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 / 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 32
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 /* Compiling in interlaced support reduces the speed
61 * of progressive decoding by about 2%. */
62 #define ALLOW_INTERLACE
67 * The maximum number of slices supported by the decoder.
68 * must be a power of 2
72 #ifdef ALLOW_INTERLACE
73 #define MB_MBAFF(h) h->mb_mbaff
74 #define MB_FIELD(h) h->mb_field_decoding_flag
75 #define FRAME_MBAFF(h) h->mb_aff_frame
76 #define FIELD_PICTURE(h) (h->picture_structure != PICT_FRAME)
84 #define FRAME_MBAFF(h) 0
85 #define FIELD_PICTURE(h) 0
87 #define IS_INTERLACED(mb_type) 0
93 #define FIELD_OR_MBAFF_PICTURE(h) (FRAME_MBAFF(h) || FIELD_PICTURE(h))
96 #define CABAC(h) h->ps.pps->cabac
99 #define CHROMA422(h) (h->ps.sps->chroma_format_idc == 2)
100 #define CHROMA444(h) (h->ps.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 + 2 * 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,
131 * Sequence parameter set
137 int chroma_format_idc;
138 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
139 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
140 int poc_type; ///< pic_order_cnt_type
141 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
142 int delta_pic_order_always_zero_flag;
143 int offset_for_non_ref_pic;
144 int offset_for_top_to_bottom_field;
145 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
146 int ref_frame_count; ///< num_ref_frames
147 int gaps_in_frame_num_allowed_flag;
148 int mb_width; ///< pic_width_in_mbs_minus1 + 1
149 int mb_height; ///< pic_height_in_map_units_minus1 + 1
150 int frame_mbs_only_flag;
151 int mb_aff; ///< mb_adaptive_frame_field_flag
152 int direct_8x8_inference_flag;
153 int crop; ///< frame_cropping_flag
155 /* those 4 are already in luma samples */
156 unsigned int crop_left; ///< frame_cropping_rect_left_offset
157 unsigned int crop_right; ///< frame_cropping_rect_right_offset
158 unsigned int crop_top; ///< frame_cropping_rect_top_offset
159 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
160 int vui_parameters_present_flag;
162 int video_signal_type_present_flag;
164 int colour_description_present_flag;
165 enum AVColorPrimaries color_primaries;
166 enum AVColorTransferCharacteristic color_trc;
167 enum AVColorSpace colorspace;
168 int timing_info_present_flag;
169 uint32_t num_units_in_tick;
171 int fixed_frame_rate_flag;
172 short offset_for_ref_frame[256]; // FIXME dyn aloc?
173 int bitstream_restriction_flag;
174 int num_reorder_frames;
175 int scaling_matrix_present;
176 uint8_t scaling_matrix4[6][16];
177 uint8_t scaling_matrix8[6][64];
178 int nal_hrd_parameters_present_flag;
179 int vcl_hrd_parameters_present_flag;
180 int pic_struct_present_flag;
181 int time_offset_length;
182 int cpb_cnt; ///< See H.264 E.1.2
183 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
184 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
185 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
186 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
187 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
188 int residual_color_transform_flag; ///< residual_colour_transform_flag
189 int constraint_set_flags; ///< constraint_set[0-3]_flag
193 * Picture parameter set
197 int cabac; ///< entropy_coding_mode_flag
198 int pic_order_present; ///< pic_order_present_flag
199 int slice_group_count; ///< num_slice_groups_minus1 + 1
200 int mb_slice_group_map_type;
201 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
202 int weighted_pred; ///< weighted_pred_flag
203 int weighted_bipred_idc;
204 int init_qp; ///< pic_init_qp_minus26 + 26
205 int init_qs; ///< pic_init_qs_minus26 + 26
206 int chroma_qp_index_offset[2];
207 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
208 int constrained_intra_pred; ///< constrained_intra_pred_flag
209 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
210 int transform_8x8_mode; ///< transform_8x8_mode_flag
211 uint8_t scaling_matrix4[6][16];
212 uint8_t scaling_matrix8[6][64];
213 uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
216 uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16];
217 uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
218 uint32_t(*dequant4_coeff[6])[16];
219 uint32_t(*dequant8_coeff[6])[64];
222 typedef struct H264ParamSets {
223 AVBufferRef *sps_list[MAX_SPS_COUNT];
224 AVBufferRef *pps_list[MAX_PPS_COUNT];
226 /* currently active parameters sets */
228 // FIXME this should properly be const
233 * Memory management control operation opcode.
