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"
34 #include "mpegvideo.h"
37 #include "rectangle.h"
39 #define interlaced_dct interlaced_dct_is_a_bad_name
40 #define mb_intra mb_intra_is_not_initialized_see_mb_type
42 #define MAX_SPS_COUNT 32
43 #define MAX_PPS_COUNT 256
45 #define MAX_MMCO_COUNT 66
47 #define MAX_DELAYED_PIC_COUNT 16
49 /* Compiling in interlaced support reduces the speed
50 * of progressive decoding by about 2%. */
51 #define ALLOW_INTERLACE
56 * The maximum number of slices supported by the decoder.
57 * must be a power of 2
61 #ifdef ALLOW_INTERLACE
62 #define MB_MBAFF h->mb_mbaff
63 #define MB_FIELD h->mb_field_decoding_flag
64 #define FRAME_MBAFF h->mb_aff_frame
65 #define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
74 #define FIELD_PICTURE 0
76 #define IS_INTERLACED(mb_type) 0
82 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
85 #define CABAC h->pps.cabac
88 #define CHROMA422 (h->sps.chroma_format_idc == 2)
89 #define CHROMA444 (h->sps.chroma_format_idc == 3)
91 #define EXTENDED_SAR 255
93 #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit
94 #define MB_TYPE_8x8DCT 0x01000000
95 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
96 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
99 * Value of Picture.reference when Picture is not a reference picture, but
100 * is held for delayed output.
102 #define DELAYED_PIC_REF 4
104 #define QP_MAX_NUM (51 + 2 * 6) // The maximum supported qp
121 NAL_AUXILIARY_SLICE = 19
128 SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
129 SEI_TYPE_PIC_TIMING = 1, ///< picture timing
130 SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
131 SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync)
135 * pic_struct in picture timing SEI message
138 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
139 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
140 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
141 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
142 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
143 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
144 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
145 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
146 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
150 * Sequence parameter set
155 int chroma_format_idc;
156 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
157 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
158 int poc_type; ///< pic_order_cnt_type
159 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
160 int delta_pic_order_always_zero_flag;
161 int offset_for_non_ref_pic;
162 int offset_for_top_to_bottom_field;
163 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
164 int ref_frame_count; ///< num_ref_frames
165 int gaps_in_frame_num_allowed_flag;
166 int mb_width; ///< pic_width_in_mbs_minus1 + 1
167 int mb_height; ///< pic_height_in_map_units_minus1 + 1
168 int frame_mbs_only_flag;
169 int mb_aff; ///< mb_adaptive_frame_field_flag
170 int direct_8x8_inference_flag;
171 int crop; ///< frame_cropping_flag
172 unsigned int crop_left; ///< frame_cropping_rect_left_offset
173 unsigned int crop_right; ///< frame_cropping_rect_right_offset
174 unsigned int crop_top; ///< frame_cropping_rect_top_offset
175 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
176 int vui_parameters_present_flag;
178 int video_signal_type_present_flag;
180 int colour_description_present_flag;
181 enum AVColorPrimaries color_primaries;
182 enum AVColorTransferCharacteristic color_trc;
183 enum AVColorSpace colorspace;
184 int timing_info_present_flag;
185 uint32_t num_units_in_tick;
187 int fixed_frame_rate_flag;
188 short offset_for_ref_frame[256]; // FIXME dyn aloc?
189 int bitstream_restriction_flag;
190 int num_reorder_frames;
191 int scaling_matrix_present;
192 uint8_t scaling_matrix4[6][16];
193 uint8_t scaling_matrix8[6][64];
194 int nal_hrd_parameters_present_flag;
195 int vcl_hrd_parameters_present_flag;
196 int pic_struct_present_flag;
197 int time_offset_length;
198 int cpb_cnt; ///< See H.264 E.1.2
199 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
200 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
201 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
202 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
203 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
204 int residual_color_transform_flag; ///< residual_colour_transform_flag
205 int constraint_set_flags; ///< constraint_set[0-3]_flag
209 * Picture parameter set
213 int cabac; ///< entropy_coding_mode_flag
214 int pic_order_present; ///< pic_order_present_flag
215 int slice_group_count; ///< num_slice_groups_minus1 + 1
216 int mb_slice_group_map_type;
217 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
218 int weighted_pred; ///< weighted_pred_flag
219 int weighted_bipred_idc;
220 int init_qp; ///< pic_init_qp_minus26 + 26
221 int init_qs; ///< pic_init_qs_minus26 + 26
222 int chroma_qp_index_offset[2];
223 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
224 int constrained_intra_pred; ///< constrained_intra_pred_flag
225 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
226 int transform_8x8_mode; ///< transform_8x8_mode_flag
227 uint8_t scaling_matrix4[6][16];
228 uint8_t scaling_matrix8[6][64];
229 uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
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
258 typedef struct H264Context {
260 H264DSPContext h264dsp;
261 int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
262 int chroma_qp[2]; // QPc
264 int qp_thresh; ///< QP threshold to skip loopfilter
270 int chroma_pred_mode;
271 int intra16x16_pred_mode;
276 int left_mb_xy[LEFT_MBS];
281 int left_type[LEFT_MBS];
283 const uint8_t *left_block;
284 int topleft_partition;
286 int8_t intra4x4_pred_mode_cache[5 * 8];
287 int8_t(*intra4x4_pred_mode);
289 unsigned int topleft_samples_available;
290 unsigned int top_samples_available;
291 unsigned int topright_samples_available;
292 unsigned int left_samples_available;
293 uint8_t (*top_borders[2])[(16 * 3) * 2];
296 * non zero coeff count cache.
