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"
39 #include "mpegutils.h"
42 #include "rectangle.h"
45 #define H264_MAX_PICTURE_COUNT 32
46 #define H264_MAX_THREADS 16
48 #define MAX_SPS_COUNT 32
49 #define MAX_PPS_COUNT 256
51 #define MAX_MMCO_COUNT 66
53 #define MAX_DELAYED_PIC_COUNT 16
55 /* Compiling in interlaced support reduces the speed
56 * of progressive decoding by about 2%. */
57 #define ALLOW_INTERLACE
62 * The maximum number of slices supported by the decoder.
63 * must be a power of 2
67 #ifdef ALLOW_INTERLACE
68 #define MB_MBAFF(h) h->mb_mbaff
69 #define MB_FIELD(h) h->mb_field_decoding_flag
70 #define FRAME_MBAFF(h) h->mb_aff_frame
71 #define FIELD_PICTURE(h) (h->picture_structure != PICT_FRAME)
79 #define FRAME_MBAFF(h) 0
80 #define FIELD_PICTURE(h) 0
82 #define IS_INTERLACED(mb_type) 0
88 #define FIELD_OR_MBAFF_PICTURE(h) (FRAME_MBAFF(h) || FIELD_PICTURE(h))
91 #define CABAC(h) h->pps.cabac
94 #define CHROMA422(h) (h->sps.chroma_format_idc == 2)
95 #define CHROMA444(h) (h->sps.chroma_format_idc == 3)
97 #define EXTENDED_SAR 255
99 #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit
100 #define MB_TYPE_8x8DCT 0x01000000
101 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
102 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
104 #define QP_MAX_NUM (51 + 2 * 6) // The maximum supported qp
117 NAL_END_SEQUENCE = 10,
119 NAL_FILLER_DATA = 12,
121 NAL_AUXILIARY_SLICE = 19,
122 NAL_FF_IGNORE = 0xff0f001,
129 SEI_TYPE_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
130 SEI_TYPE_PIC_TIMING = 1, ///< picture timing
131 SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
132 SEI_TYPE_RECOVERY_POINT = 6, ///< recovery point (frame # to decoder sync)
133 SEI_TYPE_FRAME_PACKING = 45, ///< frame packing arrangement
134 SEI_TYPE_DISPLAY_ORIENTATION = 47, ///< display orientation
138 * pic_struct in picture timing SEI message
141 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
142 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
143 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
144 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
145 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
146 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
147 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
148 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
149 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
153 * Sequence parameter set
159 int chroma_format_idc;
160 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
161 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
162 int poc_type; ///< pic_order_cnt_type
163 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
164 int delta_pic_order_always_zero_flag;
165 int offset_for_non_ref_pic;
166 int offset_for_top_to_bottom_field;
167 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
168 int ref_frame_count; ///< num_ref_frames
169 int gaps_in_frame_num_allowed_flag;
170 int mb_width; ///< pic_width_in_mbs_minus1 + 1
171 int mb_height; ///< pic_height_in_map_units_minus1 + 1
172 int frame_mbs_only_flag;
173 int mb_aff; ///< mb_adaptive_frame_field_flag
174 int direct_8x8_inference_flag;
175 int crop; ///< frame_cropping_flag
177 /* those 4 are already in luma samples */
178 unsigned int crop_left; ///< frame_cropping_rect_left_offset
179 unsigned int crop_right; ///< frame_cropping_rect_right_offset
180 unsigned int crop_top; ///< frame_cropping_rect_top_offset
181 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
182 int vui_parameters_present_flag;
184 int video_signal_type_present_flag;
186 int colour_description_present_flag;
187 enum AVColorPrimaries color_primaries;
188 enum AVColorTransferCharacteristic color_trc;
189 enum AVColorSpace colorspace;
190 int timing_info_present_flag;
191 uint32_t num_units_in_tick;
193 int fixed_frame_rate_flag;
194 short offset_for_ref_frame[256]; // FIXME dyn aloc?