235 typedef enum MMCOOpcode {
246 * Memory management control operation.
248 typedef struct MMCO {
250 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
251 int long_arg; ///< index, pic_num, or num long refs depending on opcode
254 typedef struct H264Picture {
258 AVBufferRef *qscale_table_buf;
259 int8_t *qscale_table;
261 AVBufferRef *motion_val_buf[2];
262 int16_t (*motion_val[2])[2];
264 AVBufferRef *mb_type_buf;
267 AVBufferRef *hwaccel_priv_buf;
268 void *hwaccel_picture_private; ///< hardware accelerator private data
270 AVBufferRef *ref_index_buf[2];
271 int8_t *ref_index[2];
273 int field_poc[2]; ///< top/bottom POC
274 int poc; ///< frame POC
275 int frame_num; ///< frame_num (raw frame_num from slice header)
276 int mmco_reset; /**< MMCO_RESET set this 1. Reordering code must
277 not mix pictures before and after MMCO_RESET. */
278 int pic_id; /**< pic_num (short -> no wrap version of pic_num,
279 pic_num & max_pic_num; long -> long_pic_num) */
280 int long_ref; ///< 1->long term reference 0->short term reference
281 int ref_poc[2][2][32]; ///< POCs of the frames used as reference (FIXME need per slice)
282 int ref_count[2][2]; ///< number of entries in ref_poc (FIXME need per slice)
283 int mbaff; ///< 1 -> MBAFF frame 0-> not MBAFF
284 int field_picture; ///< whether or not picture was encoded in separate fields
287 int recovered; ///< picture at IDR or recovery point + recovery count
290 typedef struct H264Ref {
301 typedef struct H264SliceContext {
302 struct H264Context *h264;
308 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
309 int slice_type_fixed;
312 int chroma_qp[2]; // QPc
313 int qp_thresh; ///< QP threshold to skip loopfilter
314 int last_qscale_diff;
317 int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
318 int slice_alpha_c0_offset;
319 int slice_beta_offset;
321 H264PredWeightTable pwt;
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 linesize, uvlinesize;
351 ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
352 ptrdiff_t mb_uvlinesize;
358 // index of the first MB of the next slice
363 int mb_field_decoding_flag;
364 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
366 int redundant_pic_count;
369 * number of neighbors (top and/or left) that used 8x8 dct
371 int neighbor_transform_size;
373 int direct_spatial_mv_pred;
381 int dist_scale_factor[32];
382 int dist_scale_factor_field[2][32];
383 int map_col_to_list0[2][16 + 32];
384 int map_col_to_list0_field[2][2][16 + 32];
387 * num_ref_idx_l0/1_active_minus1 + 1
389 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
390 unsigned int list_count;
391 H264Ref ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
392 * Reordered version of default_ref_list
393 * according to picture reordering in slice header */
397 } ref_modifications[2][32];
398 int nb_ref_modifications[2];
400 const uint8_t *intra_pcm_ptr;
401 int16_t *dc_val_base;
403 uint8_t *bipred_scratchpad;
404 uint8_t *edge_emu_buffer;
405 uint8_t (*top_borders[2])[(16 * 3) * 2];
406 int bipred_scratchpad_allocated;
407 int edge_emu_buffer_allocated;
408 int top_borders_allocated[2];
411 * non zero coeff count cache.
412 * is 64 if not available.
414 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
417 * Motion vector cache.
