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 "h264chroma.h"
41 #include "mpegutils.h"
44 #include "rectangle.h"
47 #define H264_MAX_PICTURE_COUNT 36
48 #define H264_MAX_THREADS 32
50 #define MAX_SPS_COUNT 32
51 #define MAX_PPS_COUNT 256
53 #define MAX_MMCO_COUNT 66
55 #define MAX_DELAYED_PIC_COUNT 16
57 #define MAX_MBPAIR_SIZE (256*1024) // a tighter bound could be calculated if someone cares about a few bytes
59 /* Compiling in interlaced support reduces the speed
60 * of progressive decoding by about 2%. */
61 #define ALLOW_INTERLACE
66 * The maximum number of slices supported by the decoder.
67 * must be a power of 2
71 #ifdef ALLOW_INTERLACE
72 #define MB_MBAFF(h) (h)->mb_mbaff
73 #define MB_FIELD(sl) (sl)->mb_field_decoding_flag
74 #define FRAME_MBAFF(h) (h)->mb_aff_frame
75 #define FIELD_PICTURE(h) ((h)->picture_structure != PICT_FRAME)
82 #define MB_FIELD(sl) 0
83 #define FRAME_MBAFF(h) 0
84 #define FIELD_PICTURE(h) 0
86 #define IS_INTERLACED(mb_type) 0
92 #define FIELD_OR_MBAFF_PICTURE(h) (FRAME_MBAFF(h) || FIELD_PICTURE(h))
95 #define CABAC(h) (h)->pps.cabac
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)
102 #define EXTENDED_SAR 255
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)
109 #define QP_MAX_NUM (51 + 6*6) // The maximum supported qp
122 NAL_END_SEQUENCE = 10,
124 NAL_FILLER_DATA = 12,
126 NAL_AUXILIARY_SLICE = 19,
127 NAL_FF_IGNORE = 0xff0f001,
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_REGISTERED = 4, ///< registered user data as specified by Rec. ITU-T 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_GREEN_METADATA = 56 ///< GreenMPEG information
145 * pic_struct in picture timing SEI message
148 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
149 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
150 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
151 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
152 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
153 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
154 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
155 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
156 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
160 * frame_packing_arrangement types
163 SEI_FPA_TYPE_CHECKERBOARD = 0,
164 SEI_FPA_TYPE_INTERLEAVE_COLUMN = 1,
165 SEI_FPA_TYPE_INTERLEAVE_ROW = 2,
166 SEI_FPA_TYPE_SIDE_BY_SIDE = 3,
167 SEI_FPA_TYPE_TOP_BOTTOM = 4,
168 SEI_FPA_TYPE_INTERLEAVE_TEMPORAL = 5,
173 * Sequence parameter set
179 int chroma_format_idc;
180 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
181 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
182 int poc_type; ///< pic_order_cnt_type
183 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
184 int delta_pic_order_always_zero_flag;
185 int offset_for_non_ref_pic;
186 int offset_for_top_to_bottom_field;
187 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
188 int ref_frame_count; ///< num_ref_frames
189 int gaps_in_frame_num_allowed_flag;
190 int mb_width; ///< pic_width_in_mbs_minus1 + 1
191 int mb_height; ///< pic_height_in_map_units_minus1 + 1
192 int frame_mbs_only_flag;
193 int mb_aff; ///< mb_adaptive_frame_field_flag
194 int direct_8x8_inference_flag;
195 int crop; ///< frame_cropping_flag
197 /* those 4 are already in luma samples */
198 unsigned int crop_left; ///< frame_cropping_rect_left_offset
199 unsigned int crop_right; ///< frame_cropping_rect_right_offset
200 unsigned int crop_top; ///< frame_cropping_rect_top_offset
201 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
202 int vui_parameters_present_flag;
204 int video_signal_type_present_flag;
206 int colour_description_present_flag;
207 enum AVColorPrimaries color_primaries;
208 enum AVColorTransferCharacteristic color_trc;
209 enum AVColorSpace colorspace;
210 int timing_info_present_flag;
211 uint32_t num_units_in_tick;
213 int fixed_frame_rate_flag;
214 short offset_for_ref_frame[256]; // FIXME dyn aloc?