195 int bitstream_restriction_flag;
196 int num_reorder_frames;
197 int scaling_matrix_present;
198 uint8_t scaling_matrix4[6][16];
199 uint8_t scaling_matrix8[6][64];
200 int nal_hrd_parameters_present_flag;
201 int vcl_hrd_parameters_present_flag;
202 int pic_struct_present_flag;
203 int time_offset_length;
204 int cpb_cnt; ///< See H.264 E.1.2
205 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
206 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
207 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
208 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
209 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
210 int residual_color_transform_flag; ///< residual_colour_transform_flag
211 int constraint_set_flags; ///< constraint_set[0-3]_flag
212 int new; ///< flag to keep track if the decoder context needs re-init due to changed SPS
216 * Picture parameter set
220 int cabac; ///< entropy_coding_mode_flag
221 int pic_order_present; ///< pic_order_present_flag
222 int slice_group_count; ///< num_slice_groups_minus1 + 1
223 int mb_slice_group_map_type;
224 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
225 int weighted_pred; ///< weighted_pred_flag
226 int weighted_bipred_idc;
227 int init_qp; ///< pic_init_qp_minus26 + 26
228 int init_qs; ///< pic_init_qs_minus26 + 26
229 int chroma_qp_index_offset[2];
230 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
231 int constrained_intra_pred; ///< constrained_intra_pred_flag
232 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
233 int transform_8x8_mode; ///< transform_8x8_mode_flag
234 uint8_t scaling_matrix4[6][16];
235 uint8_t scaling_matrix8[6][64];
236 uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
241 * Memory management control operation opcode.
243 typedef enum MMCOOpcode {
254 * Memory management control operation.
256 typedef struct MMCO {
258 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
259 int long_arg; ///< index, pic_num, or num long refs depending on opcode
262 typedef struct H264Picture {
266 AVBufferRef *qscale_table_buf;
267 int8_t *qscale_table;
269 AVBufferRef *motion_val_buf[2];
270 int16_t (*motion_val[2])[2];
272 AVBufferRef *mb_type_buf;
275 AVBufferRef *hwaccel_priv_buf;
276 void *hwaccel_picture_private; ///< hardware accelerator private data
278 AVBufferRef *ref_index_buf[2];
279 int8_t *ref_index[2];
281 int field_poc[2]; ///< top/bottom POC
282 int poc; ///< frame POC
283 int frame_num; ///< frame_num (raw frame_num from slice header)
284 int mmco_reset; /**< MMCO_RESET set this 1. Reordering code must
285 not mix pictures before and after MMCO_RESET. */
286 int pic_id; /**< pic_num (short -> no wrap version of pic_num,
287 pic_num & max_pic_num; long -> long_pic_num) */
288 int long_ref; ///< 1->long term reference 0->short term reference
289 int ref_poc[2][2][32]; ///< POCs of the frames used as reference (FIXME need per slice)
290 int ref_count[2][2]; ///< number of entries in ref_poc (FIXME need per slice)
291 int mbaff; ///< 1 -> MBAFF frame 0-> not MBAFF
292 int field_picture; ///< whether or not picture was encoded in separate fields
294 int needs_realloc; ///< picture needs to be reallocated (eg due to a frame size change)
296 int recovered; ///< picture at IDR or recovery point + recovery count
299 typedef struct H264SliceContext {
300 struct H264Context *h264;
304 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
305 int slice_type_fixed;
308 int chroma_qp[2]; // QPc
309 int qp_thresh; ///< QP threshold to skip loopfilter
311 // Weighted pred stuff
313 int use_weight_chroma;
314 int luma_log2_weight_denom;
315 int chroma_log2_weight_denom;
316 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
317 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
318 // The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
319 int luma_weight[48][2][2];
320 int chroma_weight[48][2][2][2];
321 int implicit_weight[48][48][2];
326 int chroma_pred_mode;
327 int intra16x16_pred_mode;
329 int8_t intra4x4_pred_mode_cache[5 * 8];
330 int8_t(*intra4x4_pred_mode);
335 int left_mb_xy[LEFT_MBS];
340 int left_type[LEFT_MBS];
342 const uint8_t *left_block;
343 int topleft_partition;
345 unsigned int topleft_samples_available;
346 unsigned int top_samples_available;
347 unsigned int topright_samples_available;
348 unsigned int left_samples_available;
350 ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
351 ptrdiff_t mb_uvlinesize;
354 * number of neighbors (top and/or left) that used 8x8 dct
356 int neighbor_transform_size;
358 int direct_spatial_mv_pred;
362 int dist_scale_factor[32];
363 int dist_scale_factor_field[2][32];
366 * non zero coeff count cache.
