2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
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.
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.
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
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #ifndef AVCODEC_H264_H
29 #define AVCODEC_H264_H
31 #include "libavutil/intreadwrite.h"
32 #include "libavutil/thread.h"
34 #include "error_resilience.h"
36 #include "h264_parse.h"
37 #include "h264chroma.h"
42 #include "mpegutils.h"
45 #include "rectangle.h"
48 #define H264_MAX_PICTURE_COUNT 36
49 #define H264_MAX_THREADS 32
51 #define MAX_SPS_COUNT 32
52 #define MAX_PPS_COUNT 256
54 #define MAX_MMCO_COUNT 66
56 #define MAX_DELAYED_PIC_COUNT 16
58 #define MAX_MBPAIR_SIZE (256*1024) // a tighter bound could be calculated if someone cares about a few bytes
60 /* Compiling in interlaced support reduces the speed
61 * of progressive decoding by about 2%. */
62 #define ALLOW_INTERLACE
67 * The maximum number of slices supported by the decoder.
68 * must be a power of 2
72 #ifdef ALLOW_INTERLACE
73 #define MB_MBAFF(h) (h)->mb_mbaff
74 #define MB_FIELD(sl) (sl)->mb_field_decoding_flag
75 #define FRAME_MBAFF(h) (h)->mb_aff_frame
76 #define FIELD_PICTURE(h) ((h)->picture_structure != PICT_FRAME)
83 #define MB_FIELD(sl) 0
84 #define FRAME_MBAFF(h) 0
85 #define FIELD_PICTURE(h) 0
87 #define IS_INTERLACED(mb_type) 0
93 #define FIELD_OR_MBAFF_PICTURE(h) (FRAME_MBAFF(h) || FIELD_PICTURE(h))
96 #define CABAC(h) (h)->pps.cabac
99 #define CHROMA(h) ((h)->sps.chroma_format_idc)
100 #define CHROMA422(h) ((h)->sps.chroma_format_idc == 2)
101 #define CHROMA444(h) ((h)->sps.chroma_format_idc == 3)
103 #define EXTENDED_SAR 255
105 #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit
106 #define MB_TYPE_8x8DCT 0x01000000
107 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
108 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
110 #define QP_MAX_NUM (51 + 6*6) // The maximum supported qp
123 NAL_END_SEQUENCE = 10,
125 NAL_FILLER_DATA = 12,
127 NAL_AUXILIARY_SLICE = 19,
128 NAL_FF_IGNORE = 0xff0f001,
135 SEI_TYPE_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
136 SEI_TYPE_PIC_TIMING = 1, ///< picture timing
137 SEI_TYPE_USER_DATA_REGISTERED = 4, ///< registered user data as specified by Rec. ITU-T T.35
138 SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
139 SEI_TYPE_RECOVERY_POINT = 6, ///< recovery point (frame # to decoder sync)
140 SEI_TYPE_FRAME_PACKING = 45, ///< frame packing arrangement
141 SEI_TYPE_DISPLAY_ORIENTATION = 47, ///< display orientation
142 SEI_TYPE_GREEN_METADATA = 56 ///< GreenMPEG information
146 * pic_struct in picture timing SEI message
149 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
150 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
151 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
152 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
153 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
154 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
155 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
156 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
157 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
161 * frame_packing_arrangement types
164 SEI_FPA_TYPE_CHECKERBOARD = 0,
165 SEI_FPA_TYPE_INTERLEAVE_COLUMN = 1,
166 SEI_FPA_TYPE_INTERLEAVE_ROW = 2,
167 SEI_FPA_TYPE_SIDE_BY_SIDE = 3,
168 SEI_FPA_TYPE_TOP_BOTTOM = 4,
169 SEI_FPA_TYPE_INTERLEAVE_TEMPORAL = 5,
174 * Sequence parameter set
180 int chroma_format_idc;
181 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
182 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
183 int poc_type; ///< pic_order_cnt_type
184 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
185 int delta_pic_order_always_zero_flag;
186 int offset_for_non_ref_pic;
187 int offset_for_top_to_bottom_field;
188 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
189 int ref_frame_count; ///< num_ref_frames
190 int gaps_in_frame_num_allowed_flag;
191 int mb_width; ///< pic_width_in_mbs_minus1 + 1
192 int mb_height; ///< pic_height_in_map_units_minus1 + 1
193 int frame_mbs_only_flag;
194 int mb_aff; ///< mb_adaptive_frame_field_flag
195 int direct_8x8_inference_flag;
196 int crop; ///< frame_cropping_flag
198 /* those 4 are already in luma samples */
199 unsigned int crop_left; ///< frame_cropping_rect_left_offset
200 unsigned int crop_right; ///< frame_cropping_rect_right_offset
201 unsigned int crop_top; ///< frame_cropping_rect_top_offset
202 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
203 int vui_parameters_present_flag;
205 int video_signal_type_present_flag;
207 int colour_description_present_flag;
208 enum AVColorPrimaries color_primaries;
209 enum AVColorTransferCharacteristic color_trc;
210 enum AVColorSpace colorspace;
211 int timing_info_present_flag;
212 uint32_t num_units_in_tick;
214 int fixed_frame_rate_flag;
215 short offset_for_ref_frame[256]; // FIXME dyn aloc?
