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/buffer.h"
32 #include "libavutil/intreadwrite.h"
33 #include "libavutil/thread.h"
35 #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
52 #define H264_MAX_THREADS 16
54 #define MAX_SPS_COUNT 32
55 #define MAX_PPS_COUNT 256
57 #define MAX_MMCO_COUNT 66
59 #define MAX_DELAYED_PIC_COUNT 16
61 /* Compiling in interlaced support reduces the speed
62 * of progressive decoding by about 2%. */
63 #define ALLOW_INTERLACE
68 * The maximum number of slices supported by the decoder.
69 * must be a power of 2
73 #ifdef ALLOW_INTERLACE
74 #define MB_MBAFF(h) h->mb_mbaff
75 #define MB_FIELD(h) h->mb_field_decoding_flag
76 #define FRAME_MBAFF(h) h->mb_aff_frame
77 #define FIELD_PICTURE(h) (h->picture_structure != PICT_FRAME)
85 #define FRAME_MBAFF(h) 0
86 #define FIELD_PICTURE(h) 0
88 #define IS_INTERLACED(mb_type) 0
94 #define FIELD_OR_MBAFF_PICTURE(h) (FRAME_MBAFF(h) || FIELD_PICTURE(h))
97 #define CABAC(h) h->ps.pps->cabac
100 #define CHROMA422(h) (h->ps.sps->chroma_format_idc == 2)
101 #define CHROMA444(h) (h->ps.sps->chroma_format_idc == 3)
103 #define EXTENDED_SAR 255
105 #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit
106 #define MB_TYPE_8x8DCT 0x01000000
107 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
108 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
110 #define QP_MAX_NUM (51 + 2 * 6) // The maximum supported qp
123 NAL_END_SEQUENCE = 10,
125 NAL_FILLER_DATA = 12,
127 NAL_AUXILIARY_SLICE = 19,
128 NAL_FF_IGNORE = 0xff0f001,
132 * Sequence parameter set
138 int chroma_format_idc;
139 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
140 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
141 int poc_type; ///< pic_order_cnt_type
142 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
143 int delta_pic_order_always_zero_flag;
144 int offset_for_non_ref_pic;
145 int offset_for_top_to_bottom_field;
146 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
147 int ref_frame_count; ///< num_ref_frames
148 int gaps_in_frame_num_allowed_flag;
149 int mb_width; ///< pic_width_in_mbs_minus1 + 1
150 int mb_height; ///< pic_height_in_map_units_minus1 + 1
151 int frame_mbs_only_flag;
152 int mb_aff; ///< mb_adaptive_frame_field_flag
153 int direct_8x8_inference_flag;
154 int crop; ///< frame_cropping_flag
156 /* those 4 are already in luma samples */
157 unsigned int crop_left; ///< frame_cropping_rect_left_offset
158 unsigned int crop_right; ///< frame_cropping_rect_right_offset
159 unsigned int crop_top; ///< frame_cropping_rect_top_offset
160 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
161 int vui_parameters_present_flag;
163 int video_signal_type_present_flag;
165 int colour_description_present_flag;
166 enum AVColorPrimaries color_primaries;
167 enum AVColorTransferCharacteristic color_trc;
168 enum AVColorSpace colorspace;
169 int timing_info_present_flag;
170 uint32_t num_units_in_tick;
172 int fixed_frame_rate_flag;
173 short offset_for_ref_frame[256]; // FIXME dyn aloc?