367 * is 64 if not available.
369 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
372 * Motion vector cache.
374 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
375 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
377 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
383 typedef struct H264Context {
384 AVCodecContext *avctx;
385 VideoDSPContext vdsp;
386 H264DSPContext h264dsp;
387 H264ChromaContext h264chroma;
388 H264QpelContext h264qpel;
393 H264Picture *cur_pic_ptr;
396 H264SliceContext *slice_ctx;
399 int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
401 /* coded dimensions -- 16 * mb w/h */
403 ptrdiff_t linesize, uvlinesize;
404 int chroma_x_shift, chroma_y_shift;
407 int coded_picture_number;
410 int context_initialized;
414 int8_t(*intra4x4_pred_mode);
416 uint8_t (*top_borders[2])[(16 * 3) * 2];
418 uint8_t (*non_zero_count)[48];
420 #define LIST_NOT_USED -1 // FIXME rename?
421 #define PART_NOT_AVAILABLE -2
424 * block_offset[ 0..23] for frame macroblocks
425 * block_offset[24..47] for field macroblocks
427 int block_offset[2 * (16 * 3)];
429 uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
431 int b_stride; // FIXME use s->b4_stride
433 SPS sps; ///< current sps
434 PPS pps; ///< current pps
436 uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
437 uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
438 uint32_t(*dequant4_coeff[6])[16];
439 uint32_t(*dequant8_coeff[6])[64];
441 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
443 // interlacing specific flags
445 int mb_field_decoding_flag;
446 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
447 int picture_structure;
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 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
459 H264Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
460 * Reordered version of default_ref_list
461 * according to picture reordering in slice header */
462 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
465 GetBitContext intra_gb;
466 GetBitContext inter_gb;
467 GetBitContext *intra_gb_ptr;
468 GetBitContext *inter_gb_ptr;
470 const uint8_t *intra_pcm_ptr;
471 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.
472 DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
473 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
479 uint8_t cabac_state[1024];
481 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
486 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
487 uint8_t *chroma_pred_mode_table;
488 int last_qscale_diff;
489 uint8_t (*mvd_table[2])[2];
490 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
491 uint8_t *direct_table;
492 uint8_t direct_cache[5 * 8];
494 uint8_t zigzag_scan[16];
495 uint8_t zigzag_scan8x8[64];
496 uint8_t zigzag_scan8x8_cavlc[64];
497 uint8_t field_scan[16];
498 uint8_t field_scan8x8[64];
499 uint8_t field_scan8x8_cavlc[64];
500 const uint8_t *zigzag_scan_q0;
501 const uint8_t *zigzag_scan8x8_q0;
502 const uint8_t *zigzag_scan8x8_cavlc_q0;
503 const uint8_t *field_scan_q0;
504 const uint8_t *field_scan8x8_q0;
505 const uint8_t *field_scan8x8_cavlc_q0;
513 int mb_height, mb_width;
521 int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
522 int slice_alpha_c0_offset;
523 int slice_beta_offset;
525 // =============================================================
526 // Things below are not used in the MB or more inner code
530 uint8_t *rbsp_buffer[2];
531 unsigned int rbsp_buffer_size[2];
534 * Used to parse AVC variant of h264
536 int is_avc; ///< this flag is != 0 if codec is avc1
537 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
539 int bit_depth_luma; ///< luma bit depth from sps to detect changes
540 int chroma_format_idc; ///< chroma format from sps to detect changes
542 SPS *sps_buffers[MAX_SPS_COUNT];
543 PPS *pps_buffers[MAX_PPS_COUNT];
545 int dequant_coeff_pps; ///< reinit tables when pps changes
547 uint16_t *slice_table_base;
552 int delta_poc_bottom;
555 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
556 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
557 int frame_num_offset; ///< for POC type 2
558 int prev_frame_num_offset; ///< for POC type 2
559 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
562 * frame_num for frames or 2 * frame_num + 1 for field pics.