216 int bitstream_restriction_flag;
217 int num_reorder_frames;
218 int scaling_matrix_present;
219 uint8_t scaling_matrix4[6][16];
220 uint8_t scaling_matrix8[6][64];
221 int nal_hrd_parameters_present_flag;
222 int vcl_hrd_parameters_present_flag;
223 int pic_struct_present_flag;
224 int time_offset_length;
225 int cpb_cnt; ///< See H.264 E.1.2
226 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
227 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
228 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
229 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
230 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
231 int residual_color_transform_flag; ///< residual_colour_transform_flag
232 int constraint_set_flags; ///< constraint_set[0-3]_flag
233 int new; ///< flag to keep track if the decoder context needs re-init due to changed SPS
239 * Picture parameter set
243 int cabac; ///< entropy_coding_mode_flag
244 int pic_order_present; ///< pic_order_present_flag
245 int slice_group_count; ///< num_slice_groups_minus1 + 1
246 int mb_slice_group_map_type;
247 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
248 int weighted_pred; ///< weighted_pred_flag
249 int weighted_bipred_idc;
250 int init_qp; ///< pic_init_qp_minus26 + 26
251 int init_qs; ///< pic_init_qs_minus26 + 26
252 int chroma_qp_index_offset[2];
253 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
254 int constrained_intra_pred; ///< constrained_intra_pred_flag
255 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
256 int transform_8x8_mode; ///< transform_8x8_mode_flag
257 uint8_t scaling_matrix4[6][16];
258 uint8_t scaling_matrix8[6][64];
259 uint8_t chroma_qp_table[2][QP_MAX_NUM+1]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
266 * Frame Packing Arrangement Type
269 int frame_packing_arrangement_id;
270 int frame_packing_arrangement_cancel_flag; ///< is previous arrangement canceled, -1 if never received
271 SEI_FpaType frame_packing_arrangement_type;
272 int frame_packing_arrangement_repetition_period;
273 int content_interpretation_type;
274 int quincunx_sampling_flag;
278 * Green MetaData Information Type
280 typedef struct GreenMetaData {
281 uint8_t green_metadata_type;
283 uint16_t num_seconds;
284 uint16_t num_pictures;
285 uint8_t percent_non_zero_macroblocks;
286 uint8_t percent_intra_coded_macroblocks;
287 uint8_t percent_six_tap_filtering;
288 uint8_t percent_alpha_point_deblocking_instance;
289 uint8_t xsd_metric_type;
290 uint16_t xsd_metric_value;
294 * Memory management control operation opcode.
296 typedef enum MMCOOpcode {
307 * Memory management control operation.
