_t ff_h264_chroma_qp[7][QP_MAX_NUM + 1]; ///< One chroma qp table for each possible bit depth (8-14).
660 extern const uint16_t ff_h264_mb_sizes[4];
661 
662 /**
663  * Decode SEI
664  */
666 
667 /**
668  * Decode SPS
669  */
671 
672 /**
673  * compute profile from sps
674  */
676 
677 /**
678  * Decode PPS
679  */
680 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
681 
682 /**
683  * Decode a network abstraction layer unit.
684  * @param consumed is the number of bytes used as input
685  * @param length is the length of the array
686  * @param dst_length is the number of decoded bytes FIXME here
687  * or a decode rbsp tailing?
688  * @return decoded bytes, might be src+1 if no escapes
689  */
691  int *dst_length, int *consumed, int length);
692 
693 /**
694  * Free any data that may have been allocated in the H264 context
695  * like SPS, PPS etc.
696  */
698 
699 /**
700  * Reconstruct bitstream slice_type.
701  */
702 int ff_h264_get_slice_type(const H264Context *h);
703 
704 /**
705  * Allocate tables.
706  * needs width/height
707  */
709 
710 /**
711  * Fill the default_ref_list.
712  */
714 
718 
719 /**
720  * Execute the reference picture marking (memory management control operations).
721  */
722 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
723 
725  int first_slice);
726 
727 int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
728 
729 /**
730  * Check if the top & left blocks are available if needed & change the
731  * dc mode so it only uses the available blocks.
732  */
734 
735 /**
736  * Check if the top & left blocks are available if needed & change the
737  * dc mode so it only uses the available blocks.
738  */
739 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma);
740 
744 void ff_h264_decode_init_vlc(void);
745 
746 /**
747  * Decode a macroblock
748  * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
749  */
751 
752 /**
753  * Decode a CABAC coded macroblock
754  * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
755  */
757 
759 
762 void ff_h264_pred_direct_motion(H264Context *const h, int *mb_type);
763 
764 void ff_h264_filter_mb_fast(H264Context *h, int mb_x, int mb_y,
765  uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
766  unsigned int linesize, unsigned int uvlinesize);
767 void ff_h264_filter_mb(H264Context *h, int mb_x, int mb_y,
768  uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
769  unsigned int linesize, unsigned int uvlinesize);
770 
771 /**
772  * Reset SEI values at the beginning of the frame.
773  *
774  * @param h H.264 context.
775  */
777 
778 /*
779  * o-o o-o
780  * / / /
781  * o-o o-o
782  * ,---'
783  * o-o o-o
784  * / / /
785  * o-o o-o
786  */
787 
788 /* Scan8 organization:
789  * 0 1 2 3 4 5 6 7
790  * 0 DY y y y y y
791  * 1 y Y Y Y Y
792  * 2 y Y Y Y Y
793  * 3 y Y Y Y Y
794  * 4 y Y Y Y Y
795  * 5 DU u u u u u
796  * 6 u U U U U
797  * 7 u U U U U
798  * 8 u U U U U
799  * 9 u U U U U
800  * 10 DV v v v v v
801  * 11 v V V V V
802  * 12 v V V V V
803  * 13 v V V V V
804  * 14 v V V V V
805  * DY/DU/DV are for luma/chroma DC.
806  */
807 
808 #define LUMA_DC_BLOCK_INDEX 48
809 #define CHROMA_DC_BLOCK_INDEX 49
810 
811 // This table must be here because scan8[constant] must be known at compiletime
812 static const uint8_t scan8[16 * 3 + 3] = {
813  4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
814  6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
815  4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
816  6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
817  4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
818  6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
819  4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
820  6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
821  4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
822  6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
823  4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
824  6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
825  0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
826 };
827 
828 static av_always_inline uint32_t pack16to32(int a, int b)
829 {
830 #if HAVE_BIGENDIAN
831  return (b & 0xFFFF) + (a << 16);
832 #else
833  return (a & 0xFFFF) + (b << 16);
834 #endif
835 }
836 
837 static av_always_inline uint16_t pack8to16(int a, int b)
838 {
839 #if HAVE_BIGENDIAN
840  return (b & 0xFF) + (a << 8);
841 #else
842  return (a & 0xFF) + (b << 8);
843 #endif
844 }
845 
846 /**
847  * Get the chroma qp.