215 int bitstream_restriction_flag;
216 int num_reorder_frames;
217 int scaling_matrix_present;
218 uint8_t scaling_matrix4[6][16];
219 uint8_t scaling_matrix8[6][64];
220 int nal_hrd_parameters_present_flag;
221 int vcl_hrd_parameters_present_flag;
222 int pic_struct_present_flag;
223 int time_offset_length;
224 int cpb_cnt; ///< See H.264 E.1.2
225 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
226 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
227 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
228 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
229 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
230 int residual_color_transform_flag; ///< residual_colour_transform_flag
231 int constraint_set_flags; ///< constraint_set[0-3]_flag
232 int new; ///< flag to keep track if the decoder context needs re-init due to changed SPS
238 * Picture parameter set
242 int cabac; ///< entropy_coding_mode_flag
243 int pic_order_present; ///< pic_order_present_flag
244 int slice_group_count; ///< num_slice_groups_minus1 + 1
245 int mb_slice_group_map_type;
246 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
247 int weighted_pred; ///< weighted_pred_flag
248 int weighted_bipred_idc;
249 int init_qp; ///< pic_init_qp_minus26 + 26
250 int init_qs; ///< pic_init_qs_minus26 + 26
251 int chroma_qp_index_offset[2];
252 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
253 int constrained_intra_pred; ///< constrained_intra_pred_flag
254 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
255 int transform_8x8_mode; ///< transform_8x8_mode_flag
256 uint8_t scaling_matrix4[6][16];
257 uint8_t scaling_matrix8[6][64];
258 uint8_t chroma_qp_table[2][QP_MAX_NUM+1]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
265 * Frame Packing Arrangement Type
268 int frame_packing_arrangement_id;
269 int frame_packing_arrangement_cancel_flag; ///< is previous arrangement canceled, -1 if never received
270 SEI_FpaType frame_packing_arrangement_type;
271 int frame_packing_arrangement_repetition_period;
272 int content_interpretation_type;
273 int quincunx_sampling_flag;
277 * Green MetaData Information Type
279 typedef struct GreenMetaData {
280 uint8_t green_metadata_type;
282 uint16_t num_seconds;
283 uint16_t num_pictures;
284 uint8_t percent_non_zero_macroblocks;
285 uint8_t percent_intra_coded_macroblocks;
286 uint8_t percent_six_tap_filtering;
287 uint8_t percent_alpha_point_deblocking_instance;
288 uint8_t xsd_metric_type;
289 uint16_t xsd_metric_value;
293 * Memory management control operation opcode.
295 typedef enum MMCOOpcode {
306 * Memory management control operation.
308 typedef struct MMCO {
310 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
311 int long_arg; ///< index, pic_num, or num long refs depending on opcode
314 typedef struct H264Picture {
318 AVBufferRef *qscale_table_buf;
319 int8_t *qscale_table;
321 AVBufferRef *motion_val_buf[2];
322 int16_t (*motion_val[2])[2];
324 AVBufferRef *mb_type_buf;
327 AVBufferRef *hwaccel_priv_buf;
328 void *hwaccel_picture_private; ///< hardware accelerator private data
330 AVBufferRef *ref_index_buf[2];
331 int8_t *ref_index[2];
333 int field_poc[2]; ///< top/bottom POC
334 int poc; ///< frame POC
335 int frame_num; ///< frame_num (raw frame_num from slice header)
336 int mmco_reset; /**< MMCO_RESET set this 1. Reordering code must
337 not mix pictures before and after MMCO_RESET. */
338 int pic_id; /**< pic_num (short -> no wrap version of pic_num,
339 pic_num & max_pic_num; long -> long_pic_num) */
340 int long_ref; ///< 1->long term reference 0->short term reference
341 int ref_poc[2][2][32]; ///< POCs of the frames/fields used as reference (FIXME need per slice)
342 int ref_count[2][2]; ///< number of entries in ref_poc (FIXME need per slice)
343 int mbaff; ///< 1 -> MBAFF frame 0-> not MBAFF
344 int field_picture; ///< whether or not picture was encoded in separate fields
347 int recovered; ///< picture at IDR or recovery point + recovery count
349 int sei_recovery_frame_cnt;
356 typedef struct H264Ref {
367 typedef struct H264SliceContext {
368 struct H264Context *h264;
