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 / MPEG-4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #ifndef AVCODEC_H264_H
29 #define AVCODEC_H264_H
31 #include "libavutil/buffer.h"
32 #include "libavutil/intreadwrite.h"
33 #include "libavutil/thread.h"
36 #include "error_resilience.h"
37 #include "h264_parse.h"
39 #include "h2645_parse.h"
40 #include "h264chroma.h"
45 #include "mpegutils.h"
48 #include "rectangle.h"
51 #define H264_MAX_PICTURE_COUNT 36
53 #define MAX_SPS_COUNT 32
54 #define MAX_PPS_COUNT 256
56 #define MAX_MMCO_COUNT 66
58 #define MAX_DELAYED_PIC_COUNT 16
60 #define MAX_MBPAIR_SIZE (256*1024) // a tighter bound could be calculated if someone cares about a few bytes
62 /* Compiling in interlaced support reduces the speed
63 * of progressive decoding by about 2%. */
64 #define ALLOW_INTERLACE
69 * The maximum number of slices supported by the decoder.
70 * must be a power of 2
74 #ifdef ALLOW_INTERLACE
75 #define MB_MBAFF(h) (h)->mb_mbaff
76 #define MB_FIELD(sl) (sl)->mb_field_decoding_flag
77 #define FRAME_MBAFF(h) (h)->mb_aff_frame
78 #define FIELD_PICTURE(h) ((h)->picture_structure != PICT_FRAME)
85 #define MB_FIELD(sl) 0
86 #define FRAME_MBAFF(h) 0
87 #define FIELD_PICTURE(h) 0
89 #define IS_INTERLACED(mb_type) 0
95 #define FIELD_OR_MBAFF_PICTURE(h) (FRAME_MBAFF(h) || FIELD_PICTURE(h))
98 #define CABAC(h) (h)->ps.pps->cabac
101 #define CHROMA(h) ((h)->ps.sps->chroma_format_idc)
102 #define CHROMA422(h) ((h)->ps.sps->chroma_format_idc == 2)
103 #define CHROMA444(h) ((h)->ps.sps->chroma_format_idc == 3)
105 #define EXTENDED_SAR 255
107 #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit
108 #define MB_TYPE_8x8DCT 0x01000000
109 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
110 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
112 #define QP_MAX_NUM (51 + 6*6) // The maximum supported qp
125 NAL_END_SEQUENCE = 10,
127 NAL_FILLER_DATA = 12,
129 NAL_AUXILIARY_SLICE = 19,
130 NAL_FF_IGNORE = 0xff0f001,
134 * Sequence parameter set
140 int chroma_format_idc;
141 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
142 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
143 int poc_type; ///< pic_order_cnt_type
144 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
145 int delta_pic_order_always_zero_flag;
146 int offset_for_non_ref_pic;
147 int offset_for_top_to_bottom_field;
148 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
149 int ref_frame_count; ///< num_ref_frames
150 int gaps_in_frame_num_allowed_flag;
151 int mb_width; ///< pic_width_in_mbs_minus1 + 1
152 int mb_height; ///< pic_height_in_map_units_minus1 + 1
153 int frame_mbs_only_flag;
154 int mb_aff; ///< mb_adaptive_frame_field_flag
155 int direct_8x8_inference_flag;
156 int crop; ///< frame_cropping_flag
158 /* those 4 are already in luma samples */
159 unsigned int crop_left; ///< frame_cropping_rect_left_offset
160 unsigned int crop_right; ///< frame_cropping_rect_right_offset
161 unsigned int crop_top; ///< frame_cropping_rect_top_offset
162 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
163 int vui_parameters_present_flag;
165 int video_signal_type_present_flag;
167 int colour_description_present_flag;
168 enum AVColorPrimaries color_primaries;
169 enum AVColorTransferCharacteristic color_trc;
170 enum AVColorSpace colorspace;
171 int timing_info_present_flag;
172 uint32_t num_units_in_tick;
174 int fixed_frame_rate_flag;
175 short offset_for_ref_frame[256]; // FIXME dyn aloc?