848  */
850 {
851  return h->pps.chroma_qp_table[t][qscale];
852 }
853 
854 /**
855  * Get the predicted intra4x4 prediction mode.
856  */
858 {
859  const int index8 = scan8[n];
860  const int left = h->intra4x4_pred_mode_cache[index8 - 1];
861  const int top = h->intra4x4_pred_mode_cache[index8 - 8];
862  const int min = FFMIN(left, top);
863 
864  tprintf(h->avctx, "mode:%d %d min:%d\n", left, top, min);
865 
866  if (min < 0)
867  return DC_PRED;
868  else
869  return min;
870 }
871 
873 {
874  int8_t *i4x4 = h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
875  int8_t *i4x4_cache = h->intra4x4_pred_mode_cache;
876 
877  AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
878  i4x4[4] = i4x4_cache[7 + 8 * 3];
879  i4x4[5] = i4x4_cache[7 + 8 * 2];
880  i4x4[6] = i4x4_cache[7 + 8 * 1];
881 }
882 
884 {
885  const int mb_xy = h->mb_xy;
886  uint8_t *nnz = h->non_zero_count[mb_xy];
887  uint8_t *nnz_cache = h->non_zero_count_cache;
888 
889  AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
890  AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
891  AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
892  AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
893  AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
894  AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
895  AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
896  AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
897 
898  if (!h->chroma_y_shift) {
899  AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
900  AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
901  AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
902  AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
903  }
904 }
905 
907  int b_stride,
908  int b_xy, int b8_xy,
909  int mb_type, int list)
910 {
911  int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
912  int16_t(*mv_src)[2] = &h->mv_cache[list][scan8[0]];
913  AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
914  AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
915  AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
916  AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
917  if (CABAC(h)) {
918  uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8 * h->mb_xy
919  : h->mb2br_xy[h->mb_xy]];
920  uint8_t(*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
921  if (IS_SKIP(mb_type)) {
922  AV_ZERO128(mvd_dst);
923  } else {
924  AV_COPY64(mvd_dst, mvd_src + 8 * 3);
925  AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
926  AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
927  AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
928  }
929  }
930 
931  {
932  int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
933  int8_t *ref_cache = h->ref_cache[list];
934  ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
935  ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
936  ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
937  ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
938  }
939 }
940 
941 static av_always_inline void write_back_motion(H264Context *h, int mb_type)
942 {
943  const int b_stride = h->b_stride;
944  const int b_xy = 4 * h->mb_x + 4 * h->mb_y * h->b_stride; // try mb2b(8)_xy
945  const int b8_xy = 4 * h->mb_xy;
946 
947  if (USES_LIST(mb_type, 0)) {
948  write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 0);
949  } else {
950  fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
951  2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
952  }
953  if (USES_LIST(mb_type, 1))
954  write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 1);
955 
956  if (h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) {
957  if (IS_8X8(mb_type)) {
958  uint8_t *direct_table = &h->direct_table[4 * h->mb_xy];
959  direct_table[1] = h->sub_mb_type[1] >> 1;
960  direct_table[2] = h->sub_mb_type[2] >> 1;
961  direct_table[3] = h->sub_mb_type[3] >> 1;
962  }
963  }
964 }
965 
967 {
969  return !(AV_RN64A(h->sub_mb_type) &
971  0x0001000100010001ULL));
972  else
973  return !(AV_RN64A(h->sub_mb_type) &
975  0x0001000100010001ULL));
976 }
977 
978 void ff_h264_draw_horiz_band(H264Context *h, int y, int height);
979 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc);
980 
981 #endif /* AVCODEC_H264_H */