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 Libav.
7 * Libav 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 * Libav 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 Libav; 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"
33 #include "error_resilience.h"
35 #include "h264chroma.h"
40 #include "mpegutils.h"
43 #include "rectangle.h"
46 #define H264_MAX_PICTURE_COUNT 32
47 #define H264_MAX_THREADS 16
49 #define MAX_SPS_COUNT 32
50 #define MAX_PPS_COUNT 256
52 #define MAX_MMCO_COUNT 66
54 #define MAX_DELAYED_PIC_COUNT 16
56 /* Compiling in interlaced support reduces the speed
57 * of progressive decoding by about 2%. */
58 #define ALLOW_INTERLACE
63 * The maximum number of slices supported by the decoder.
64 * must be a power of 2
68 #ifdef ALLOW_INTERLACE
69 #define MB_MBAFF(h) h->mb_mbaff
70 #define MB_FIELD(h) h->mb_field_decoding_flag
71 #define FRAME_MBAFF(h) h->mb_aff_frame
72 #define FIELD_PICTURE(h) (h->picture_structure != PICT_FRAME)
80 #define FRAME_MBAFF(h) 0
81 #define FIELD_PICTURE(h) 0
83 #define IS_INTERLACED(mb_type) 0
89 #define FIELD_OR_MBAFF_PICTURE(h) (FRAME_MBAFF(h) || FIELD_PICTURE(h))
92 #define CABAC(h) h->pps.cabac
95 #define CHROMA422(h) (h->sps.chroma_format_idc == 2)
96 #define CHROMA444(h) (h->sps.chroma_format_idc == 3)
98 #define EXTENDED_SAR 255
100 #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit
101 #define MB_TYPE_8x8DCT 0x01000000
102 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
103 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
105 #define QP_MAX_NUM (51 + 2 * 6) // The maximum supported qp
118 NAL_END_SEQUENCE = 10,
120 NAL_FILLER_DATA = 12,
122 NAL_AUXILIARY_SLICE = 19,
123 NAL_FF_IGNORE = 0xff0f001,
130 SEI_TYPE_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
131 SEI_TYPE_PIC_TIMING = 1, ///< picture timing
132 SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
133 SEI_TYPE_RECOVERY_POINT = 6, ///< recovery point (frame # to decoder sync)
134 SEI_TYPE_FRAME_PACKING = 45, ///< frame packing arrangement
135 SEI_TYPE_DISPLAY_ORIENTATION = 47, ///< display orientation
139 * pic_struct in picture timing SEI message
142 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
143 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
144 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
145 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
146 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
147 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
148 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
149 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
150 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
154 * Sequence parameter set
160 int chroma_format_idc;
161 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
162 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
163 int poc_type; ///< pic_order_cnt_type
164 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
165 int delta_pic_order_always_zero_flag;
166 int offset_for_non_ref_pic;
167 int offset_for_top_to_bottom_field;
168 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
169 int ref_frame_count; ///< num_ref_frames
170 int gaps_in_frame_num_allowed_flag;
171 int mb_width; ///< pic_width_in_mbs_minus1 + 1
172 int mb_height; ///< pic_height_in_map_units_minus1 + 1
173 int frame_mbs_only_flag;
174 int mb_aff; ///< mb_adaptive_frame_field_flag
175 int direct_8x8_inference_flag;
176 int crop; ///< frame_cropping_flag
178 /* those 4 are already in luma samples */
179 unsigned int crop_left; ///< frame_cropping_rect_left_offset
180 unsigned int crop_right; ///< frame_cropping_rect_right_offset
181 unsigned int crop_top; ///< frame_cropping_rect_top_offset
182 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
183 int vui_parameters_present_flag;
185 int video_signal_type_present_flag;
187 int colour_description_present_flag;
188 enum AVColorPrimaries color_primaries;
189 enum AVColorTransferCharacteristic color_trc;
190 enum AVColorSpace colorspace;
191 int timing_info_present_flag;
192 uint32_t num_units_in_tick;
194 int fixed_frame_rate_flag;
195 short offset_for_ref_frame[256]; // FIXME dyn aloc?