374 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
375 int slice_type_fixed;
378 int chroma_qp[2]; // QPc
379 int qp_thresh; ///< QP threshold to skip loopfilter
380 int last_qscale_diff;
383 int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
384 int slice_alpha_c0_offset;
385 int slice_beta_offset;
387 // Weighted pred stuff
389 int use_weight_chroma;
390 int luma_log2_weight_denom;
391 int chroma_log2_weight_denom;
392 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
393 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
394 // The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
395 int luma_weight[48][2][2];
396 int chroma_weight[48][2][2][2];
397 int implicit_weight[48][48][2];
402 int chroma_pred_mode;
403 int intra16x16_pred_mode;
405 int8_t intra4x4_pred_mode_cache[5 * 8];
406 int8_t(*intra4x4_pred_mode);
411 int left_mb_xy[LEFT_MBS];
416 int left_type[LEFT_MBS];
418 const uint8_t *left_block;
419 int topleft_partition;
421 unsigned int topleft_samples_available;
422 unsigned int top_samples_available;
423 unsigned int topright_samples_available;
424 unsigned int left_samples_available;
426 ptrdiff_t linesize, uvlinesize;
427 ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
428 ptrdiff_t mb_uvlinesize;
434 // index of the first MB of the next slice
439 int mb_field_decoding_flag;
440 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
442 int redundant_pic_count;
445 * number of neighbors (top and/or left) that used 8x8 dct
447 int neighbor_transform_size;
449 int direct_spatial_mv_pred;
457 int dist_scale_factor[32];
458 int dist_scale_factor_field[2][32];
459 int map_col_to_list0[2][16 + 32];
460 int map_col_to_list0_field[2][2][16 + 32];
463 * num_ref_idx_l0/1_active_minus1 + 1
465 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
466 unsigned int list_count;
467 H264Ref ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
468 * Reordered version of default_ref_list
469 * according to picture reordering in slice header */
470 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
472 const uint8_t *intra_pcm_ptr;
473 int16_t *dc_val_base;
475 uint8_t *bipred_scratchpad;
476 uint8_t *edge_emu_buffer;
477 uint8_t (*top_borders[2])[(16 * 3) * 2];
478 int bipred_scratchpad_allocated;
479 int edge_emu_buffer_allocated;
480 int top_borders_allocated[2];
483 * non zero coeff count cache.
484 * is 64 if not available.
486 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
489 * Motion vector cache.
491 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
492 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
493 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
494 uint8_t direct_cache[5 * 8];
496 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
498 ///< as a dct coefficient is int32_t in high depth, we need to reserve twice the space.
499 DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2];
500 DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
501 ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either
502 ///< check that i is not too large or ensure that there is some unused stuff after mb
503 int16_t mb_padding[256 * 2];
505 uint8_t (*mvd_table[2])[2];
511 uint8_t cabac_state[1024];
514 // rbsp buffer used for this slice
515 uint8_t *rbsp_buffer;
516 unsigned int rbsp_buffer_size;
522 typedef struct H264Context {
524 AVCodecContext *avctx;
525 VideoDSPContext vdsp;
526 H264DSPContext h264dsp;
527 H264ChromaContext h264chroma;
528 H264QpelContext h264qpel;
531 H264Picture DPB[H264_MAX_PICTURE_COUNT];
532 H264Picture *cur_pic_ptr;
534 H264Picture last_pic_for_ec;
536 H264SliceContext *slice_ctx;
539 int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
541 /* coded dimensions -- 16 * mb w/h */
543 int chroma_x_shift, chroma_y_shift;
546 * Backup frame properties: needed, because they can be different
547 * between returned frame and last decoded frame.
551 enum AVPixelFormat backup_pix_fmt;
554 int coded_picture_number;
557 int context_initialized;
561 int8_t(*intra4x4_pred_mode);
564 uint8_t (*non_zero_count)[48];
566 #define LIST_NOT_USED -1 // FIXME rename?