419 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
420 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
421 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
422 uint8_t direct_cache[5 * 8];
424 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
426 ///< as a DCT coefficient is int32_t in high depth, we need to reserve twice the space.
427 DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2];
428 DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
429 ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either
430 ///< check that i is not too large or ensure that there is some unused stuff after mb
431 int16_t mb_padding[256 * 2];
433 uint8_t (*mvd_table[2])[2];
439 uint8_t cabac_state[1024];
446 typedef struct H264Context {
447 const AVClass *class;
448 AVCodecContext *avctx;
449 VideoDSPContext vdsp;
450 H264DSPContext h264dsp;
451 H264ChromaContext h264chroma;
452 H264QpelContext h264qpel;
454 H264Picture DPB[H264_MAX_PICTURE_COUNT];
455 H264Picture *cur_pic_ptr;
458 H264SliceContext *slice_ctx;
463 int pixel_shift; ///< 0 for 8-bit H.264, 1 for high-bit-depth H.264
465 /* coded dimensions -- 16 * mb w/h */
467 int chroma_x_shift, chroma_y_shift;
470 int coded_picture_number;
472 int context_initialized;
475 /* Set when slice threading is used and at least one slice uses deblocking
476 * mode 1 (i.e. across slice boundaries). Then we disable the loop filter
477 * during normal MB decoding and execute it serially at the end.
481 int8_t(*intra4x4_pred_mode);
484 uint8_t (*non_zero_count)[48];
486 #define LIST_NOT_USED -1 // FIXME rename?
487 #define PART_NOT_AVAILABLE -2
490 * block_offset[ 0..23] for frame macroblocks
491 * block_offset[24..47] for field macroblocks
493 int block_offset[2 * (16 * 3)];
495 uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
497 int b_stride; // FIXME use s->b4_stride
499 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
501 // interlacing specific flags
503 int picture_structure;
506 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
508 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
511 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
512 uint8_t *chroma_pred_mode_table;
513 uint8_t (*mvd_table[2])[2];
514 uint8_t *direct_table;
516 uint8_t zigzag_scan[16];
517 uint8_t zigzag_scan8x8[64];
518 uint8_t zigzag_scan8x8_cavlc[64];
519 uint8_t field_scan[16];
520 uint8_t field_scan8x8[64];
521 uint8_t field_scan8x8_cavlc[64];
522 const uint8_t *zigzag_scan_q0;
523 const uint8_t *zigzag_scan8x8_q0;
524 const uint8_t *zigzag_scan8x8_cavlc_q0;
525 const uint8_t *field_scan_q0;
526 const uint8_t *field_scan8x8_q0;
527 const uint8_t *field_scan8x8_cavlc_q0;
530 int mb_height, mb_width;
534 // =============================================================
535 // Things below are not used in the MB or more inner code
541 * Used to parse AVC variant of H.264
543 int is_avc; ///< this flag is != 0 if codec is avc1
544 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
546 int bit_depth_luma; ///< luma bit depth from sps to detect changes
547 int chroma_format_idc; ///< chroma format from sps to detect changes
551 uint16_t *slice_table_base;
556 * frame_num for frames or 2 * frame_num + 1 for field pics.
561 * max_frame_num or 2 * max_frame_num for field pics.
565 H264Picture *short_ref[32];
566 H264Picture *long_ref[32];
567 H264Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
568 int last_pocs[MAX_DELAYED_PIC_COUNT];
569 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
583 * @name Members for slice based multithreading
587 * current slice number, used to initialize slice_num of each thread/context
594 * Complement sei_pic_struct
595 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
596 * However, soft telecined frames may have these values.
597 * This is used in an attempt to flag soft telecine progressive.
599 int prev_interlaced_frame;
602 * recovery_frame is the frame_num at which the next frame should
603 * be fully constructed.
605 * Set to -1 when not expecting a recovery point.
610 * We have seen an IDR, so all the following frames in coded order are correctly
613 #define FRAME_RECOVERED_IDR (1 << 0)
615 * Sufficient number of frames have been decoded since a SEI recovery point,
616 * so all the following frames in presentation order are correct.