196 int bitstream_restriction_flag;
197 int num_reorder_frames;
198 int scaling_matrix_present;
199 uint8_t scaling_matrix4[6][16];
200 uint8_t scaling_matrix8[6][64];
201 int nal_hrd_parameters_present_flag;
202 int vcl_hrd_parameters_present_flag;
203 int pic_struct_present_flag;
204 int time_offset_length;
205 int cpb_cnt; ///< See H.264 E.1.2
206 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
207 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
208 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
209 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
210 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
211 int residual_color_transform_flag; ///< residual_colour_transform_flag
212 int constraint_set_flags; ///< constraint_set[0-3]_flag
213 int new; ///< flag to keep track if the decoder context needs re-init due to changed SPS
217 * Picture parameter set
221 int cabac; ///< entropy_coding_mode_flag
222 int pic_order_present; ///< pic_order_present_flag
223 int slice_group_count; ///< num_slice_groups_minus1 + 1
224 int mb_slice_group_map_type;
225 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
226 int weighted_pred; ///< weighted_pred_flag
227 int weighted_bipred_idc;
228 int init_qp; ///< pic_init_qp_minus26 + 26
229 int init_qs; ///< pic_init_qs_minus26 + 26
230 int chroma_qp_index_offset[2];
231 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
232 int constrained_intra_pred; ///< constrained_intra_pred_flag
233 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
234 int transform_8x8_mode; ///< transform_8x8_mode_flag
235 uint8_t scaling_matrix4[6][16];
236 uint8_t scaling_matrix8[6][64];
237 uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
242 * Memory management control operation opcode.
244 typedef enum MMCOOpcode {
255 * Memory management control operation.
257 typedef struct MMCO {
259 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
260 int long_arg; ///< index, pic_num, or num long refs depending on opcode
263 typedef struct H264Picture {
267 AVBufferRef *qscale_table_buf;
268 int8_t *qscale_table;
270 AVBufferRef *motion_val_buf[2];
271 int16_t (*motion_val[2])[2];
273 AVBufferRef *mb_type_buf;
276 AVBufferRef *hwaccel_priv_buf;
277 void *hwaccel_picture_private; ///< hardware accelerator private data
279 AVBufferRef *ref_index_buf[2];
280 int8_t *ref_index[2];
282 int field_poc[2]; ///< top/bottom POC
283 int poc; ///< frame POC
284 int frame_num; ///< frame_num (raw frame_num from slice header)
285 int mmco_reset; /**< MMCO_RESET set this 1. Reordering code must
286 not mix pictures before and after MMCO_RESET. */
287 int pic_id; /**< pic_num (short -> no wrap version of pic_num,
288 pic_num & max_pic_num; long -> long_pic_num) */
289 int long_ref; ///< 1->long term reference 0->short term reference
290 int ref_poc[2][2][32]; ///< POCs of the frames used as reference (FIXME need per slice)
291 int ref_count[2][2]; ///< number of entries in ref_poc (FIXME need per slice)
292 int mbaff; ///< 1 -> MBAFF frame 0-> not MBAFF
293 int field_picture; ///< whether or not picture was encoded in separate fields
295 int needs_realloc; ///< picture needs to be reallocated (eg due to a frame size change)
297 int recovered; ///< picture at IDR or recovery point + recovery count
303 typedef struct H264Context {
304 AVCodecContext *avctx;
306 VideoDSPContext vdsp;
307 H264DSPContext h264dsp;
308 H264ChromaContext h264chroma;
309 H264QpelContext h264qpel;
310 ParseContext parse_context;
315 H264Picture *cur_pic_ptr;
318 int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
319 int chroma_qp[2]; // QPc
321 int qp_thresh; ///< QP threshold to skip loopfilter
323 /* coded dimensions -- 16 * mb w/h */
325 ptrdiff_t linesize, uvlinesize;
326 int chroma_x_shift, chroma_y_shift;
330 int data_partitioning;
331 int coded_picture_number;
334 int context_initialized;
342 int chroma_pred_mode;
343 int intra16x16_pred_mode;
348 int left_mb_xy[LEFT_MBS];
353 int left_type[LEFT_MBS];
355 const uint8_t *left_block;
356 int topleft_partition;
358 int8_t intra4x4_pred_mode_cache[5 * 8];
359 int8_t(*intra4x4_pred_mode);
361 unsigned int topleft_samples_available;
362 unsigned int top_samples_available;
363 unsigned int topright_samples_available;
364 unsigned int left_samples_available;
365 uint8_t (*top_borders[2])[(16 * 3) * 2];
368 * non zero coeff count cache.