567 #define PART_NOT_AVAILABLE -2
570 * block_offset[ 0..23] for frame macroblocks
571 * block_offset[24..47] for field macroblocks
573 int block_offset[2 * (16 * 3)];
575 uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
577 int b_stride; // FIXME use s->b4_stride
580 unsigned current_sps_id; ///< id of the current SPS
581 SPS sps; ///< current sps
582 PPS pps; ///< current pps
584 int au_pps_id; ///< pps_id of current access unit
586 uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
587 uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
588 uint32_t(*dequant4_coeff[6])[16];
589 uint32_t(*dequant8_coeff[6])[64];
591 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
593 // interlacing specific flags
595 int picture_structure;
598 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
600 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
603 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
604 uint8_t *chroma_pred_mode_table;
605 uint8_t (*mvd_table[2])[2];
606 uint8_t *direct_table;
608 uint8_t zigzag_scan[16];
609 uint8_t zigzag_scan8x8[64];
610 uint8_t zigzag_scan8x8_cavlc[64];
611 uint8_t field_scan[16];
612 uint8_t field_scan8x8[64];
613 uint8_t field_scan8x8_cavlc[64];
614 uint8_t zigzag_scan_q0[16];
615 uint8_t zigzag_scan8x8_q0[64];
616 uint8_t zigzag_scan8x8_cavlc_q0[64];
617 uint8_t field_scan_q0[16];
618 uint8_t field_scan8x8_q0[64];
619 uint8_t field_scan8x8_cavlc_q0[64];
624 int mb_height, mb_width;
628 // =============================================================
629 // Things below are not used in the MB or more inner code
635 * Used to parse AVC variant of h264
637 int is_avc; ///< this flag is != 0 if codec is avc1
638 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
640 int bit_depth_luma; ///< luma bit depth from sps to detect changes
641 int chroma_format_idc; ///< chroma format from sps to detect changes
643 SPS *sps_buffers[MAX_SPS_COUNT];
644 PPS *pps_buffers[MAX_PPS_COUNT];
646 int dequant_coeff_pps; ///< reinit tables when pps changes
648 uint16_t *slice_table_base;
653 int delta_poc_bottom;
656 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
657 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
658 int frame_num_offset; ///< for POC type 2
659 int prev_frame_num_offset; ///< for POC type 2
660 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
663 * frame_num for frames or 2 * frame_num + 1 for field pics.
668 * max_frame_num or 2 * max_frame_num for field pics.
672 H264Picture *short_ref[32];
673 H264Picture *long_ref[32];
674 H264Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
675 int last_pocs[MAX_DELAYED_PIC_COUNT];
676 H264Picture *next_output_pic;
677 int next_outputed_poc;
680 * memory management control operations buffer.
682 MMCO mmco[MAX_MMCO_COUNT];
686 int long_ref_count; ///< number of actual long term references
687 int short_ref_count; ///< number of actual short term references
690 * @name Members for slice based multithreading
694 * current slice number, used to initialize slice_num of each thread/context
699 * Max number of threads / contexts.
700 * This is equal to AVCodecContext.thread_count unless
701 * multithreaded decoding is impossible, in which case it is
706 int slice_context_count;
709 * 1 if the single thread fallback warning has already been
710 * displayed, 0 otherwise.
712 int single_decode_warning;
714 enum AVPictureType pict_type;
719 * pic_struct in picture timing SEI message
721 SEI_PicStructType sei_pic_struct;
724 * Complement sei_pic_struct
725 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
726 * However, soft telecined frames may have these values.
727 * This is used in an attempt to flag soft telecine progressive.
729 int prev_interlaced_frame;
732 * frame_packing_arrangment SEI message
734 int sei_frame_packing_present;
735 int frame_packing_arrangement_type;
736 int content_interpretation_type;
737 int quincunx_subsampling;
740 * display orientation SEI message
742 int sei_display_orientation_present;
743 int sei_anticlockwise_rotation;
744 int sei_hflip, sei_vflip;
747 * User data registered by Rec. ITU-T T.35 SEI
749 int sei_reguserdata_afd_present;
750 uint8_t active_format_description;
751 int a53_caption_size;
752 uint8_t *a53_caption;
755 * Bit set of clock types for fields/frames in picture timing SEI message.