618 #define FRAME_RECOVERED_SEI (1 << 1)
620 int frame_recovered; ///< Initial frame has been completely recovered
622 /* for frame threading, this is set to 1
623 * after finish_setup() has been called, so we cannot modify
624 * some context properties (which are supposed to stay constant between
632 AVBufferPool *qscale_table_pool;
633 AVBufferPool *mb_type_pool;
634 AVBufferPool *motion_val_pool;
635 AVBufferPool *ref_index_pool;
636 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
639 extern const uint16_t ff_h264_mb_sizes[4];
644 int ff_h264_decode_seq_parameter_set(GetBitContext *gb, AVCodecContext *avctx,
650 int ff_h264_decode_picture_parameter_set(GetBitContext *gb, AVCodecContext *avctx,
651 H264ParamSets *ps, int bit_length);
654 * Reconstruct bitstream slice_type.
656 int ff_h264_get_slice_type(const H264SliceContext *sl);
662 int ff_h264_alloc_tables(H264Context *h);
664 int ff_h264_decode_ref_pic_list_reordering(const H264Context *h, H264SliceContext *sl);
665 int ff_h264_build_ref_list(const H264Context *h, H264SliceContext *sl);
666 void ff_h264_remove_all_refs(H264Context *h);
669 * Execute the reference picture marking (memory management control operations).
671 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
673 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb,
676 int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
678 void ff_h264_hl_decode_mb(const H264Context *h, H264SliceContext *sl);
679 int ff_h264_decode_init(AVCodecContext *avctx);
680 void ff_h264_decode_init_vlc(void);
683 * Decode a macroblock
684 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
686 int ff_h264_decode_mb_cavlc(const H264Context *h, H264SliceContext *sl);
689 * Decode a CABAC coded macroblock
690 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
692 int ff_h264_decode_mb_cabac(const H264Context *h, H264SliceContext *sl);
694 void ff_h264_init_cabac_states(const H264Context *h, H264SliceContext *sl);
696 void ff_h264_init_dequant_tables(H264Context *h);
698 void ff_h264_direct_dist_scale_factor(const H264Context *const h, H264SliceContext *sl);
699 void ff_h264_direct_ref_list_init(const H264Context *const h, H264SliceContext *sl);
700 void ff_h264_pred_direct_motion(const H264Context *const h, H264SliceContext *sl,
703 void ff_h264_filter_mb_fast(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
704 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
705 unsigned int linesize, unsigned int uvlinesize);
706 void ff_h264_filter_mb(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
707 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
708 unsigned int linesize, unsigned int uvlinesize);
720 /* Scan8 organization:
737 * DY/DU/DV are for luma/chroma DC.
740 #define LUMA_DC_BLOCK_INDEX 48
741 #define CHROMA_DC_BLOCK_INDEX 49
743 // This table must be here because scan8[constant] must be known at compiletime
744 static const uint8_t scan8[16 * 3 + 3] = {
745 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
746 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
747 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
748 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
749 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
750 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
751 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
752 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
753 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
754 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
755 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
756 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
757 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
760 static av_always_inline uint32_t pack16to32(int a, int b)
763 return (b & 0xFFFF) + (a << 16);
765 return (a & 0xFFFF) + (b << 16);
769 static av_always_inline uint16_t pack8to16(int a, int b)
772 return (b & 0xFF) + (a << 8);
774 return (a & 0xFF) + (b << 8);
781 static av_always_inline int get_chroma_qp(const H264Context *h, int t, int qscale)
783 return h->ps.pps->chroma_qp_table[t][qscale];
787 * Get the predicted intra4x4 prediction mode.