176 int bitstream_restriction_flag;
177 int num_reorder_frames;
178 int scaling_matrix_present;
179 uint8_t scaling_matrix4[6][16];
180 uint8_t scaling_matrix8[6][64];
181 int nal_hrd_parameters_present_flag;
182 int vcl_hrd_parameters_present_flag;
183 int pic_struct_present_flag;
184 int time_offset_length;
185 int cpb_cnt; ///< See H.264 E.1.2
186 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
187 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
188 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
189 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
190 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
191 int residual_color_transform_flag; ///< residual_colour_transform_flag
192 int constraint_set_flags; ///< constraint_set[0-3]_flag
198 * Picture parameter set
202 int cabac; ///< entropy_coding_mode_flag
203 int pic_order_present; ///< pic_order_present_flag
204 int slice_group_count; ///< num_slice_groups_minus1 + 1
205 int mb_slice_group_map_type;
206 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
207 int weighted_pred; ///< weighted_pred_flag
208 int weighted_bipred_idc;
209 int init_qp; ///< pic_init_qp_minus26 + 26
210 int init_qs; ///< pic_init_qs_minus26 + 26
211 int chroma_qp_index_offset[2];
212 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
213 int constrained_intra_pred; ///< constrained_intra_pred_flag
214 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
215 int transform_8x8_mode; ///< transform_8x8_mode_flag
216 uint8_t scaling_matrix4[6][16];
217 uint8_t scaling_matrix8[6][64];
218 uint8_t chroma_qp_table[2][QP_MAX_NUM+1]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
223 uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16];
224 uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
225 uint32_t(*dequant4_coeff[6])[16];
226 uint32_t(*dequant8_coeff[6])[64];
229 typedef struct H264ParamSets {
230 AVBufferRef *sps_list[MAX_SPS_COUNT];
231 AVBufferRef *pps_list[MAX_PPS_COUNT];
233 AVBufferRef *pps_ref;
234 AVBufferRef *sps_ref;
235 /* currently active parameters sets */
237 // FIXME this should properly be const
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/fields 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
296 int recovered; ///< picture at IDR or recovery point + recovery count
298 int sei_recovery_frame_cnt;
305 typedef struct H264Ref {
316 typedef struct H264SliceContext {
317 struct H264Context *h264;
323 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
324 int slice_type_fixed;
327 int chroma_qp[2]; // QPc
328 int qp_thresh; ///< QP threshold to skip loopfilter
329 int last_qscale_diff;
332 int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
333 int slice_alpha_c0_offset;
334 int slice_beta_offset;
336 H264PredWeightTable pwt;
341 int chroma_pred_mode;
342 int intra16x16_pred_mode;
344 int8_t intra4x4_pred_mode_cache[5 * 8];
345 int8_t(*intra4x4_pred_mode);
350 int left_mb_xy[LEFT_MBS];
355 int left_type[LEFT_MBS];
357 const uint8_t *left_block;
358 int topleft_partition;
360 unsigned int topleft_samples_available;
361 unsigned int top_samples_available;
362 unsigned int topright_samples_available;
363 unsigned int left_samples_available;
365 ptrdiff_t linesize, uvlinesize;
366 ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
367 ptrdiff_t mb_uvlinesize;
373 // index of the first MB of the next slice
378 int mb_field_decoding_flag;
379 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
381 int redundant_pic_count;
384 * number of neighbors (top and/or left) that used 8x8 dct
386 int neighbor_transform_size;
388 int direct_spatial_mv_pred;
396 int dist_scale_factor[32];
397 int dist_scale_factor_field[2][32];
398 int map_col_to_list0[2][16 + 32];
399 int map_col_to_list0_field[2][2][16 + 32];
402 * num_ref_idx_l0/1_active_minus1 + 1
404 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
405 unsigned int list_count;
406 H264Ref ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
407 * Reordered version of default_ref_list
408 * according to picture reordering in slice header */
410 const uint8_t *intra_pcm_ptr;
411 int16_t *dc_val_base;
413 uint8_t *bipred_scratchpad;
414 uint8_t *edge_emu_buffer;
415 uint8_t (*top_borders[2])[(16 * 3) * 2];
416 int bipred_scratchpad_allocated;
417 int edge_emu_buffer_allocated;
418 int top_borders_allocated[2];
421 * non zero coeff count cache.
422 * is 64 if not available.
424 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
427 * Motion vector cache.
429 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
430 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
431 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
432 uint8_t direct_cache[5 * 8];
434 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
436 ///< as a DCT coefficient is int32_t in high depth, we need to reserve twice the space.