756 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
762 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
764 int sei_dpb_output_delay;
767 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
769 int sei_cpb_removal_delay;
772 * recovery_frame_cnt from SEI message
774 * Set to -1 if no recovery point SEI message found or to number of frames
775 * before playback synchronizes. Frames having recovery point are key
778 int sei_recovery_frame_cnt;
781 * Are the SEI recovery points looking valid.
783 int valid_recovery_point;
788 * recovery_frame is the frame_num at which the next frame should
789 * be fully constructed.
791 * Set to -1 when not expecting a recovery point.
796 * We have seen an IDR, so all the following frames in coded order are correctly
799 #define FRAME_RECOVERED_IDR (1 << 0)
801 * Sufficient number of frames have been decoded since a SEI recovery point,
802 * so all the following frames in presentation order are correct.
804 #define FRAME_RECOVERED_SEI (1 << 1)
806 int frame_recovered; ///< Initial frame has been completely recovered
808 int has_recovery_point;
812 /* for frame threading, this is set to 1
813 * after finish_setup() has been called, so we cannot modify
814 * some context properties (which are supposed to stay constant between
819 int sei_buffering_period_present; ///< Buffering period SEI flag
820 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
822 int cur_chroma_format_idc;
823 int cur_bit_depth_luma;
824 int16_t slice_row[MAX_SLICES]; ///< to detect when MAX_SLICES is too low
826 uint8_t parse_history[6];
827 int parse_history_count;
832 AVBufferPool *qscale_table_pool;
833 AVBufferPool *mb_type_pool;
834 AVBufferPool *motion_val_pool;
835 AVBufferPool *ref_index_pool;
837 /* Motion Estimation */
838 qpel_mc_func (*qpel_put)[16];
839 qpel_mc_func (*qpel_avg)[16];
842 GreenMetaData sei_green_metadata;
846 extern const uint8_t ff_h264_chroma_qp[7][QP_MAX_NUM + 1]; ///< One chroma qp table for each possible bit depth (8-14).
847 extern const uint16_t ff_h264_mb_sizes[4];
852 int ff_h264_decode_sei(H264Context *h);
857 int ff_h264_decode_seq_parameter_set(H264Context *h, int ignore_truncation);
860 * compute profile from sps
862 int ff_h264_get_profile(SPS *sps);
867 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
870 * Decode a network abstraction layer unit.
871 * @param consumed is the number of bytes used as input
872 * @param length is the length of the array
873 * @param dst_length is the number of decoded bytes FIXME here
874 * or a decode rbsp tailing?
875 * @return decoded bytes, might be src+1 if no escapes
877 const uint8_t *ff_h264_decode_nal(H264Context *h, H264SliceContext *sl, const uint8_t *src,
878 int *dst_length, int *consumed, int length);
881 * Free any data that may have been allocated in the H264 context
884 void ff_h264_free_context(H264Context *h);
887 * Reconstruct bitstream slice_type.
889 int ff_h264_get_slice_type(const H264SliceContext *sl);
895 int ff_h264_alloc_tables(H264Context *h);
897 int ff_h264_decode_ref_pic_list_reordering(H264Context *h, H264SliceContext *sl);
898 void ff_h264_fill_mbaff_ref_list(H264Context *h, H264SliceContext *sl);
899 void ff_h264_remove_all_refs(H264Context *h);
902 * Execute the reference picture marking (memory management control operations).
904 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
906 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb,
909 int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
912 * Check if the top & left blocks are available if needed & change the
913 * dc mode so it only uses the available blocks.
915 int ff_h264_check_intra4x4_pred_mode(const H264Context *h, H264SliceContext *sl);
918 * Check if the top & left blocks are available if needed & change the
919 * dc mode so it only uses the available blocks.