369 * is 64 if not available.
371 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
373 uint8_t (*non_zero_count)[48];
376 * Motion vector cache.
378 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
379 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
380 #define LIST_NOT_USED -1 // FIXME rename?
381 #define PART_NOT_AVAILABLE -2
384 * number of neighbors (top and/or left) that used 8x8 dct
386 int neighbor_transform_size;
389 * block_offset[ 0..23] for frame macroblocks
390 * block_offset[24..47] for field macroblocks
392 int block_offset[2 * (16 * 3)];
394 uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
396 int b_stride; // FIXME use s->b4_stride
398 ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
399 ptrdiff_t mb_uvlinesize;
401 SPS sps; ///< current sps
402 PPS pps; ///< current pps
404 uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
405 uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
406 uint32_t(*dequant4_coeff[6])[16];
407 uint32_t(*dequant8_coeff[6])[64];
410 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
412 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
413 int slice_type_fixed;
415 // interlacing specific flags
417 int mb_field_decoding_flag;
418 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
419 int picture_structure;
422 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
424 // Weighted pred stuff
426 int use_weight_chroma;
427 int luma_log2_weight_denom;
428 int chroma_log2_weight_denom;
429 // The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
430 int luma_weight[48][2][2];
431 int chroma_weight[48][2][2][2];
432 int implicit_weight[48][48][2];
434 int direct_spatial_mv_pred;
437 int dist_scale_factor[32];
438 int dist_scale_factor_field[2][32];
439 int map_col_to_list0[2][16 + 32];
440 int map_col_to_list0_field[2][2][16 + 32];
443 * num_ref_idx_l0/1_active_minus1 + 1
445 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
446 unsigned int list_count;
447 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
448 H264Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
449 * Reordered version of default_ref_list
450 * according to picture reordering in slice header */
451 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
454 GetBitContext intra_gb;
455 GetBitContext inter_gb;
456 GetBitContext *intra_gb_ptr;
457 GetBitContext *inter_gb_ptr;
459 const uint8_t *intra_pcm_ptr;
460 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.
461 DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
462 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
468 uint8_t cabac_state[1024];
470 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
475 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
476 uint8_t *chroma_pred_mode_table;
477 int last_qscale_diff;
478 uint8_t (*mvd_table[2])[2];
479 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
480 uint8_t *direct_table;
481 uint8_t direct_cache[5 * 8];
483 uint8_t zigzag_scan[16];
484 uint8_t zigzag_scan8x8[64];
485 uint8_t zigzag_scan8x8_cavlc[64];
486 uint8_t field_scan[16];
487 uint8_t field_scan8x8[64];
488 uint8_t field_scan8x8_cavlc[64];
489 const uint8_t *zigzag_scan_q0;
490 const uint8_t *zigzag_scan8x8_q0;
491 const uint8_t *zigzag_scan8x8_cavlc_q0;
492 const uint8_t *field_scan_q0;
493 const uint8_t *field_scan8x8_q0;
494 const uint8_t *field_scan8x8_cavlc_q0;
502 int mb_height, mb_width;
510 int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
511 int slice_alpha_c0_offset;
512 int slice_beta_offset;
514 // =============================================================
515 // Things below are not used in the MB or more inner code
519 uint8_t *rbsp_buffer[2];
520 unsigned int rbsp_buffer_size[2];
523 * Used to parse AVC variant of h264
525 int is_avc; ///< this flag is != 0 if codec is avc1
526 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
527 int got_first; ///< this flag is != 0 if we've parsed a frame
529 int bit_depth_luma; ///< luma bit depth from sps to detect changes
530 int chroma_format_idc; ///< chroma format from sps to detect changes
532 SPS *sps_buffers[MAX_SPS_COUNT];
533 PPS *pps_buffers[MAX_PPS_COUNT];
535 int dequant_coeff_pps; ///< reinit tables when pps changes
537 uint16_t *slice_table_base;
542 int delta_poc_bottom;
545 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
546 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
547 int frame_num_offset; ///< for POC type 2
548 int prev_frame_num_offset; ///< for POC type 2
549 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
552 * frame_num for frames or 2 * frame_num + 1 for field pics.