437 DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2];
438 DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
439 ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either
440 ///< check that i is not too large or ensure that there is some unused stuff after mb
441 int16_t mb_padding[256 * 2];
443 uint8_t (*mvd_table[2])[2];
449 uint8_t cabac_state[1024];
456 typedef struct H264Context {
457 const AVClass *class;
458 AVCodecContext *avctx;
459 VideoDSPContext vdsp;
460 H264DSPContext h264dsp;
461 H264ChromaContext h264chroma;
462 H264QpelContext h264qpel;
464 H264Picture DPB[H264_MAX_PICTURE_COUNT];
465 H264Picture *cur_pic_ptr;
467 H264Picture last_pic_for_ec;
469 H264SliceContext *slice_ctx;
474 int pixel_shift; ///< 0 for 8-bit H.264, 1 for high-bit-depth H.264
476 /* coded dimensions -- 16 * mb w/h */
478 int chroma_x_shift, chroma_y_shift;
481 * Backup frame properties: needed, because they can be different
482 * between returned frame and last decoded frame.
486 enum AVPixelFormat backup_pix_fmt;
489 int coded_picture_number;
491 int context_initialized;
494 /* Set when slice threading is used and at least one slice uses deblocking
495 * mode 1 (i.e. across slice boundaries). Then we disable the loop filter
496 * during normal MB decoding and execute it serially at the end.
500 int8_t(*intra4x4_pred_mode);
503 uint8_t (*non_zero_count)[48];
505 #define LIST_NOT_USED -1 // FIXME rename?
506 #define PART_NOT_AVAILABLE -2
509 * block_offset[ 0..23] for frame macroblocks
510 * block_offset[24..47] for field macroblocks
512 int block_offset[2 * (16 * 3)];
514 uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
516 int b_stride; // FIXME use s->b4_stride
518 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
520 // interlacing specific flags
522 int picture_structure;
525 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
527 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
530 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
531 uint8_t *chroma_pred_mode_table;
532 uint8_t (*mvd_table[2])[2];
533 uint8_t *direct_table;
535 uint8_t zigzag_scan[16];
536 uint8_t zigzag_scan8x8[64];
537 uint8_t zigzag_scan8x8_cavlc[64];
538 uint8_t field_scan[16];
539 uint8_t field_scan8x8[64];
540 uint8_t field_scan8x8_cavlc[64];
541 uint8_t zigzag_scan_q0[16];
542 uint8_t zigzag_scan8x8_q0[64];
543 uint8_t zigzag_scan8x8_cavlc_q0[64];
544 uint8_t field_scan_q0[16];
545 uint8_t field_scan8x8_q0[64];
546 uint8_t field_scan8x8_cavlc_q0[64];
549 int mb_height, mb_width;
553 // =============================================================
554 // Things below are not used in the MB or more inner code
560 * Used to parse AVC variant of H.264
562 int is_avc; ///< this flag is != 0 if codec is avc1
563 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
565 int bit_depth_luma; ///< luma bit depth from sps to detect changes
566 int chroma_format_idc; ///< chroma format from sps to detect changes
570 uint16_t *slice_table_base;
575 * frame_num for frames or 2 * frame_num + 1 for field pics.
580 * max_frame_num or 2 * max_frame_num for field pics.
584 H264Ref default_ref[2];
585 H264Picture *short_ref[32];
586 H264Picture *long_ref[32];
587 H264Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
588 int last_pocs[MAX_DELAYED_PIC_COUNT];
589 H264Picture *next_output_pic;
590 int next_outputed_poc;
593 * memory management control operations buffer.
595 MMCO mmco[MAX_MMCO_COUNT];
599 int long_ref_count; ///< number of actual long term references
600 int short_ref_count; ///< number of actual short term references
603 * @name Members for slice based multithreading
607 * current slice number, used to initialize slice_num of each thread/context
612 * Max number of threads / contexts.
613 * This is equal to AVCodecContext.thread_count unless
614 * multithreaded decoding is impossible, in which case it is
620 * 1 if the single thread fallback warning has already been
621 * displayed, 0 otherwise.
623 int single_decode_warning;
628 * Complement sei_pic_struct
629 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
630 * However, soft telecined frames may have these values.
631 * This is used in an attempt to flag soft telecine progressive.
633 int prev_interlaced_frame;
636 * Are the SEI recovery points looking valid.
638 int valid_recovery_point;
641 * recovery_frame is the frame_num at which the next frame should
642 * be fully constructed.
644 * Set to -1 when not expecting a recovery point.