309 typedef struct MMCO {
311 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
312 int long_arg; ///< index, pic_num, or num long refs depending on opcode
315 typedef struct H264Picture {
319 AVBufferRef *qscale_table_buf;
320 int8_t *qscale_table;
322 AVBufferRef *motion_val_buf[2];
323 int16_t (*motion_val[2])[2];
325 AVBufferRef *mb_type_buf;
328 AVBufferRef *hwaccel_priv_buf;
329 void *hwaccel_picture_private; ///< hardware accelerator private data
331 AVBufferRef *ref_index_buf[2];
332 int8_t *ref_index[2];
334 int field_poc[2]; ///< top/bottom POC
335 int poc; ///< frame POC
336 int frame_num; ///< frame_num (raw frame_num from slice header)
337 int mmco_reset; /**< MMCO_RESET set this 1. Reordering code must
338 not mix pictures before and after MMCO_RESET. */
339 int pic_id; /**< pic_num (short -> no wrap version of pic_num,
340 pic_num & max_pic_num; long -> long_pic_num) */
341 int long_ref; ///< 1->long term reference 0->short term reference
342 int ref_poc[2][2][32]; ///< POCs of the frames/fields used as reference (FIXME need per slice)
343 int ref_count[2][2]; ///< number of entries in ref_poc (FIXME need per slice)
344 int mbaff; ///< 1 -> MBAFF frame 0-> not MBAFF
345 int field_picture; ///< whether or not picture was encoded in separate fields
348 int recovered; ///< picture at IDR or recovery point + recovery count
350 int sei_recovery_frame_cnt;
357 typedef struct H264Ref {
368 typedef struct H264SliceContext {
369 struct H264Context *h264;
375 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
376 int slice_type_fixed;
379 int chroma_qp[2]; // QPc
380 int qp_thresh; ///< QP threshold to skip loopfilter
381 int last_qscale_diff;
384 int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
385 int slice_alpha_c0_offset;
386 int slice_beta_offset;
388 H264PredWeightTable pwt;
393 int chroma_pred_mode;
394 int intra16x16_pred_mode;
396 int8_t intra4x4_pred_mode_cache[5 * 8];
397 int8_t(*intra4x4_pred_mode);
402 int left_mb_xy[LEFT_MBS];
407 int left_type[LEFT_MBS];
409 const uint8_t *left_block;
410 int topleft_partition;
412 unsigned int topleft_samples_available;
413 unsigned int top_samples_available;
414 unsigned int topright_samples_available;
415 unsigned int left_samples_available;
417 ptrdiff_t linesize, uvlinesize;
418 ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
419 ptrdiff_t mb_uvlinesize;
425 // index of the first MB of the next slice
430 int mb_field_decoding_flag;
431 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
433 int redundant_pic_count;
436 * number of neighbors (top and/or left) that used 8x8 dct
438 int neighbor_transform_size;
440 int direct_spatial_mv_pred;
448 int dist_scale_factor[32];
449 int dist_scale_factor_field[2][32];
450 int map_col_to_list0[2][16 + 32];
451 int map_col_to_list0_field[2][2][16 + 32];
454 * num_ref_idx_l0/1_active_minus1 + 1
456 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
457 unsigned int list_count;
458 H264Ref ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
459 * Reordered version of default_ref_list
460 * according to picture reordering in slice header */
461 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
463 const uint8_t *intra_pcm_ptr;
464 int16_t *dc_val_base;
466 uint8_t *bipred_scratchpad;
467 uint8_t *edge_emu_buffer;
468 uint8_t (*top_borders[2])[(16 * 3) * 2];
469 int bipred_scratchpad_allocated;
470 int edge_emu_buffer_allocated;
471 int top_borders_allocated[2];
474 * non zero coeff count cache.
475 * is 64 if not available.
477 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
480 * Motion vector cache.
482 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
483 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
484 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
485 uint8_t direct_cache[5 * 8];
487 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
489 ///< as a dct coefficient is int32_t in high depth, we need to reserve twice the space.
490 DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2];
491 DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
492 ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either
493 ///< check that i is not too large or ensure that there is some unused stuff after mb
494 int16_t mb_padding[256 * 2];
496 uint8_t (*mvd_table[2])[2];
502 uint8_t cabac_state[1024];
505 // rbsp buffer used for this slice
506 uint8_t *rbsp_buffer;
507 unsigned int rbsp_buffer_size;
513 typedef struct H264Context {
514 const AVClass *class;
515 AVCodecContext *avctx;
516 VideoDSPContext vdsp;
517 H264DSPContext h264dsp;
518 H264ChromaContext h264chroma;
519 H264QpelContext h264qpel;
522 H264Picture DPB[H264_MAX_PICTURE_COUNT];
523 H264Picture *cur_pic_ptr;
525 H264Picture last_pic_for_ec;
527 H264SliceContext *slice_ctx;
530 int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
532 /* coded dimensions -- 16 * mb w/h */
534 int chroma_x_shift, chroma_y_shift;
537 * Backup frame properties: needed, because they can be different
538 * between returned frame and last decoded frame.