174 int bitstream_restriction_flag;
175 int num_reorder_frames;
176 int scaling_matrix_present;
177 uint8_t scaling_matrix4[6][16];
178 uint8_t scaling_matrix8[6][64];
179 int nal_hrd_parameters_present_flag;
180 int vcl_hrd_parameters_present_flag;
181 int pic_struct_present_flag;
182 int time_offset_length;
183 int cpb_cnt; ///< See H.264 E.1.2
184 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
185 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
186 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
187 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
188 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
189 int residual_color_transform_flag; ///< residual_colour_transform_flag
190 int constraint_set_flags; ///< constraint_set[0-3]_flag
194 * Picture parameter set
198 int cabac; ///< entropy_coding_mode_flag
199 int pic_order_present; ///< pic_order_present_flag
200 int slice_group_count; ///< num_slice_groups_minus1 + 1
201 int mb_slice_group_map_type;
202 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
203 int weighted_pred; ///< weighted_pred_flag
204 int weighted_bipred_idc;
205 int init_qp; ///< pic_init_qp_minus26 + 26
206 int init_qs; ///< pic_init_qs_minus26 + 26
207 int chroma_qp_index_offset[2];
208 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
209 int constrained_intra_pred; ///< constrained_intra_pred_flag
210 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
211 int transform_8x8_mode; ///< transform_8x8_mode_flag
212 uint8_t scaling_matrix4[6][16];
213 uint8_t scaling_matrix8[6][64];
214 uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
217 uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16];
218 uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
219 uint32_t(*dequant4_coeff[6])[16];
220 uint32_t(*dequant8_coeff[6])[64];
223 typedef struct H264ParamSets {
224 AVBufferRef *sps_list[MAX_SPS_COUNT];
225 AVBufferRef *pps_list[MAX_PPS_COUNT];
227 /* currently active parameters sets */
229 // FIXME this should properly be const
234 * Memory management control operation opcode.
236 typedef enum MMCOOpcode {
247 * Memory management control operation.
249 typedef struct MMCO {
251 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
252 int long_arg; ///< index, pic_num, or num long refs depending on opcode
255 typedef struct H264Picture {
259 AVBufferRef *qscale_table_buf;
260 int8_t *qscale_table;
262 AVBufferRef *motion_val_buf[2];
263 int16_t (*motion_val[2])[2];
265 AVBufferRef *mb_type_buf;
268 AVBufferRef *hwaccel_priv_buf;
269 void *hwaccel_picture_private; ///< hardware accelerator private data
271 AVBufferRef *ref_index_buf[2];
272 int8_t *ref_index[2];
274 int field_poc[2]; ///< top/bottom POC
275 int poc; ///< frame POC
276 int frame_num; ///< frame_num (raw frame_num from slice header)
277 int mmco_reset; /**< MMCO_RESET set this 1. Reordering code must
278 not mix pictures before and after MMCO_RESET. */
279 int pic_id; /**< pic_num (short -> no wrap version of pic_num,
280 pic_num & max_pic_num; long -> long_pic_num) */
281 int long_ref; ///< 1->long term reference 0->short term reference
282 int ref_poc[2][2][32]; ///< POCs of the frames used as reference (FIXME need per slice)
283 int ref_count[2][2]; ///< number of entries in ref_poc (FIXME need per slice)
284 int mbaff; ///< 1 -> MBAFF frame 0-> not MBAFF
285 int field_picture; ///< whether or not picture was encoded in separate fields
288 int recovered; ///< picture at IDR or recovery point + recovery count
291 typedef struct H264Ref {
302 typedef struct H264SliceContext {
303 struct H264Context *h264;
309 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
310 int slice_type_fixed;
313 int chroma_qp[2]; // QPc
314 int qp_thresh; ///< QP threshold to skip loopfilter
315 int last_qscale_diff;
318 int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
319 int slice_alpha_c0_offset;
320 int slice_beta_offset;
322 H264PredWeightTable pwt;
327 int chroma_pred_mode;
328 int intra16x16_pred_mode;
330 int8_t intra4x4_pred_mode_cache[5 * 8];
331 int8_t(*intra4x4_pred_mode);
336 int left_mb_xy[LEFT_MBS];
341 int left_type[LEFT_MBS];
343 const uint8_t *left_block;
344 int topleft_partition;
346 unsigned int topleft_samples_available;
347 unsigned int top_samples_available;
348 unsigned int topright_samples_available;
349 unsigned int left_samples_available;
351 ptrdiff_t linesize, uvlinesize;
352 ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
353 ptrdiff_t mb_uvlinesize;
359 // index of the first MB of the next slice
364 int mb_field_decoding_flag;
365 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
367 int redundant_pic_count;
370 * number of neighbors (top and/or left) that used 8x8 dct
372 int neighbor_transform_size;
374 int direct_spatial_mv_pred;
382 int dist_scale_factor[32];
383 int dist_scale_factor_field[2][32];
384 int map_col_to_list0[2][16 + 32];
385 int map_col_to_list0_field[2][2][16 + 32];
388 * num_ref_idx_l0/1_active_minus1 + 1
390 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
391 unsigned int list_count;
392 H264Ref ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
393 * Reordered version of default_ref_list
394 * according to picture reordering in slice header */
395 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
397 const uint8_t *intra_pcm_ptr;
398 int16_t *dc_val_base;
400 uint8_t *bipred_scratchpad;
401 uint8_t *edge_emu_buffer;
402 uint8_t (*top_borders[2])[(16 * 3) * 2];
403 int bipred_scratchpad_allocated;
404 int edge_emu_buffer_allocated;
405 int top_borders_allocated[2];
408 * non zero coeff count cache.