649 * We have seen an IDR, so all the following frames in coded order are correctly
652 #define FRAME_RECOVERED_IDR (1 << 0)
654 * Sufficient number of frames have been decoded since a SEI recovery point,
655 * so all the following frames in presentation order are correct.
657 #define FRAME_RECOVERED_SEI (1 << 1)
659 int frame_recovered; ///< Initial frame has been completely recovered
661 int has_recovery_point;
665 /* for frame threading, this is set to 1
666 * after finish_setup() has been called, so we cannot modify
667 * some context properties (which are supposed to stay constant between
671 int cur_chroma_format_idc;
672 int cur_bit_depth_luma;
673 int16_t slice_row[MAX_SLICES]; ///< to detect when MAX_SLICES is too low
679 AVBufferPool *qscale_table_pool;
680 AVBufferPool *mb_type_pool;
681 AVBufferPool *motion_val_pool;
682 AVBufferPool *ref_index_pool;
683 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
686 extern const uint16_t ff_h264_mb_sizes[4];
689 * Uninit H264 param sets structure.
692 void ff_h264_ps_uninit(H264ParamSets *ps);
697 int ff_h264_decode_seq_parameter_set(GetBitContext *gb, AVCodecContext *avctx,
698 H264ParamSets *ps, int ignore_truncation);
703 int ff_h264_decode_picture_parameter_set(GetBitContext *gb, AVCodecContext *avctx,
704 H264ParamSets *ps, int bit_length);
707 * Reconstruct bitstream slice_type.
709 int ff_h264_get_slice_type(const H264SliceContext *sl);
715 int ff_h264_alloc_tables(H264Context *h);
717 int ff_h264_decode_ref_pic_list_reordering(H264Context *h, H264SliceContext *sl);
718 void ff_h264_remove_all_refs(H264Context *h);
721 * Execute the reference picture marking (memory management control operations).
723 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
725 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb,
728 int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
730 void ff_h264_hl_decode_mb(const H264Context *h, H264SliceContext *sl);
731 int ff_h264_decode_init(AVCodecContext *avctx);
732 void ff_h264_decode_init_vlc(void);
735 * Decode a macroblock
736 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
738 int ff_h264_decode_mb_cavlc(const H264Context *h, H264SliceContext *sl);
741 * Decode a CABAC coded macroblock
742 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
744 int ff_h264_decode_mb_cabac(const H264Context *h, H264SliceContext *sl);
746 void ff_h264_init_cabac_states(const H264Context *h, H264SliceContext *sl);
748 void ff_h264_direct_dist_scale_factor(const H264Context *const h, H264SliceContext *sl);
749 void ff_h264_direct_ref_list_init(const H264Context *const h, H264SliceContext *sl);
750 void ff_h264_pred_direct_motion(const H264Context *const h, H264SliceContext *sl,
753 void ff_h264_filter_mb_fast(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
754 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
755 unsigned int linesize, unsigned int uvlinesize);
756 void ff_h264_filter_mb(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
757 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
758 unsigned int linesize, unsigned int uvlinesize);
770 /* Scan8 organization:
787 * DY/DU/DV are for luma/chroma DC.
790 #define LUMA_DC_BLOCK_INDEX 48
791 #define CHROMA_DC_BLOCK_INDEX 49
793 // This table must be here because scan8[constant] must be known at compiletime
794 static const uint8_t scan8[16 * 3 + 3] = {
795 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
796 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
797 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
798 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
799 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
800 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
801 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
802 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
803 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
804 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
805 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
806 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
807 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
810 static av_always_inline uint32_t pack16to32(unsigned a, unsigned b)
813 return (b & 0xFFFF) + (a << 16);
815 return (a & 0xFFFF) + (b << 16);
819 static av_always_inline uint16_t pack8to16(unsigned a, unsigned b)
822 return (b & 0xFF) + (a << 8);
824 return (a & 0xFF) + (b << 8);
831 static av_always_inline int get_chroma_qp(const H264Context *h, int t, int qscale)
833 return h->ps.pps->chroma_qp_table[t][qscale];
837 * Get the predicted intra4x4 prediction mode.