567 * max_frame_num or 2 * max_frame_num for field pics.
571 int redundant_pic_count;
573 H264Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
574 H264Picture *short_ref[32];
575 H264Picture *long_ref[32];
576 H264Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
577 int last_pocs[MAX_DELAYED_PIC_COUNT];
578 H264Picture *next_output_pic;
580 int next_outputed_poc;
583 * memory management control operations buffer.
585 MMCO mmco[MAX_MMCO_COUNT];
589 int long_ref_count; ///< number of actual long term references
590 int short_ref_count; ///< number of actual short term references
595 * @name Members for slice based multithreading
598 struct H264Context *thread_context[H264_MAX_THREADS];
601 * current slice number, used to initalize slice_num of each thread/context
606 * Max number of threads / contexts.
607 * This is equal to AVCodecContext.thread_count unless
608 * multithreaded decoding is impossible, in which case it is
613 int slice_context_count;
616 * 1 if the single thread fallback warning has already been
617 * displayed, 0 otherwise.
619 int single_decode_warning;
621 enum AVPictureType pict_type;
627 * pic_struct in picture timing SEI message
629 SEI_PicStructType sei_pic_struct;
632 * Complement sei_pic_struct
633 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
634 * However, soft telecined frames may have these values.
635 * This is used in an attempt to flag soft telecine progressive.
637 int prev_interlaced_frame;
640 * frame_packing_arrangment SEI message
642 int sei_frame_packing_present;
643 int frame_packing_arrangement_type;
644 int content_interpretation_type;
645 int quincunx_subsampling;
648 * display orientation SEI message
650 int sei_display_orientation_present;
651 int sei_anticlockwise_rotation;
652 int sei_hflip, sei_vflip;
655 * Bit set of clock types for fields/frames in picture timing SEI message.
656 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
662 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
664 int sei_dpb_output_delay;
667 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
669 int sei_cpb_removal_delay;
672 * recovery_frame_cnt from SEI message
674 * Set to -1 if no recovery point SEI message found or to number of frames
675 * before playback synchronizes. Frames having recovery point are key
678 int sei_recovery_frame_cnt;
681 * recovery_frame is the frame_num at which the next frame should
682 * be fully constructed.
684 * Set to -1 when not expecting a recovery point.
689 * We have seen an IDR, so all the following frames in coded order are correctly
692 #define FRAME_RECOVERED_IDR (1 << 0)
694 * Sufficient number of frames have been decoded since a SEI recovery point,
695 * so all the following frames in presentation order are correct.
697 #define FRAME_RECOVERED_SEI (1 << 1)
699 int frame_recovered; ///< Initial frame has been completely recovered
703 int sei_buffering_period_present; ///< Buffering period SEI flag
704 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
706 int cur_chroma_format_idc;
707 uint8_t *bipred_scratchpad;
708 uint8_t *edge_emu_buffer;
709 int16_t *dc_val_base;
711 AVBufferPool *qscale_table_pool;
712 AVBufferPool *mb_type_pool;
713 AVBufferPool *motion_val_pool;
714 AVBufferPool *ref_index_pool;
716 /* Motion Estimation */
717 qpel_mc_func (*qpel_put)[16];
718 qpel_mc_func (*qpel_avg)[16];
721 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).
722 extern const uint16_t ff_h264_mb_sizes[4];
727 int ff_h264_decode_sei(H264Context *h);
732 int ff_h264_decode_seq_parameter_set(H264Context *h);
735 * compute profile from sps
737 int ff_h264_get_profile(SPS *sps);
742 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
745 * Decode a network abstraction layer unit.
746 * @param consumed is the number of bytes used as input
747 * @param length is the length of the array
748 * @param dst_length is the number of decoded bytes FIXME here
749 * or a decode rbsp tailing?