557 * max_frame_num or 2 * max_frame_num for field pics.
561 int redundant_pic_count;
563 H264Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
564 H264Picture *short_ref[32];
565 H264Picture *long_ref[32];
566 H264Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
567 int last_pocs[MAX_DELAYED_PIC_COUNT];
568 H264Picture *next_output_pic;
570 int next_outputed_poc;
573 * memory management control operations buffer.
575 MMCO mmco[MAX_MMCO_COUNT];
579 int long_ref_count; ///< number of actual long term references
580 int short_ref_count; ///< number of actual short term references
585 * @name Members for slice based multithreading
588 struct H264Context *thread_context[H264_MAX_THREADS];
591 * current slice number, used to initalize slice_num of each thread/context
596 * Max number of threads / contexts.
597 * This is equal to AVCodecContext.thread_count unless
598 * multithreaded decoding is impossible, in which case it is
603 int slice_context_count;
606 * 1 if the single thread fallback warning has already been
607 * displayed, 0 otherwise.
609 int single_decode_warning;
611 enum AVPictureType pict_type;
617 * pic_struct in picture timing SEI message
619 SEI_PicStructType sei_pic_struct;
622 * Complement sei_pic_struct
623 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
624 * However, soft telecined frames may have these values.
625 * This is used in an attempt to flag soft telecine progressive.
627 int prev_interlaced_frame;
630 * frame_packing_arrangment SEI message
632 int sei_frame_packing_present;
633 int frame_packing_arrangement_type;
634 int content_interpretation_type;
635 int quincunx_subsampling;
638 * display orientation SEI message
640 int sei_display_orientation_present;
641 int sei_anticlockwise_rotation;
642 int sei_hflip, sei_vflip;
645 * Bit set of clock types for fields/frames in picture timing SEI message.
646 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
652 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
654 int sei_dpb_output_delay;
657 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
659 int sei_cpb_removal_delay;
662 * recovery_frame_cnt from SEI message
664 * Set to -1 if no recovery point SEI message found or to number of frames
665 * before playback synchronizes. Frames having recovery point are key
668 int sei_recovery_frame_cnt;
671 * recovery_frame is the frame_num at which the next frame should
672 * be fully constructed.
674 * Set to -1 when not expecting a recovery point.
679 * We have seen an IDR, so all the following frames in coded order are correctly
682 #define FRAME_RECOVERED_IDR (1 << 0)
684 * Sufficient number of frames have been decoded since a SEI recovery point,
685 * so all the following frames in presentation order are correct.
687 #define FRAME_RECOVERED_SEI (1 << 1)
689 int frame_recovered; ///< Initial frame has been completely recovered
691 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
692 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
695 int sei_buffering_period_present; ///< Buffering period SEI flag
696 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
698 int cur_chroma_format_idc;
699 uint8_t *bipred_scratchpad;
700 uint8_t *edge_emu_buffer;
701 int16_t *dc_val_base;
703 AVBufferPool *qscale_table_pool;
704 AVBufferPool *mb_type_pool;
705 AVBufferPool *motion_val_pool;
706 AVBufferPool *ref_index_pool;
708 /* Motion Estimation */
709 qpel_mc_func (*qpel_put)[16];
710 qpel_mc_func (*qpel_avg)[16];
713 extern const uint8_t ff_h264_chroma_qp[3][QP_MAX_NUM + 1]; ///< One chroma qp table for each supported bit depth (8, 9, 10).