839 static av_always_inline int pred_intra_mode(const H264Context *h,
840 H264SliceContext *sl, int n)
842 const int index8 = scan8[n];
843 const int left = sl->intra4x4_pred_mode_cache[index8 - 1];
844 const int top = sl->intra4x4_pred_mode_cache[index8 - 8];
845 const int min = FFMIN(left, top);
847 ff_tlog(h->avctx, "mode:%d %d min:%d\n", left, top, min);
855 static av_always_inline void write_back_intra_pred_mode(const H264Context *h,
856 H264SliceContext *sl)
858 int8_t *i4x4 = sl->intra4x4_pred_mode + h->mb2br_xy[sl->mb_xy];
859 int8_t *i4x4_cache = sl->intra4x4_pred_mode_cache;
861 AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
862 i4x4[4] = i4x4_cache[7 + 8 * 3];
863 i4x4[5] = i4x4_cache[7 + 8 * 2];
864 i4x4[6] = i4x4_cache[7 + 8 * 1];
867 static av_always_inline void write_back_non_zero_count(const H264Context *h,
868 H264SliceContext *sl)
870 const int mb_xy = sl->mb_xy;
871 uint8_t *nnz = h->non_zero_count[mb_xy];
872 uint8_t *nnz_cache = sl->non_zero_count_cache;
874 AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
875 AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
876 AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
877 AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
878 AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
879 AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
880 AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
881 AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
883 if (!h->chroma_y_shift) {
884 AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
885 AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
886 AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
887 AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
891 static av_always_inline void write_back_motion_list(const H264Context *h,
892 H264SliceContext *sl,
895 int mb_type, int list)
897 int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
898 int16_t(*mv_src)[2] = &sl->mv_cache[list][scan8[0]];
899 AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
900 AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
901 AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
902 AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
904 uint8_t (*mvd_dst)[2] = &sl->mvd_table[list][FMO ? 8 * sl->mb_xy
905 : h->mb2br_xy[sl->mb_xy]];
906 uint8_t(*mvd_src)[2] = &sl->mvd_cache[list][scan8[0]];
907 if (IS_SKIP(mb_type)) {
910 AV_COPY64(mvd_dst, mvd_src + 8 * 3);
911 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
912 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
913 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
918 int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
919 int8_t *ref_cache = sl->ref_cache[list];
920 ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
921 ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
922 ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
923 ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
927 static av_always_inline void write_back_motion(const H264Context *h,
928 H264SliceContext *sl,
931 const int b_stride = h->b_stride;
932 const int b_xy = 4 * sl->mb_x + 4 * sl->mb_y * h->b_stride; // try mb2b(8)_xy
933 const int b8_xy = 4 * sl->mb_xy;
935 if (USES_LIST(mb_type, 0)) {
936 write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 0);
938 fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
939 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
941 if (USES_LIST(mb_type, 1))
942 write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 1);
944 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) {
945 if (IS_8X8(mb_type)) {
946 uint8_t *direct_table = &h->direct_table[4 * sl->mb_xy];
947 direct_table[1] = sl->sub_mb_type[1] >> 1;
948 direct_table[2] = sl->sub_mb_type[2] >> 1;
949 direct_table[3] = sl->sub_mb_type[3] >> 1;
954 static av_always_inline int get_dct8x8_allowed(const H264Context *h, H264SliceContext *sl)
956 if (h->ps.sps->direct_8x8_inference_flag)
957 return !(AV_RN64A(sl->sub_mb_type) &
958 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
959 0x0001000100010001ULL));
961 return !(AV_RN64A(sl->sub_mb_type) &
962 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
963 0x0001000100010001ULL));
966 static inline int find_start_code(const uint8_t *buf, int buf_size,
967 int buf_index, int next_avc)
971 buf_index = avpriv_find_start_code(buf + buf_index, buf + next_avc + 1, &state) - buf - 1;
973 return FFMIN(buf_index, buf_size);
976 int ff_h264_field_end(H264Context *h, H264SliceContext *sl, int in_setup);
978 int ff_h264_ref_picture(H264Context *h, H264Picture *dst, H264Picture *src);
979 void ff_h264_unref_picture(H264Context *h, H264Picture *pic);
981 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl);
983 void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl, int y, int height);
985 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl);
986 #define SLICE_SINGLETHREAD 1
987 #define SLICE_SKIPED 2
989 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count);
990 int ff_h264_update_thread_context(AVCodecContext *dst,
991 const AVCodecContext *src);
993 void ff_h264_flush_change(H264Context *h);
995 void ff_h264_free_tables(H264Context *h);
997 void ff_h264_set_erpic(ERPicture *dst, H264Picture *src);
999 #endif /* AVCODEC_H264_H */