542 enum AVPixelFormat backup_pix_fmt;
545 int coded_picture_number;
548 int context_initialized;
552 int8_t(*intra4x4_pred_mode);
555 uint8_t (*non_zero_count)[48];
557 #define LIST_NOT_USED -1 // FIXME rename?
558 #define PART_NOT_AVAILABLE -2
561 * block_offset[ 0..23] for frame macroblocks
562 * block_offset[24..47] for field macroblocks
564 int block_offset[2 * (16 * 3)];
566 uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
568 int b_stride; // FIXME use s->b4_stride
571 unsigned current_sps_id; ///< id of the current SPS
572 SPS sps; ///< current sps
573 PPS pps; ///< current pps
575 int au_pps_id; ///< pps_id of current access unit
577 uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
578 uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
579 uint32_t(*dequant4_coeff[6])[16];
580 uint32_t(*dequant8_coeff[6])[64];
582 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
584 // interlacing specific flags
586 int picture_structure;
589 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
591 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
594 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
595 uint8_t *chroma_pred_mode_table;
596 uint8_t (*mvd_table[2])[2];
597 uint8_t *direct_table;
599 uint8_t zigzag_scan[16];
600 uint8_t zigzag_scan8x8[64];
601 uint8_t zigzag_scan8x8_cavlc[64];
602 uint8_t field_scan[16];
603 uint8_t field_scan8x8[64];
604 uint8_t field_scan8x8_cavlc[64];
605 uint8_t zigzag_scan_q0[16];
606 uint8_t zigzag_scan8x8_q0[64];
607 uint8_t zigzag_scan8x8_cavlc_q0[64];
608 uint8_t field_scan_q0[16];
609 uint8_t field_scan8x8_q0[64];
610 uint8_t field_scan8x8_cavlc_q0[64];
615 int mb_height, mb_width;
619 // =============================================================
620 // Things below are not used in the MB or more inner code
626 * Used to parse AVC variant of h264
628 int is_avc; ///< this flag is != 0 if codec is avc1
629 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
631 int bit_depth_luma; ///< luma bit depth from sps to detect changes
632 int chroma_format_idc; ///< chroma format from sps to detect changes
634 SPS *sps_buffers[MAX_SPS_COUNT];
635 PPS *pps_buffers[MAX_PPS_COUNT];
637 int dequant_coeff_pps; ///< reinit tables when pps changes
639 uint16_t *slice_table_base;
644 int delta_poc_bottom;
647 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
648 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
649 int frame_num_offset; ///< for POC type 2
650 int prev_frame_num_offset; ///< for POC type 2
651 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
654 * frame_num for frames or 2 * frame_num + 1 for field pics.
659 * max_frame_num or 2 * max_frame_num for field pics.
663 H264Ref default_ref[2];
664 H264Picture *short_ref[32];
665 H264Picture *long_ref[32];
666 H264Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
667 int last_pocs[MAX_DELAYED_PIC_COUNT];
668 H264Picture *next_output_pic;
669 int next_outputed_poc;
672 * memory management control operations buffer.
674 MMCO mmco[MAX_MMCO_COUNT];
678 int long_ref_count; ///< number of actual long term references
679 int short_ref_count; ///< number of actual short term references
682 * @name Members for slice based multithreading
686 * current slice number, used to initialize slice_num of each thread/context
691 * Max number of threads / contexts.
692 * This is equal to AVCodecContext.thread_count unless
693 * multithreaded decoding is impossible, in which case it is
698 int slice_context_count;
701 * 1 if the single thread fallback warning has already been
702 * displayed, 0 otherwise.
704 int single_decode_warning;
706 enum AVPictureType pict_type;
711 * pic_struct in picture timing SEI message
713 SEI_PicStructType sei_pic_struct;
716 * Complement sei_pic_struct
717 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
718 * However, soft telecined frames may have these values.
719 * This is used in an attempt to flag soft telecine progressive.