297 * is 64 if not available.
299 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
301 uint8_t (*non_zero_count)[48];
304 * Motion vector cache.
306 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
307 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
308 #define LIST_NOT_USED -1 // FIXME rename?
309 #define PART_NOT_AVAILABLE -2
312 * number of neighbors (top and/or left) that used 8x8 dct
314 int neighbor_transform_size;
317 * block_offset[ 0..23] for frame macroblocks
318 * block_offset[24..47] for field macroblocks
320 int block_offset[2 * (16 * 3)];
322 uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
324 int b_stride; // FIXME use s->b4_stride
326 int mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
332 SPS sps; ///< current sps
337 PPS pps; // FIXME move to Picture perhaps? (->no) do we need that?
339 uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
340 uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
341 uint32_t(*dequant4_coeff[6])[16];
342 uint32_t(*dequant8_coeff[6])[64];
345 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
347 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
348 int slice_type_fixed;
350 // interlacing specific flags
352 int mb_field_decoding_flag;
353 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
355 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
357 // Weighted pred stuff
359 int use_weight_chroma;
360 int luma_log2_weight_denom;
361 int chroma_log2_weight_denom;
362 // The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
363 int luma_weight[48][2][2];
364 int chroma_weight[48][2][2][2];
365 int implicit_weight[48][48][2];
367 int direct_spatial_mv_pred;
370 int dist_scale_factor[16];
371 int dist_scale_factor_field[2][32];
372 int map_col_to_list0[2][16 + 32];
373 int map_col_to_list0_field[2][2][16 + 32];
376 * num_ref_idx_l0/1_active_minus1 + 1
378 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
379 unsigned int list_count;
380 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
381 Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
382 * Reordered version of default_ref_list
383 * according to picture reordering in slice header */
384 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
387 GetBitContext intra_gb;
388 GetBitContext inter_gb;
389 GetBitContext *intra_gb_ptr;
390 GetBitContext *inter_gb_ptr;
392 DECLARE_ALIGNED(16, DCTELEM, mb)[16 * 48 * 2]; ///< as a dct coeffecient is int32_t in high depth, we need to reserve twice the space.
393 DECLARE_ALIGNED(16, DCTELEM, mb_luma_dc)[3][16 * 2];
394 DCTELEM 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
400 uint8_t cabac_state[1024];
402 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
407 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
408 uint8_t *chroma_pred_mode_table;
409 int last_qscale_diff;
410 uint8_t (*mvd_table[2])[2];
411 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
412 uint8_t *direct_table;
413 uint8_t direct_cache[5 * 8];
415 uint8_t zigzag_scan[16];
416 uint8_t zigzag_scan8x8[64];
417 uint8_t zigzag_scan8x8_cavlc[64];
418 uint8_t field_scan[16];
419 uint8_t field_scan8x8[64];
420 uint8_t field_scan8x8_cavlc[64];
421 const uint8_t *zigzag_scan_q0;
422 const uint8_t *zigzag_scan8x8_q0;
423 const uint8_t *zigzag_scan8x8_cavlc_q0;
424 const uint8_t *field_scan_q0;
425 const uint8_t *field_scan8x8_q0;
426 const uint8_t *field_scan8x8_cavlc_q0;
435 int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
436 int slice_alpha_c0_offset;
437 int slice_beta_offset;
439 // =============================================================
440 // Things below are not used in the MB or more inner code
444 uint8_t *rbsp_buffer[2];
445 unsigned int rbsp_buffer_size[2];
448 * Used to parse AVC variant of h264
450 int is_avc; ///< this flag is != 0 if codec is avc1
451 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
452 int got_first; ///< this flag is != 0 if we've parsed a frame
454 SPS *sps_buffers[MAX_SPS_COUNT];
455 PPS *pps_buffers[MAX_PPS_COUNT];
457 int dequant_coeff_pps; ///< reinit tables when pps changes
459 uint16_t *slice_table_base;
464 int delta_poc_bottom;
467 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
468 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
469 int frame_num_offset; ///< for POC type 2
470 int prev_frame_num_offset; ///< for POC type 2
471 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
474 * frame_num for frames or 2 * frame_num + 1 for field pics.