750 * @return decoded bytes, might be src+1 if no escapes
752 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
753 int *dst_length, int *consumed, int length);
756 * Free any data that may have been allocated in the H264 context
759 void ff_h264_free_context(H264Context *h);
762 * Reconstruct bitstream slice_type.
764 int ff_h264_get_slice_type(const H264SliceContext *sl);
770 int ff_h264_alloc_tables(H264Context *h);
773 * Fill the default_ref_list.
775 int ff_h264_fill_default_ref_list(H264Context *h, H264SliceContext *sl);
777 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
778 void ff_h264_fill_mbaff_ref_list(H264Context *h, H264SliceContext *sl);
779 void ff_h264_remove_all_refs(H264Context *h);
782 * Execute the reference picture marking (memory management control operations).
784 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
786 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb,
789 int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
792 * Check if the top & left blocks are available if needed & change the
793 * dc mode so it only uses the available blocks.
795 int ff_h264_check_intra4x4_pred_mode(H264Context *h, H264SliceContext *sl);
798 * Check if the top & left blocks are available if needed & change the
799 * dc mode so it only uses the available blocks.
801 int ff_h264_check_intra_pred_mode(H264Context *h, H264SliceContext *sl,
802 int mode, int is_chroma);
804 void ff_h264_hl_decode_mb(H264Context *h, H264SliceContext *sl);
805 int ff_h264_decode_extradata(H264Context *h);
806 int ff_h264_decode_init(AVCodecContext *avctx);
807 void ff_h264_decode_init_vlc(void);
810 * Decode a macroblock
811 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
813 int ff_h264_decode_mb_cavlc(H264Context *h, H264SliceContext *sl);
816 * Decode a CABAC coded macroblock
817 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
819 int ff_h264_decode_mb_cabac(H264Context *h, H264SliceContext *sl);
821 void ff_h264_init_cabac_states(H264Context *h, H264SliceContext *sl);
823 void h264_init_dequant_tables(H264Context *h);
825 void ff_h264_direct_dist_scale_factor(H264Context *const h, H264SliceContext *sl);
826 void ff_h264_direct_ref_list_init(H264Context *const h, H264SliceContext *sl);
827 void ff_h264_pred_direct_motion(H264Context *const h, H264SliceContext *sl,
830 void ff_h264_filter_mb_fast(H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
831 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
832 unsigned int linesize, unsigned int uvlinesize);
833 void ff_h264_filter_mb(H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
834 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
835 unsigned int linesize, unsigned int uvlinesize);
838 * Reset SEI values at the beginning of the frame.
840 * @param h H.264 context.
842 void ff_h264_reset_sei(H264Context *h);
854 /* Scan8 organization:
871 * DY/DU/DV are for luma/chroma DC.
874 #define LUMA_DC_BLOCK_INDEX 48
875 #define CHROMA_DC_BLOCK_INDEX 49
877 // This table must be here because scan8[constant] must be known at compiletime
878 static const uint8_t scan8[16 * 3 + 3] = {
879 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
880 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
881 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
882 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
883 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
884 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
885 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
886 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
887 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
888 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
889 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
890 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
891 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
894 static av_always_inline uint32_t pack16to32(int a, int b)
897 return (b & 0xFFFF) + (a << 16);
899 return (a & 0xFFFF) + (b << 16);
903 static av_always_inline uint16_t pack8to16(int a, int b)
906 return (b & 0xFF) + (a << 8);
908 return (a & 0xFF) + (b << 8);
915 static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale)
917 return h->pps.chroma_qp_table[t][qscale];
921 * Get the predicted intra4x4 prediction mode.