721 int prev_interlaced_frame;
724 * frame_packing_arrangment SEI message
726 int sei_frame_packing_present;
727 int frame_packing_arrangement_type;
728 int content_interpretation_type;
729 int quincunx_subsampling;
732 * display orientation SEI message
734 int sei_display_orientation_present;
735 int sei_anticlockwise_rotation;
736 int sei_hflip, sei_vflip;
739 * User data registered by Rec. ITU-T T.35 SEI
741 int sei_reguserdata_afd_present;
742 uint8_t active_format_description;
743 int a53_caption_size;
744 uint8_t *a53_caption;
747 * Bit set of clock types for fields/frames in picture timing SEI message.
748 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
754 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
756 int sei_dpb_output_delay;
759 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
761 int sei_cpb_removal_delay;
764 * recovery_frame_cnt from SEI message
766 * Set to -1 if no recovery point SEI message found or to number of frames
767 * before playback synchronizes. Frames having recovery point are key
770 int sei_recovery_frame_cnt;
773 * Are the SEI recovery points looking valid.
775 int valid_recovery_point;
780 * recovery_frame is the frame_num at which the next frame should
781 * be fully constructed.
783 * Set to -1 when not expecting a recovery point.
788 * We have seen an IDR, so all the following frames in coded order are correctly
791 #define FRAME_RECOVERED_IDR (1 << 0)
793 * Sufficient number of frames have been decoded since a SEI recovery point,
794 * so all the following frames in presentation order are correct.
796 #define FRAME_RECOVERED_SEI (1 << 1)
798 int frame_recovered; ///< Initial frame has been completely recovered
800 int has_recovery_point;
804 /* for frame threading, this is set to 1
805 * after finish_setup() has been called, so we cannot modify
806 * some context properties (which are supposed to stay constant between
811 int sei_buffering_period_present; ///< Buffering period SEI flag
812 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
814 int cur_chroma_format_idc;
815 int cur_bit_depth_luma;
816 int16_t slice_row[MAX_SLICES]; ///< to detect when MAX_SLICES is too low
818 uint8_t parse_history[6];
819 int parse_history_count;
824 AVBufferPool *qscale_table_pool;
825 AVBufferPool *mb_type_pool;
826 AVBufferPool *motion_val_pool;
827 AVBufferPool *ref_index_pool;
829 /* Motion Estimation */
830 qpel_mc_func (*qpel_put)[16];
831 qpel_mc_func (*qpel_avg)[16];
834 GreenMetaData sei_green_metadata;
838 extern const uint8_t ff_h264_chroma_qp[7][QP_MAX_NUM + 1]; ///< One chroma qp table for each possible bit depth (8-14).
839 extern const uint16_t ff_h264_mb_sizes[4];
844 int ff_h264_decode_sei(H264Context *h);
849 int ff_h264_decode_seq_parameter_set(H264Context *h, int ignore_truncation);
852 * compute profile from sps
854 int ff_h264_get_profile(SPS *sps);
859 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
862 * Decode a network abstraction layer unit.
863 * @param consumed is the number of bytes used as input
864 * @param length is the length of the array
865 * @param dst_length is the number of decoded bytes FIXME here
866 * or a decode rbsp tailing?
867 * @return decoded bytes, might be src+1 if no escapes
869 const uint8_t *ff_h264_decode_nal(H264Context *h, H264SliceContext *sl, const uint8_t *src,
870 int *dst_length, int *consumed, int length);
873 * Free any data that may have been allocated in the H264 context
876 void ff_h264_free_context(H264Context *h);
879 * Reconstruct bitstream slice_type.
881 int ff_h264_get_slice_type(const H264SliceContext *sl);
887 int ff_h264_alloc_tables(H264Context *h);
889 int ff_h264_decode_ref_pic_list_reordering(H264Context *h, H264SliceContext *sl);
890 void ff_h264_fill_mbaff_ref_list(H264Context *h, H264SliceContext *sl);
891 void ff_h264_remove_all_refs(H264Context *h);
894 * Execute the reference picture marking (memory management control operations).
896 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
898 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb,
901 int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
904 * Check if the top & left blocks are available if needed & change the
905 * dc mode so it only uses the available blocks.
907 int ff_h264_check_intra4x4_pred_mode(const H264Context *h, H264SliceContext *sl);
910 * Check if the top & left blocks are available if needed & change the
911 * dc mode so it only uses the available blocks.