409 * is 64 if not available.
411 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
414 * Motion vector cache.
416 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
417 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
418 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
419 uint8_t direct_cache[5 * 8];
421 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
423 ///< as a dct coeffecient is int32_t in high depth, we need to reserve twice the space.
424 DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2];
425 DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
426 ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either
427 ///< check that i is not too large or ensure that there is some unused stuff after mb
428 int16_t mb_padding[256 * 2];
430 uint8_t (*mvd_table[2])[2];
436 uint8_t cabac_state[1024];
439 // rbsp buffer used for this slice
440 uint8_t *rbsp_buffer;
441 unsigned int rbsp_buffer_size;
447 typedef struct H264Context {
448 const AVClass *class;
449 AVCodecContext *avctx;
450 VideoDSPContext vdsp;
451 H264DSPContext h264dsp;
452 H264ChromaContext h264chroma;
453 H264QpelContext h264qpel;
456 H264Picture DPB[H264_MAX_PICTURE_COUNT];
457 H264Picture *cur_pic_ptr;
460 H264SliceContext *slice_ctx;
465 int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
467 /* coded dimensions -- 16 * mb w/h */
469 int chroma_x_shift, chroma_y_shift;
472 int coded_picture_number;
474 int context_initialized;
478 int8_t(*intra4x4_pred_mode);
481 uint8_t (*non_zero_count)[48];
483 #define LIST_NOT_USED -1 // FIXME rename?
484 #define PART_NOT_AVAILABLE -2
487 * block_offset[ 0..23] for frame macroblocks
488 * block_offset[24..47] for field macroblocks
490 int block_offset[2 * (16 * 3)];
492 uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
494 int b_stride; // FIXME use s->b4_stride
496 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
498 // interlacing specific flags
500 int picture_structure;
503 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
505 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
508 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
509 uint8_t *chroma_pred_mode_table;
510 uint8_t (*mvd_table[2])[2];
511 uint8_t *direct_table;
513 uint8_t zigzag_scan[16];
514 uint8_t zigzag_scan8x8[64];
515 uint8_t zigzag_scan8x8_cavlc[64];
516 uint8_t field_scan[16];
517 uint8_t field_scan8x8[64];
518 uint8_t field_scan8x8_cavlc[64];
519 const uint8_t *zigzag_scan_q0;
520 const uint8_t *zigzag_scan8x8_q0;
521 const uint8_t *zigzag_scan8x8_cavlc_q0;
522 const uint8_t *field_scan_q0;
523 const uint8_t *field_scan8x8_q0;
524 const uint8_t *field_scan8x8_cavlc_q0;
527 int mb_height, mb_width;
531 // =============================================================
532 // Things below are not used in the MB or more inner code
538 * Used to parse AVC variant of h264
540 int is_avc; ///< this flag is != 0 if codec is avc1
541 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
543 int bit_depth_luma; ///< luma bit depth from sps to detect changes
544 int chroma_format_idc; ///< chroma format from sps to detect changes
548 uint16_t *slice_table_base;
553 * frame_num for frames or 2 * frame_num + 1 for field pics.
558 * max_frame_num or 2 * max_frame_num for field pics.
562 H264Picture *short_ref[32];
563 H264Picture *long_ref[32];
564 H264Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
565 int last_pocs[MAX_DELAYED_PIC_COUNT];
566 H264Picture *next_output_pic;
567 int next_outputed_poc;
570 * memory management control operations buffer.
572 MMCO mmco[MAX_MMCO_COUNT];
576 int long_ref_count; ///< number of actual long term references
577 int short_ref_count; ///< number of actual short term references
580 * @name Members for slice based multithreading
584 * current slice number, used to initalize slice_num of each thread/context
589 * Max number of threads / contexts.
590 * This is equal to AVCodecContext.thread_count unless
591 * multithreaded decoding is impossible, in which case it is
596 int slice_context_count;
599 * 1 if the single thread fallback warning has already been
600 * displayed, 0 otherwise.