479 * max_frame_num or 2 * max_frame_num for field pics.
483 int redundant_pic_count;
485 Picture *short_ref[32];
486 Picture *long_ref[32];
487 Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
488 Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
489 int last_pocs[MAX_DELAYED_PIC_COUNT];
490 Picture *next_output_pic;
492 int next_outputed_poc;
495 * memory management control operations buffer.
497 MMCO mmco[MAX_MMCO_COUNT];
501 int long_ref_count; ///< number of actual long term references
502 int short_ref_count; ///< number of actual short term references
507 * @name Members for slice based multithreading
510 struct H264Context *thread_context[MAX_THREADS];
513 * current slice number, used to initalize slice_num of each thread/context
518 * Max number of threads / contexts.
519 * This is equal to AVCodecContext.thread_count unless
520 * multithreaded decoding is impossible, in which case it is
526 * 1 if the single thread fallback warning has already been
527 * displayed, 0 otherwise.
529 int single_decode_warning;
535 * pic_struct in picture timing SEI message
537 SEI_PicStructType sei_pic_struct;
540 * Complement sei_pic_struct
541 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
542 * However, soft telecined frames may have these values.
543 * This is used in an attempt to flag soft telecine progressive.
545 int prev_interlaced_frame;
548 * Bit set of clock types for fields/frames in picture timing SEI message.
549 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
555 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
557 int sei_dpb_output_delay;
560 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
562 int sei_cpb_removal_delay;
565 * recovery_frame_cnt from SEI message
567 * Set to -1 if no recovery point SEI message found or to number of frames
568 * before playback synchronizes. Frames having recovery point are key
571 int sei_recovery_frame_cnt;
573 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
574 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
577 int sei_buffering_period_present; ///< Buffering period SEI flag
578 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
580 int cur_chroma_format_idc;
583 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).
584 extern const uint16_t ff_h264_mb_sizes[4];
589 int ff_h264_decode_sei(H264Context *h);
594 int ff_h264_decode_seq_parameter_set(H264Context *h);
597 * compute profile from sps
599 int ff_h264_get_profile(SPS *sps);
604 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
607 * Decode a network abstraction layer unit.
608 * @param consumed is the number of bytes used as input
609 * @param length is the length of the array
610 * @param dst_length is the number of decoded bytes FIXME here
611 * or a decode rbsp tailing?
612 * @return decoded bytes, might be src+1 if no escapes
614 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
615 int *dst_length, int *consumed, int length);
618 * Free any data that may have been allocated in the H264 context
621 av_cold void ff_h264_free_context(H264Context *h);
624 * Reconstruct bitstream slice_type.
626 int ff_h264_get_slice_type(const H264Context *h);
632 int ff_h264_alloc_tables(H264Context *h);
635 * Fill the default_ref_list.
637 int ff_h264_fill_default_ref_list(H264Context *h);
639 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
640 void ff_h264_fill_mbaff_ref_list(H264Context *h);
641 void ff_h264_remove_all_refs(H264Context *h);
644 * Execute the reference picture marking (memory management control operations).
646 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
648 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
650 void ff_generate_sliding_window_mmcos(H264Context *h);
653 * Check if the top & left blocks are available if needed & change the
654 * dc mode so it only uses the available blocks.
656 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
659 * Check if the top & left blocks are available if needed & change the
660 * dc mode so it only uses the available blocks.
662 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma);
664 void ff_h264_hl_decode_mb(H264Context *h);
665 int ff_h264_frame_start(H264Context *h);
666 int ff_h264_decode_extradata(H264Context *h);
667 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
668 av_cold void ff_h264_decode_init_vlc(void);
671 * Decode a macroblock
672 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
674 int ff_h264_decode_mb_cavlc(H264Context *h);
677 * Decode a CABAC coded macroblock
678 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
680 int ff_h264_decode_mb_cabac(H264Context *h);
682 void ff_h264_init_cabac_states(H264Context *h);
684 void ff_h264_direct_dist_scale_factor(H264Context *const h);
685 void ff_h264_direct_ref_list_init(H264Context *const h);
686 void ff_h264_pred_direct_motion(H264Context *const h, int *mb_type);
688 void ff_h264_filter_mb_fast(H264Context *h, int mb_x, int mb_y,
689 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
690 unsigned int linesize, unsigned int uvlinesize);
691 void ff_h264_filter_mb(H264Context *h, int mb_x, int mb_y,
692 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
693 unsigned int linesize, unsigned int uvlinesize);
696 * Reset SEI values at the beginning of the frame.