913 int ff_h264_check_intra_pred_mode(const H264Context *h, H264SliceContext *sl,
914 int mode, int is_chroma);
916 void ff_h264_hl_decode_mb(const H264Context *h, H264SliceContext *sl);
917 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size);
918 int ff_h264_decode_init(AVCodecContext *avctx);
919 void ff_h264_decode_init_vlc(void);
922 * Decode a macroblock
923 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
925 int ff_h264_decode_mb_cavlc(const H264Context *h, H264SliceContext *sl);
928 * Decode a CABAC coded macroblock
929 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
931 int ff_h264_decode_mb_cabac(const H264Context *h, H264SliceContext *sl);
933 void ff_h264_init_cabac_states(const H264Context *h, H264SliceContext *sl);
935 void ff_h264_init_dequant_tables(H264Context *h);
937 void ff_h264_direct_dist_scale_factor(const H264Context *const h, H264SliceContext *sl);
938 void ff_h264_direct_ref_list_init(const H264Context *const h, H264SliceContext *sl);
939 void ff_h264_pred_direct_motion(const H264Context *const h, H264SliceContext *sl,
942 void ff_h264_filter_mb_fast(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
943 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
944 unsigned int linesize, unsigned int uvlinesize);
945 void ff_h264_filter_mb(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
946 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
947 unsigned int linesize, unsigned int uvlinesize);
950 * Reset SEI values at the beginning of the frame.
952 * @param h H.264 context.
954 void ff_h264_reset_sei(H264Context *h);
957 * Get stereo_mode string from the h264 frame_packing_arrangement
958 * @param h H.264 context.
960 const char* ff_h264_sei_stereo_mode(H264Context *h);
972 /* Scan8 organization:
989 * DY/DU/DV are for luma/chroma DC.
992 #define LUMA_DC_BLOCK_INDEX 48
993 #define CHROMA_DC_BLOCK_INDEX 49
995 // This table must be here because scan8[constant] must be known at compiletime
996 static const uint8_t scan8[16 * 3 + 3] = {
997 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
998 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
999 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
1000 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
1001 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
1002 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
1003 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
1004 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
1005 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
1006 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
1007 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
1008 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
1009 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
1012 static av_always_inline uint32_t pack16to32(unsigned a, unsigned b)
1015 return (b & 0xFFFF) + (a << 16);
1017 return (a & 0xFFFF) + (b << 16);
1021 static av_always_inline uint16_t pack8to16(unsigned a, unsigned b)
1024 return (b & 0xFF) + (a << 8);
1026 return (a & 0xFF) + (b << 8);
1031 * Get the chroma qp.
1033 static av_always_inline int get_chroma_qp(const H264Context *h, int t, int qscale)
1035 return h->pps.chroma_qp_table[t][qscale];
1039 * Get the predicted intra4x4 prediction mode.
1041 static av_always_inline int pred_intra_mode(const H264Context *h,
1042 H264SliceContext *sl, int n)
1044 const int index8 = scan8[n];
1045 const int left = sl->intra4x4_pred_mode_cache[index8 - 1];
1046 const int top = sl->intra4x4_pred_mode_cache[index8 - 8];
1047 const int min = FFMIN(left, top);
1049 ff_tlog(h->avctx, "mode:%d %d min:%d\n", left, top, min);
1057 static av_always_inline void write_back_intra_pred_mode(const H264Context *h,
1058 H264SliceContext *sl)
1060 int8_t *i4x4 = sl->intra4x4_pred_mode + h->mb2br_xy[sl->mb_xy];
1061 int8_t *i4x4_cache = sl->intra4x4_pred_mode_cache;
1063 AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
1064 i4x4[4] = i4x4_cache[7 + 8 * 3];
1065 i4x4[5] = i4x4_cache[7 + 8 * 2];
1066 i4x4[6] = i4x4_cache[7 + 8 * 1];
1069 static av_always_inline void write_back_non_zero_count(const H264Context *h,
1070 H264SliceContext *sl)
1072 const int mb_xy = sl->mb_xy;
1073 uint8_t *nnz = h->non_zero_count[mb_xy];
1074 uint8_t *nnz_cache = sl->non_zero_count_cache;
1076 AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
1077 AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
1078 AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
1079 AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
1080 AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
1081 AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
1082 AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