602 int single_decode_warning;
604 enum AVPictureType pict_type;
609 * Complement sei_pic_struct
610 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
611 * However, soft telecined frames may have these values.
612 * This is used in an attempt to flag soft telecine progressive.
614 int prev_interlaced_frame;
617 * recovery_frame is the frame_num at which the next frame should
618 * be fully constructed.
620 * Set to -1 when not expecting a recovery point.
625 * We have seen an IDR, so all the following frames in coded order are correctly
628 #define FRAME_RECOVERED_IDR (1 << 0)
630 * Sufficient number of frames have been decoded since a SEI recovery point,
631 * so all the following frames in presentation order are correct.
633 #define FRAME_RECOVERED_SEI (1 << 1)
635 int frame_recovered; ///< Initial frame has been completely recovered
637 /* for frame threading, this is set to 1
638 * after finish_setup() has been called, so we cannot modify
639 * some context properties (which are supposed to stay constant between
647 AVBufferPool *qscale_table_pool;
648 AVBufferPool *mb_type_pool;
649 AVBufferPool *motion_val_pool;
650 AVBufferPool *ref_index_pool;
652 /* Motion Estimation */
653 qpel_mc_func (*qpel_put)[16];
654 qpel_mc_func (*qpel_avg)[16];
657 extern const uint16_t ff_h264_mb_sizes[4];
662 int ff_h264_decode_seq_parameter_set(GetBitContext *gb, AVCodecContext *avctx,
668 int ff_h264_decode_picture_parameter_set(GetBitContext *gb, AVCodecContext *avctx,
669 H264ParamSets *ps, int bit_length);
672 * Free any data that may have been allocated in the H264 context
675 void ff_h264_free_context(H264Context *h);
678 * Reconstruct bitstream slice_type.
680 int ff_h264_get_slice_type(const H264SliceContext *sl);
686 int ff_h264_alloc_tables(H264Context *h);
688 int ff_h264_decode_ref_pic_list_reordering(const H264Context *h, H264SliceContext *sl);
689 void ff_h264_fill_mbaff_ref_list(H264SliceContext *sl);
690 void ff_h264_remove_all_refs(H264Context *h);
693 * Execute the reference picture marking (memory management control operations).
695 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
697 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb,
700 int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
702 void ff_h264_hl_decode_mb(const H264Context *h, H264SliceContext *sl);
703 int ff_h264_decode_init(AVCodecContext *avctx);
704 void ff_h264_decode_init_vlc(void);
707 * Decode a macroblock
708 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
710 int ff_h264_decode_mb_cavlc(const H264Context *h, H264SliceContext *sl);
713 * Decode a CABAC coded macroblock
714 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
716 int ff_h264_decode_mb_cabac(const H264Context *h, H264SliceContext *sl);
718 void ff_h264_init_cabac_states(const H264Context *h, H264SliceContext *sl);
720 void ff_h264_init_dequant_tables(H264Context *h);
722 void ff_h264_direct_dist_scale_factor(const H264Context *const h, H264SliceContext *sl);
723 void ff_h264_direct_ref_list_init(const H264Context *const h, H264SliceContext *sl);
724 void ff_h264_pred_direct_motion(const H264Context *const h, H264SliceContext *sl,
727 void ff_h264_filter_mb_fast(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
728 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
729 unsigned int linesize, unsigned int uvlinesize);
730 void ff_h264_filter_mb(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
731 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
732 unsigned int linesize, unsigned int uvlinesize);
744 /* Scan8 organization:
761 * DY/DU/DV are for luma/chroma DC.
764 #define LUMA_DC_BLOCK_INDEX 48
765 #define CHROMA_DC_BLOCK_INDEX 49
767 // This table must be here because scan8[constant] must be known at compiletime
768 static const uint8_t scan8[16 * 3 + 3] = {
769 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
770 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
771 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
772 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
773 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
774 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
775 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
776 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
777 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
778 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
779 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
780 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
781 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
784 static av_always_inline uint32_t pack16to32(int a, int b)
787 return (b & 0xFFFF) + (a << 16);
789 return (a & 0xFFFF) + (b << 16);
793 static av_always_inline uint16_t pack8to16(int a, int b)
796 return (b & 0xFF) + (a << 8);
798 return (a & 0xFF) + (b << 8);
805 static av_always_inline int get_chroma_qp(const H264Context *h, int t, int qscale)
807 return h->ps.pps->chroma_qp_table[t][qscale];
811 * Get the predicted intra4x4 prediction mode.