789 static av_always_inline int pred_intra_mode(const H264Context *h,
790 H264SliceContext *sl, int n)
792 const int index8 = scan8[n];
793 const int left = sl->intra4x4_pred_mode_cache[index8 - 1];
794 const int top = sl->intra4x4_pred_mode_cache[index8 - 8];
795 const int min = FFMIN(left, top);
797 ff_tlog(h->avctx, "mode:%d %d min:%d\n", left, top, min);
805 static av_always_inline void write_back_intra_pred_mode(const H264Context *h,
806 H264SliceContext *sl)
808 int8_t *i4x4 = sl->intra4x4_pred_mode + h->mb2br_xy[sl->mb_xy];
809 int8_t *i4x4_cache = sl->intra4x4_pred_mode_cache;
811 AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
812 i4x4[4] = i4x4_cache[7 + 8 * 3];
813 i4x4[5] = i4x4_cache[7 + 8 * 2];
814 i4x4[6] = i4x4_cache[7 + 8 * 1];
817 static av_always_inline void write_back_non_zero_count(const H264Context *h,
818 H264SliceContext *sl)
820 const int mb_xy = sl->mb_xy;
821 uint8_t *nnz = h->non_zero_count[mb_xy];
822 uint8_t *nnz_cache = sl->non_zero_count_cache;
824 AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
825 AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
826 AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
827 AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
828 AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
829 AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
830 AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
831 AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
833 if (!h->chroma_y_shift) {
834 AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
835 AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
836 AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
837 AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
841 static av_always_inline void write_back_motion_list(const H264Context *h,
842 H264SliceContext *sl,
845 int mb_type, int list)
847 int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
848 int16_t(*mv_src)[2] = &sl->mv_cache[list][scan8[0]];
849 AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
850 AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
851 AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
852 AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
854 uint8_t (*mvd_dst)[2] = &sl->mvd_table[list][FMO ? 8 * sl->mb_xy
855 : h->mb2br_xy[sl->mb_xy]];
856 uint8_t(*mvd_src)[2] = &sl->mvd_cache[list][scan8[0]];
857 if (IS_SKIP(mb_type)) {
860 AV_COPY64(mvd_dst, mvd_src + 8 * 3);
861 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
862 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
863 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
868 int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
869 int8_t *ref_cache = sl->ref_cache[list];
870 ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
871 ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
872 ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
873 ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
877 static av_always_inline void write_back_motion(const H264Context *h,
878 H264SliceContext *sl,
881 const int b_stride = h->b_stride;
882 const int b_xy = 4 * sl->mb_x + 4 * sl->mb_y * h->b_stride; // try mb2b(8)_xy
883 const int b8_xy = 4 * sl->mb_xy;
885 if (USES_LIST(mb_type, 0)) {
886 write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 0);
888 fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
889 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
891 if (USES_LIST(mb_type, 1))
892 write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 1);
894 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) {
895 if (IS_8X8(mb_type)) {
896 uint8_t *direct_table = &h->direct_table[4 * sl->mb_xy];
897 direct_table[1] = sl->sub_mb_type[1] >> 1;
898 direct_table[2] = sl->sub_mb_type[2] >> 1;
899 direct_table[3] = sl->sub_mb_type[3] >> 1;
904 static av_always_inline int get_dct8x8_allowed(const H264Context *h, H264SliceContext *sl)
906 if (h->ps.sps->direct_8x8_inference_flag)
907 return !(AV_RN64A(sl->sub_mb_type) &
908 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
909 0x0001000100010001ULL));
911 return !(AV_RN64A(sl->sub_mb_type) &
912 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
913 0x0001000100010001ULL));
916 int ff_h264_field_end(H264Context *h, H264SliceContext *sl, int in_setup);
918 int ff_h264_ref_picture(H264Context *h, H264Picture *dst, H264Picture *src);
919 void ff_h264_unref_picture(H264Context *h, H264Picture *pic);
921 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl);
923 void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl, int y, int height);
925 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl);
926 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count);
927 int ff_h264_update_thread_context(AVCodecContext *dst,
928 const AVCodecContext *src);
930 void ff_h264_flush_change(H264Context *h);
932 void ff_h264_free_tables(H264Context *h);
934 #endif /* AVCODEC_H264_H */