1083 AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
1085 if (!h->chroma_y_shift) {
1086 AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
1087 AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
1088 AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
1089 AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
1093 static av_always_inline void write_back_motion_list(const H264Context *h,
1094 H264SliceContext *sl,
1096 int b_xy, int b8_xy,
1097 int mb_type, int list)
1099 int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
1100 int16_t(*mv_src)[2] = &sl->mv_cache[list][scan8[0]];
1101 AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
1102 AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
1103 AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
1104 AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
1106 uint8_t (*mvd_dst)[2] = &sl->mvd_table[list][FMO ? 8 * sl->mb_xy
1107 : h->mb2br_xy[sl->mb_xy]];
1108 uint8_t(*mvd_src)[2] = &sl->mvd_cache[list][scan8[0]];
1109 if (IS_SKIP(mb_type)) {
1110 AV_ZERO128(mvd_dst);
1112 AV_COPY64(mvd_dst, mvd_src + 8 * 3);
1113 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
1114 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
1115 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
1120 int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
1121 int8_t *ref_cache = sl->ref_cache[list];
1122 ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
1123 ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
1124 ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
1125 ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
1129 static av_always_inline void write_back_motion(const H264Context *h,
1130 H264SliceContext *sl,
1133 const int b_stride = h->b_stride;
1134 const int b_xy = 4 * sl->mb_x + 4 * sl->mb_y * h->b_stride; // try mb2b(8)_xy
1135 const int b8_xy = 4 * sl->mb_xy;
1137 if (USES_LIST(mb_type, 0)) {
1138 write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 0);
1140 fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
1141 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1143 if (USES_LIST(mb_type, 1))
1144 write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 1);
1146 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) {
1147 if (IS_8X8(mb_type)) {
1148 uint8_t *direct_table = &h->direct_table[4 * sl->mb_xy];
1149 direct_table[1] = sl->sub_mb_type[1] >> 1;
1150 direct_table[2] = sl->sub_mb_type[2] >> 1;
1151 direct_table[3] = sl->sub_mb_type[3] >> 1;
1156 static av_always_inline int get_dct8x8_allowed(const H264Context *h, H264SliceContext *sl)
1158 if (h->sps.direct_8x8_inference_flag)
1159 return !(AV_RN64A(sl->sub_mb_type) &
1160 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
1161 0x0001000100010001ULL));
1163 return !(AV_RN64A(sl->sub_mb_type) &
1164 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
1165 0x0001000100010001ULL));
1168 static inline int find_start_code(const uint8_t *buf, int buf_size,
1169 int buf_index, int next_avc)
1171 uint32_t state = -1;
1173 buf_index = avpriv_find_start_code(buf + buf_index, buf + next_avc + 1, &state) - buf - 1;
1175 return FFMIN(buf_index, buf_size);
1178 static inline int get_avc_nalsize(H264Context *h, const uint8_t *buf,
1179 int buf_size, int *buf_index)
1183 if (*buf_index >= buf_size - h->nal_length_size) {
1184 // the end of the buffer is reached, refill it.
1185 return AVERROR(EAGAIN);
1188 for (i = 0; i < h->nal_length_size; i++)
1189 nalsize = ((unsigned)nalsize << 8) | buf[(*buf_index)++];
1190 if (nalsize <= 0 || nalsize > buf_size - *buf_index) {
1191 av_log(h->avctx, AV_LOG_ERROR,
1192 "AVC: nal size %d\n", nalsize);
1193 return AVERROR_INVALIDDATA;
1198 int ff_h264_field_end(H264Context *h, H264SliceContext *sl, int in_setup);
1200 int ff_h264_ref_picture(H264Context *h, H264Picture *dst, H264Picture *src);
1201 void ff_h264_unref_picture(H264Context *h, H264Picture *pic);
1203 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl);
1205 void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl, int y, int height);
1206 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc);
1207 int ff_set_ref_count(H264Context *h, H264SliceContext *sl);
1209 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl);
1210 #define SLICE_SINGLETHREAD 1
1211 #define SLICE_SKIPED 2
1213 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count);
1214 int ff_h264_update_thread_context(AVCodecContext *dst,
1215 const AVCodecContext *src);
1217 void ff_h264_flush_change(H264Context *h);
1219 void ff_h264_free_tables(H264Context *h);
1221 void ff_h264_set_erpic(ERPicture *dst, H264Picture *src);
1223 #endif /* AVCODEC_H264_H */