714 extern const uint16_t ff_h264_mb_sizes[4];
719 int ff_h264_decode_sei(H264Context *h);
724 int ff_h264_decode_seq_parameter_set(H264Context *h);
727 * compute profile from sps
729 int ff_h264_get_profile(SPS *sps);
734 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
737 * Decode a network abstraction layer unit.
738 * @param consumed is the number of bytes used as input
739 * @param length is the length of the array
740 * @param dst_length is the number of decoded bytes FIXME here
741 * or a decode rbsp tailing?
742 * @return decoded bytes, might be src+1 if no escapes
744 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
745 int *dst_length, int *consumed, int length);
748 * Free any data that may have been allocated in the H264 context
751 void ff_h264_free_context(H264Context *h);
754 * Reconstruct bitstream slice_type.
756 int ff_h264_get_slice_type(const H264Context *h);
762 int ff_h264_alloc_tables(H264Context *h);
765 * Fill the default_ref_list.
767 int ff_h264_fill_default_ref_list(H264Context *h);
769 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
770 void ff_h264_fill_mbaff_ref_list(H264Context *h);
771 void ff_h264_remove_all_refs(H264Context *h);
774 * Execute the reference picture marking (memory management control operations).
776 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
778 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb,
781 int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
784 * Check if the top & left blocks are available if needed & change the
785 * dc mode so it only uses the available blocks.
787 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
790 * Check if the top & left blocks are available if needed & change the
791 * dc mode so it only uses the available blocks.
793 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma);
795 void ff_h264_hl_decode_mb(H264Context *h);
796 int ff_h264_decode_extradata(H264Context *h);
797 int ff_h264_decode_init(AVCodecContext *avctx);
798 void ff_h264_decode_init_vlc(void);
801 * Decode a macroblock
802 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
804 int ff_h264_decode_mb_cavlc(H264Context *h);
807 * Decode a CABAC coded macroblock
808 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
810 int ff_h264_decode_mb_cabac(H264Context *h);
812 void ff_h264_init_cabac_states(H264Context *h);
814 void h264_init_dequant_tables(H264Context *h);
816 void ff_h264_direct_dist_scale_factor(H264Context *const h);
817 void ff_h264_direct_ref_list_init(H264Context *const h);
818 void ff_h264_pred_direct_motion(H264Context *const h, int *mb_type);
820 void ff_h264_filter_mb_fast(H264Context *h, int mb_x, int mb_y,
821 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
822 unsigned int linesize, unsigned int uvlinesize);
823 void ff_h264_filter_mb(H264Context *h, int mb_x, int mb_y,
824 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
825 unsigned int linesize, unsigned int uvlinesize);
828 * Reset SEI values at the beginning of the frame.
830 * @param h H.264 context.
832 void ff_h264_reset_sei(H264Context *h);
844 /* Scan8 organization:
861 * DY/DU/DV are for luma/chroma DC.
864 #define LUMA_DC_BLOCK_INDEX 48
865 #define CHROMA_DC_BLOCK_INDEX 49
867 // This table must be here because scan8[constant] must be known at compiletime
868 static const uint8_t scan8[16 * 3 + 3] = {
869 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
870 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
871 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
872 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
873 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
874 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
875 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
876 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
877 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
878 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
879 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
880 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
881 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
884 static av_always_inline uint32_t pack16to32(int a, int b)
887 return (b & 0xFFFF) + (a << 16);
889 return (a & 0xFFFF) + (b << 16);
893 static av_always_inline uint16_t pack8to16(int a, int b)
896 return (b & 0xFF) + (a << 8);
898 return (a & 0xFF) + (b << 8);
905 static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale)
907 return h->pps.chroma_qp_table[t][qscale];
911 * Get the predicted intra4x4 prediction mode.