698 * @param h H.264 context.
700 void ff_h264_reset_sei(H264Context *h);
712 /* Scan8 organization:
729 * DY/DU/DV are for luma/chroma DC.
732 #define LUMA_DC_BLOCK_INDEX 48
733 #define CHROMA_DC_BLOCK_INDEX 49
735 // This table must be here because scan8[constant] must be known at compiletime
736 static const uint8_t scan8[16 * 3 + 3] = {
737 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
738 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
739 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
740 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
741 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
742 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
743 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
744 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
745 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
746 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
747 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
748 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
749 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
752 static av_always_inline uint32_t pack16to32(int a, int b)
755 return (b & 0xFFFF) + (a << 16);
757 return (a & 0xFFFF) + (b << 16);
761 static av_always_inline uint16_t pack8to16(int a, int b)
764 return (b & 0xFF) + (a << 8);
766 return (a & 0xFF) + (b << 8);
773 static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale)
775 return h->pps.chroma_qp_table[t][qscale];
779 * Get the predicted intra4x4 prediction mode.
781 static av_always_inline int pred_intra_mode(H264Context *h, int n)
783 const int index8 = scan8[n];
784 const int left = h->intra4x4_pred_mode_cache[index8 - 1];
785 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
786 const int min = FFMIN(left, top);
788 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left, top, min);
796 static av_always_inline void write_back_intra_pred_mode(H264Context *h)
798 int8_t *i4x4 = h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
799 int8_t *i4x4_cache = h->intra4x4_pred_mode_cache;
801 AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
802 i4x4[4] = i4x4_cache[7 + 8 * 3];
803 i4x4[5] = i4x4_cache[7 + 8 * 2];
804 i4x4[6] = i4x4_cache[7 + 8 * 1];
807 static av_always_inline void write_back_non_zero_count(H264Context *h)
809 const int mb_xy = h->mb_xy;
810 uint8_t *nnz = h->non_zero_count[mb_xy];
811 uint8_t *nnz_cache = h->non_zero_count_cache;
813 AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
814 AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
815 AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
816 AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
817 AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
818 AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
819 AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
820 AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
822 if (!h->s.chroma_y_shift) {
823 AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
824 AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
825 AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
826 AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
830 static av_always_inline void write_back_motion_list(H264Context *h,
831 MpegEncContext *const s,
834 int mb_type, int list)
836 int16_t(*mv_dst)[2] = &s->current_picture.f.motion_val[list][b_xy];
837 int16_t(*mv_src)[2] = &h->mv_cache[list][scan8[0]];
838 AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
839 AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
840 AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
841 AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
843 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8 * h->mb_xy
844 : h->mb2br_xy[h->mb_xy]];
845 uint8_t(*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
846 if (IS_SKIP(mb_type)) {
849 AV_COPY64(mvd_dst, mvd_src + 8 * 3);
850 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
851 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
852 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
857 int8_t *ref_index = &s->current_picture.f.ref_index[list][b8_xy];
858 int8_t *ref_cache = h->ref_cache[list];
859 ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
860 ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
861 ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
862 ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
866 static av_always_inline void write_back_motion(H264Context *h, int mb_type)
868 MpegEncContext *const s = &h->s;
869 const int b_stride = h->b_stride;
870 const int b_xy = 4 * s->mb_x + 4 * s->mb_y * h->b_stride; // try mb2b(8)_xy
871 const int b8_xy = 4 * h->mb_xy;
873 if (USES_LIST(mb_type, 0)) {
874 write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 0);
876 fill_rectangle(&s->current_picture.f.ref_index[0][b8_xy],
877 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
879 if (USES_LIST(mb_type, 1))
880 write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 1);
882 if (h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC) {
883 if (IS_8X8(mb_type)) {
884 uint8_t *direct_table = &h->direct_table[4 * h->mb_xy];
885 direct_table[1] = h->sub_mb_type[1] >> 1;
886 direct_table[2] = h->sub_mb_type[2] >> 1;
887 direct_table[3] = h->sub_mb_type[3] >> 1;
892 static av_always_inline int get_dct8x8_allowed(H264Context *h)
894 if (h->sps.direct_8x8_inference_flag)
895 return !(AV_RN64A(h->sub_mb_type) &
896 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
897 0x0001000100010001ULL));
899 return !(AV_RN64A(h->sub_mb_type) &
900 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
901 0x0001000100010001ULL));
904 #endif /* AVCODEC_H264_H */