813 static av_always_inline int pred_intra_mode(const H264Context *h,
814 H264SliceContext *sl, int n)
816 const int index8 = scan8[n];
817 const int left = sl->intra4x4_pred_mode_cache[index8 - 1];
818 const int top = sl->intra4x4_pred_mode_cache[index8 - 8];
819 const int min = FFMIN(left, top);
821 ff_tlog(h->avctx, "mode:%d %d min:%d\n", left, top, min);
829 static av_always_inline void write_back_intra_pred_mode(const H264Context *h,
830 H264SliceContext *sl)
832 int8_t *i4x4 = sl->intra4x4_pred_mode + h->mb2br_xy[sl->mb_xy];
833 int8_t *i4x4_cache = sl->intra4x4_pred_mode_cache;
835 AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
836 i4x4[4] = i4x4_cache[7 + 8 * 3];
837 i4x4[5] = i4x4_cache[7 + 8 * 2];
838 i4x4[6] = i4x4_cache[7 + 8 * 1];
841 static av_always_inline void write_back_non_zero_count(const H264Context *h,
842 H264SliceContext *sl)
844 const int mb_xy = sl->mb_xy;
845 uint8_t *nnz = h->non_zero_count[mb_xy];
846 uint8_t *nnz_cache = sl->non_zero_count_cache;
848 AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
849 AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
850 AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
851 AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
852 AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
853 AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
854 AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
855 AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
857 if (!h->chroma_y_shift) {
858 AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
859 AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
860 AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
861 AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
865 static av_always_inline void write_back_motion_list(const H264Context *h,
866 H264SliceContext *sl,
869 int mb_type, int list)
871 int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
872 int16_t(*mv_src)[2] = &sl->mv_cache[list][scan8[0]];
873 AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
874 AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
875 AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
876 AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
878 uint8_t (*mvd_dst)[2] = &sl->mvd_table[list][FMO ? 8 * sl->mb_xy
879 : h->mb2br_xy[sl->mb_xy]];
880 uint8_t(*mvd_src)[2] = &sl->mvd_cache[list][scan8[0]];
881 if (IS_SKIP(mb_type)) {
884 AV_COPY64(mvd_dst, mvd_src + 8 * 3);
885 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
886 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
887 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
892 int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
893 int8_t *ref_cache = sl->ref_cache[list];
894 ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
895 ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
896 ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
897 ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
901 static av_always_inline void write_back_motion(const H264Context *h,
902 H264SliceContext *sl,
905 const int b_stride = h->b_stride;
906 const int b_xy = 4 * sl->mb_x + 4 * sl->mb_y * h->b_stride; // try mb2b(8)_xy
907 const int b8_xy = 4 * sl->mb_xy;
909 if (USES_LIST(mb_type, 0)) {
910 write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 0);
912 fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
913 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
915 if (USES_LIST(mb_type, 1))
916 write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 1);
918 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) {
919 if (IS_8X8(mb_type)) {
920 uint8_t *direct_table = &h->direct_table[4 * sl->mb_xy];
921 direct_table[1] = sl->sub_mb_type[1] >> 1;
922 direct_table[2] = sl->sub_mb_type[2] >> 1;
923 direct_table[3] = sl->sub_mb_type[3] >> 1;
928 static av_always_inline int get_dct8x8_allowed(const H264Context *h, H264SliceContext *sl)
930 if (h->ps.sps->direct_8x8_inference_flag)
931 return !(AV_RN64A(sl->sub_mb_type) &
932 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
933 0x0001000100010001ULL));
935 return !(AV_RN64A(sl->sub_mb_type) &
936 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
937 0x0001000100010001ULL));
940 int ff_h264_field_end(H264Context *h, H264SliceContext *sl, int in_setup);
942 int ff_h264_ref_picture(H264Context *h, H264Picture *dst, H264Picture *src);
943 void ff_h264_unref_picture(H264Context *h, H264Picture *pic);
945 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl);
947 void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl, int y, int height);
949 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl);
950 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count);
951 int ff_h264_update_thread_context(AVCodecContext *dst,
952 const AVCodecContext *src);
954 void ff_h264_flush_change(H264Context *h);
956 void ff_h264_free_tables(H264Context *h);
958 #endif /* AVCODEC_H264_H */