913 static av_always_inline int pred_intra_mode(H264Context *h, int n)
915 const int index8 = scan8[n];
916 const int left = h->intra4x4_pred_mode_cache[index8 - 1];
917 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
918 const int min = FFMIN(left, top);
920 tprintf(h->avctx, "mode:%d %d min:%d\n", left, top, min);
928 static av_always_inline void write_back_intra_pred_mode(H264Context *h)
930 int8_t *i4x4 = h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
931 int8_t *i4x4_cache = h->intra4x4_pred_mode_cache;
933 AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
934 i4x4[4] = i4x4_cache[7 + 8 * 3];
935 i4x4[5] = i4x4_cache[7 + 8 * 2];
936 i4x4[6] = i4x4_cache[7 + 8 * 1];
939 static av_always_inline void write_back_non_zero_count(H264Context *h)
941 const int mb_xy = h->mb_xy;
942 uint8_t *nnz = h->non_zero_count[mb_xy];
943 uint8_t *nnz_cache = h->non_zero_count_cache;
945 AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
946 AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
947 AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
948 AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
949 AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
950 AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
951 AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
952 AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
954 if (!h->chroma_y_shift) {
955 AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
956 AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
957 AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
958 AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
962 static av_always_inline void write_back_motion_list(H264Context *h,
965 int mb_type, int list)
967 int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
968 int16_t(*mv_src)[2] = &h->mv_cache[list][scan8[0]];
969 AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
970 AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
971 AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
972 AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
974 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8 * h->mb_xy
975 : h->mb2br_xy[h->mb_xy]];
976 uint8_t(*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
977 if (IS_SKIP(mb_type)) {
980 AV_COPY64(mvd_dst, mvd_src + 8 * 3);
981 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
982 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
983 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
988 int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
989 int8_t *ref_cache = h->ref_cache[list];
990 ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
991 ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
992 ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
993 ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
997 static av_always_inline void write_back_motion(H264Context *h, int mb_type)
999 const int b_stride = h->b_stride;
1000 const int b_xy = 4 * h->mb_x + 4 * h->mb_y * h->b_stride; // try mb2b(8)_xy
1001 const int b8_xy = 4 * h->mb_xy;
1003 if (USES_LIST(mb_type, 0)) {
1004 write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 0);
1006 fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
1007 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1009 if (USES_LIST(mb_type, 1))
1010 write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 1);
1012 if (h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) {
1013 if (IS_8X8(mb_type)) {
1014 uint8_t *direct_table = &h->direct_table[4 * h->mb_xy];
1015 direct_table[1] = h->sub_mb_type[1] >> 1;
1016 direct_table[2] = h->sub_mb_type[2] >> 1;
1017 direct_table[3] = h->sub_mb_type[3] >> 1;
1022 static av_always_inline int get_dct8x8_allowed(H264Context *h)
1024 if (h->sps.direct_8x8_inference_flag)
1025 return !(AV_RN64A(h->sub_mb_type) &
1026 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
1027 0x0001000100010001ULL));
1029 return !(AV_RN64A(h->sub_mb_type) &
1030 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
1031 0x0001000100010001ULL));
1034 int ff_h264_field_end(H264Context *h, int in_setup);
1036 int ff_h264_ref_picture(H264Context *h, H264Picture *dst, H264Picture *src);
1037 void ff_h264_unref_picture(H264Context *h, H264Picture *pic);
1039 int ff_h264_context_init(H264Context *h);
1040 int ff_h264_set_parameter_from_sps(H264Context *h);
1042 void ff_h264_draw_horiz_band(H264Context *h, int y, int height);
1043 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc);
1044 int ff_pred_weight_table(H264Context *h);
1045 int ff_set_ref_count(H264Context *h);
1047 int ff_h264_decode_slice_header(H264Context *h, H264Context *h0);
1048 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count);
1049 int ff_h264_update_thread_context(AVCodecContext *dst,
1050 const AVCodecContext *src);
1052 void ff_h264_flush_change(H264Context *h);
1054 void ff_h264_free_tables(H264Context *h, int free_rbsp);
1056 #endif /* AVCODEC_H264_H */