921 int ff_h264_check_intra_pred_mode(const H264Context *h, H264SliceContext *sl,
922 int mode, int is_chroma);
924 void ff_h264_hl_decode_mb(const H264Context *h, H264SliceContext *sl);
925 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size);
926 int ff_h264_decode_init(AVCodecContext *avctx);
927 void ff_h264_decode_init_vlc(void);
930 * Decode a macroblock
931 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
933 int ff_h264_decode_mb_cavlc(const H264Context *h, H264SliceContext *sl);
936 * Decode a CABAC coded macroblock
937 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
939 int ff_h264_decode_mb_cabac(const H264Context *h, H264SliceContext *sl);
941 void ff_h264_init_cabac_states(const H264Context *h, H264SliceContext *sl);
943 void ff_h264_init_dequant_tables(H264Context *h);
945 void ff_h264_direct_dist_scale_factor(const H264Context *const h, H264SliceContext *sl);
946 void ff_h264_direct_ref_list_init(const H264Context *const h, H264SliceContext *sl);
947 void ff_h264_pred_direct_motion(const H264Context *const h, H264SliceContext *sl,
950 void ff_h264_filter_mb_fast(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
951 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
952 unsigned int linesize, unsigned int uvlinesize);
953 void ff_h264_filter_mb(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
954 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
955 unsigned int linesize, unsigned int uvlinesize);
958 * Reset SEI values at the beginning of the frame.
960 * @param h H.264 context.
962 void ff_h264_reset_sei(H264Context *h);
965 * Get stereo_mode string from the h264 frame_packing_arrangement
966 * @param h H.264 context.
968 const char* ff_h264_sei_stereo_mode(H264Context *h);
980 /* Scan8 organization:
997 * DY/DU/DV are for luma/chroma DC.
1000 #define LUMA_DC_BLOCK_INDEX 48
1001 #define CHROMA_DC_BLOCK_INDEX 49
1003 // This table must be here because scan8[constant] must be known at compiletime
1004 static const uint8_t scan8[16 * 3 + 3] = {
1005 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
1006 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
1007 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
1008 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
1009 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
1010 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
1011 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
1012 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
1013 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
1014 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
1015 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
1016 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
1017 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
1020 static av_always_inline uint32_t pack16to32(unsigned a, unsigned b)
1023 return (b & 0xFFFF) + (a << 16);
1025 return (a & 0xFFFF) + (b << 16);
1029 static av_always_inline uint16_t pack8to16(unsigned a, unsigned b)
1032 return (b & 0xFF) + (a << 8);
1034 return (a & 0xFF) + (b << 8);
1039 * Get the chroma qp.
1041 static av_always_inline int get_chroma_qp(const H264Context *h, int t, int qscale)
1043 return h->pps.chroma_qp_table[t][qscale];
1047 * Get the predicted intra4x4 prediction mode.
1049 static av_always_inline int pred_intra_mode(const H264Context *h,
1050 H264SliceContext *sl, int n)
1052 const int index8 = scan8[n];
1053 const int left = sl->intra4x4_pred_mode_cache[index8 - 1];
1054 const int top = sl->intra4x4_pred_mode_cache[index8 - 8];
1055 const int min = FFMIN(left, top);
1057 ff_tlog(h->avctx, "mode:%d %d min:%d\n", left, top, min);
1065 static av_always_inline void write_back_intra_pred_mode(const H264Context *h,
1066 H264SliceContext *sl)
1068 int8_t *i4x4 = sl->intra4x4_pred_mode + h->mb2br_xy[sl->mb_xy];
1069 int8_t *i4x4_cache = sl->intra4x4_pred_mode_cache;
1071 AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
1072 i4x4[4] = i4x4_cache[7 + 8 * 3];
1073 i4x4[5] = i4x4_cache[7 + 8 * 2];
1074 i4x4[6] = i4x4_cache[7 + 8 * 1];
1077 static av_always_inline void write_back_non_zero_count(const H264Context *h,
1078 H264SliceContext *sl)
1080 const int mb_xy = sl->mb_xy;
1081 uint8_t *nnz = h->non_zero_count[mb_xy];
1082 uint8_t *nnz_cache = sl->non_zero_count_cache;
1084 AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
1085 AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
1086 AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
1087 AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
1088 AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
1089 AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
1090 AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