923 static av_always_inline int pred_intra_mode(H264Context *h,
924 H264SliceContext *sl, int n)
926 const int index8 = scan8[n];
927 const int left = sl->intra4x4_pred_mode_cache[index8 - 1];
928 const int top = sl->intra4x4_pred_mode_cache[index8 - 8];
929 const int min = FFMIN(left, top);
931 tprintf(h->avctx, "mode:%d %d min:%d\n", left, top, min);
939 static av_always_inline void write_back_intra_pred_mode(H264Context *h,
940 H264SliceContext *sl)
942 int8_t *i4x4 = sl->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
943 int8_t *i4x4_cache = sl->intra4x4_pred_mode_cache;
945 AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
946 i4x4[4] = i4x4_cache[7 + 8 * 3];
947 i4x4[5] = i4x4_cache[7 + 8 * 2];
948 i4x4[6] = i4x4_cache[7 + 8 * 1];
951 static av_always_inline void write_back_non_zero_count(H264Context *h,
952 H264SliceContext *sl)
954 const int mb_xy = h->mb_xy;
955 uint8_t *nnz = h->non_zero_count[mb_xy];
956 uint8_t *nnz_cache = sl->non_zero_count_cache;
958 AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
959 AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
960 AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
961 AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
962 AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
963 AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
964 AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
965 AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
967 if (!h->chroma_y_shift) {
968 AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
969 AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
970 AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
971 AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
975 static av_always_inline void write_back_motion_list(H264Context *h,
976 H264SliceContext *sl,
979 int mb_type, int list)
981 int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
982 int16_t(*mv_src)[2] = &sl->mv_cache[list][scan8[0]];
983 AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
984 AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
985 AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
986 AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
988 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8 * h->mb_xy
989 : h->mb2br_xy[h->mb_xy]];
990 uint8_t(*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
991 if (IS_SKIP(mb_type)) {
994 AV_COPY64(mvd_dst, mvd_src + 8 * 3);
995 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
996 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
997 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
1002 int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
1003 int8_t *ref_cache = sl->ref_cache[list];
1004 ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
1005 ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
1006 ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
1007 ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
1011 static av_always_inline void write_back_motion(H264Context *h,
1012 H264SliceContext *sl,
1015 const int b_stride = h->b_stride;
1016 const int b_xy = 4 * h->mb_x + 4 * h->mb_y * h->b_stride; // try mb2b(8)_xy
1017 const int b8_xy = 4 * h->mb_xy;
1019 if (USES_LIST(mb_type, 0)) {
1020 write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 0);
1022 fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
1023 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1025 if (USES_LIST(mb_type, 1))
1026 write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 1);
1028 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) {
1029 if (IS_8X8(mb_type)) {
1030 uint8_t *direct_table = &h->direct_table[4 * h->mb_xy];
1031 direct_table[1] = sl->sub_mb_type[1] >> 1;
1032 direct_table[2] = sl->sub_mb_type[2] >> 1;
1033 direct_table[3] = sl->sub_mb_type[3] >> 1;
1038 static av_always_inline int get_dct8x8_allowed(H264Context *h, H264SliceContext *sl)
1040 if (h->sps.direct_8x8_inference_flag)
1041 return !(AV_RN64A(sl->sub_mb_type) &
1042 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
1043 0x0001000100010001ULL));
1045 return !(AV_RN64A(sl->sub_mb_type) &
1046 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
1047 0x0001000100010001ULL));
1050 int ff_h264_field_end(H264Context *h, int in_setup);
1052 int ff_h264_ref_picture(H264Context *h, H264Picture *dst, H264Picture *src);
1053 void ff_h264_unref_picture(H264Context *h, H264Picture *pic);
1055 int ff_h264_context_init(H264Context *h);
1056 int ff_h264_set_parameter_from_sps(H264Context *h);
1058 void ff_h264_draw_horiz_band(H264Context *h, int y, int height);
1059 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc);
1060 int ff_pred_weight_table(H264Context *h, H264SliceContext *sl);
1061 int ff_set_ref_count(H264Context *h, H264SliceContext *sl);
1063 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Context *h0);
1064 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count);
1065 int ff_h264_update_thread_context(AVCodecContext *dst,
1066 const AVCodecContext *src);
1068 void ff_h264_flush_change(H264Context *h);
1070 void ff_h264_free_tables(H264Context *h, int free_rbsp);
1072 #endif /* AVCODEC_H264_H */