1091 AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
1093 if (!h->chroma_y_shift) {
1094 AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
1095 AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
1096 AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
1097 AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
1101 static av_always_inline void write_back_motion_list(const H264Context *h,
1102 H264SliceContext *sl,
1104 int b_xy, int b8_xy,
1105 int mb_type, int list)
1107 int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
1108 int16_t(*mv_src)[2] = &sl->mv_cache[list][scan8[0]];
1109 AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
1110 AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
1111 AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
1112 AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
1114 uint8_t (*mvd_dst)[2] = &sl->mvd_table[list][FMO ? 8 * sl->mb_xy
1115 : h->mb2br_xy[sl->mb_xy]];
1116 uint8_t(*mvd_src)[2] = &sl->mvd_cache[list][scan8[0]];
1117 if (IS_SKIP(mb_type)) {
1118 AV_ZERO128(mvd_dst);
1120 AV_COPY64(mvd_dst, mvd_src + 8 * 3);
1121 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
1122 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
1123 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
1128 int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
1129 int8_t *ref_cache = sl->ref_cache[list];
1130 ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
1131 ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
1132 ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
1133 ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
1137 static av_always_inline void write_back_motion(const H264Context *h,
1138 H264SliceContext *sl,
1141 const int b_stride = h->b_stride;
1142 const int b_xy = 4 * sl->mb_x + 4 * sl->mb_y * h->b_stride; // try mb2b(8)_xy
1143 const int b8_xy = 4 * sl->mb_xy;
1145 if (USES_LIST(mb_type, 0)) {
1146 write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 0);
1148 fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
1149 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1151 if (USES_LIST(mb_type, 1))
1152 write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 1);
1154 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) {
1155 if (IS_8X8(mb_type)) {
1156 uint8_t *direct_table = &h->direct_table[4 * sl->mb_xy];
1157 direct_table[1] = sl->sub_mb_type[1] >> 1;
1158 direct_table[2] = sl->sub_mb_type[2] >> 1;
1159 direct_table[3] = sl->sub_mb_type[3] >> 1;
1164 static av_always_inline int get_dct8x8_allowed(const H264Context *h, H264SliceContext *sl)
1166 if (h->sps.direct_8x8_inference_flag)
1167 return !(AV_RN64A(sl->sub_mb_type) &
1168 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
1169 0x0001000100010001ULL));
1171 return !(AV_RN64A(sl->sub_mb_type) &
1172 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
1173 0x0001000100010001ULL));
1176 static inline int find_start_code(const uint8_t *buf, int buf_size,
1177 int buf_index, int next_avc)
1179 uint32_t state = -1;
1181 buf_index = avpriv_find_start_code(buf + buf_index, buf + next_avc + 1, &state) - buf - 1;
1183 return FFMIN(buf_index, buf_size);
1186 static inline int get_avc_nalsize(H264Context *h, const uint8_t *buf,
1187 int buf_size, int *buf_index)
1191 if (*buf_index >= buf_size - h->nal_length_size) {
1192 // the end of the buffer is reached, refill it.
1193 return AVERROR(EAGAIN);
1196 for (i = 0; i < h->nal_length_size; i++)
1197 nalsize = ((unsigned)nalsize << 8) | buf[(*buf_index)++];
1198 if (nalsize <= 0 || nalsize > buf_size - *buf_index) {
1199 av_log(h->avctx, AV_LOG_ERROR,
1200 "AVC: nal size %d\n", nalsize);
1201 return AVERROR_INVALIDDATA;
1206 int ff_h264_field_end(H264Context *h, H264SliceContext *sl, int in_setup);
1208 int ff_h264_ref_picture(H264Context *h, H264Picture *dst, H264Picture *src);
1209 void ff_h264_unref_picture(H264Context *h, H264Picture *pic);
1211 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl);
1213 void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl, int y, int height);
1214 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc);
1215 int ff_pred_weight_table(H264Context *h, H264SliceContext *sl);
1216 int ff_set_ref_count(H264Context *h, H264SliceContext *sl);
1218 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl);
1219 #define SLICE_SINGLETHREAD 1
1220 #define SLICE_SKIPED 2
1222 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count);
1223 int ff_h264_update_thread_context(AVCodecContext *dst,
1224 const AVCodecContext *src);
1226 void ff_h264_flush_change(H264Context *h);
1228 void ff_h264_free_tables(H264Context *h);
1230 void ff_h264_set_erpic(ERPicture *dst, H264Picture *src);
1232 #endif /* AVCODEC_H264_H */