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 / 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"
35 #include "h264chroma.h"
39 #include "rectangle.h"
41 #define MAX_SPS_COUNT 32
42 #define MAX_PPS_COUNT 256
44 #define MAX_MMCO_COUNT 66
46 #define MAX_DELAYED_PIC_COUNT 16
48 #define MAX_MBPAIR_SIZE (256*1024) // a tighter bound could be calculated if someone cares about a few bytes
50 /* Compiling in interlaced support reduces the speed
51 * of progressive decoding by about 2%. */
52 #define ALLOW_INTERLACE
57 * The maximum number of slices supported by the decoder.
58 * must be a power of 2
62 #ifdef ALLOW_INTERLACE
63 #define MB_MBAFF h->mb_mbaff
64 #define MB_FIELD h->mb_field_decoding_flag
65 #define FRAME_MBAFF h->mb_aff_frame
66 #define FIELD_PICTURE (h->picture_structure != PICT_FRAME)
75 #define FIELD_PICTURE 0
77 #define IS_INTERLACED(mb_type) 0
83 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
86 #define CABAC h->pps.cabac
89 #define CHROMA (h->sps.chroma_format_idc)
90 #define CHROMA422 (h->sps.chroma_format_idc == 2)
91 #define CHROMA444 (h->sps.chroma_format_idc == 3)
93 #define EXTENDED_SAR 255
95 #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit
96 #define MB_TYPE_8x8DCT 0x01000000
97 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
98 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
100 #define QP_MAX_NUM (51 + 6*6) // The maximum supported qp
117 NAL_AUXILIARY_SLICE = 19,
118 NAL_FF_IGNORE = 0xff0f001,
125 SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
126 SEI_TYPE_PIC_TIMING = 1, ///< picture timing
127 SEI_TYPE_USER_DATA_ITU_T_T35 = 4, ///< user data registered by ITU-T Recommendation T.35
128 SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
129 SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync)
133 * pic_struct in picture timing SEI message
136 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
137 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
138 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
139 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
140 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
141 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
142 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
143 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
144 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
148 * Sequence parameter set
153 int chroma_format_idc;
154 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
155 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
156 int poc_type; ///< pic_order_cnt_type
157 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
158 int delta_pic_order_always_zero_flag;
159 int offset_for_non_ref_pic;
160 int offset_for_top_to_bottom_field;
161 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
162 int ref_frame_count; ///< num_ref_frames
163 int gaps_in_frame_num_allowed_flag;
164 int mb_width; ///< pic_width_in_mbs_minus1 + 1
165 int mb_height; ///< pic_height_in_map_units_minus1 + 1
166 int frame_mbs_only_flag;
167 int mb_aff; ///< mb_adaptive_frame_field_flag
168 int direct_8x8_inference_flag;
169 int crop; ///< frame_cropping_flag
170 unsigned int crop_left; ///< frame_cropping_rect_left_offset
171 unsigned int crop_right; ///< frame_cropping_rect_right_offset
172 unsigned int crop_top; ///< frame_cropping_rect_top_offset
173 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
174 int vui_parameters_present_flag;
176 int video_signal_type_present_flag;
178 int colour_description_present_flag;
179 enum AVColorPrimaries color_primaries;
180 enum AVColorTransferCharacteristic color_trc;
181 enum AVColorSpace colorspace;
182 int timing_info_present_flag;
183 uint32_t num_units_in_tick;
185 int fixed_frame_rate_flag;
186 short offset_for_ref_frame[256]; // FIXME dyn aloc?
187 int bitstream_restriction_flag;
188 int num_reorder_frames;
189 int scaling_matrix_present;
190 uint8_t scaling_matrix4[6][16];
191 uint8_t scaling_matrix8[6][64];
192 int nal_hrd_parameters_present_flag;
193 int vcl_hrd_parameters_present_flag;
194 int pic_struct_present_flag;
195 int time_offset_length;
196 int cpb_cnt; ///< See H.264 E.1.2
197 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
198 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
199 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
200 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
201 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
202 int residual_color_transform_flag; ///< residual_colour_transform_flag
203 int constraint_set_flags; ///< constraint_set[0-3]_flag
204 int new; ///< flag to keep track if the decoder context needs re-init due to changed SPS
208 * Picture parameter set
212 int cabac; ///< entropy_coding_mode_flag
213 int pic_order_present; ///< pic_order_present_flag
214 int slice_group_count; ///< num_slice_groups_minus1 + 1
215 int mb_slice_group_map_type;
216 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
217 int weighted_pred; ///< weighted_pred_flag
218 int weighted_bipred_idc;
219 int init_qp; ///< pic_init_qp_minus26 + 26
220 int init_qs; ///< pic_init_qs_minus26 + 26
221 int chroma_qp_index_offset[2];
222 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
223 int constrained_intra_pred; ///< constrained_intra_pred_flag
224 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
225 int transform_8x8_mode; ///< transform_8x8_mode_flag
226 uint8_t scaling_matrix4[6][16];
227 uint8_t scaling_matrix8[6][64];
228 uint8_t chroma_qp_table[2][QP_MAX_NUM+1]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
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
257 typedef struct H264Context {
258 AVCodecContext *avctx;
260 VideoDSPContext vdsp;
261 H264DSPContext h264dsp;
262 H264ChromaContext h264chroma;
263 H264QpelContext h264qpel;
265 ParseContext parse_context;
270 Picture *cur_pic_ptr;
273 int picture_range_start, picture_range_end;
275 int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
276 int chroma_qp[2]; // QPc
278 int qp_thresh; ///< QP threshold to skip loopfilter
281 int linesize, uvlinesize;
282 int chroma_x_shift, chroma_y_shift;
286 int data_partitioning;
287 int coded_picture_number;
290 int context_initialized;
298 int chroma_pred_mode;
299 int intra16x16_pred_mode;
304 int left_mb_xy[LEFT_MBS];
309 int left_type[LEFT_MBS];
311 const uint8_t *left_block;
312 int topleft_partition;
314 int8_t intra4x4_pred_mode_cache[5 * 8];
315 int8_t(*intra4x4_pred_mode);
317 unsigned int topleft_samples_available;
318 unsigned int top_samples_available;
319 unsigned int topright_samples_available;
320 unsigned int left_samples_available;
321 uint8_t (*top_borders[2])[(16 * 3) * 2];
324 * non zero coeff count cache.
325 * is 64 if not available.
327 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
329 uint8_t (*non_zero_count)[48];
332 * Motion vector cache.
334 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
335 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
336 #define LIST_NOT_USED -1 // FIXME rename?
337 #define PART_NOT_AVAILABLE -2
340 * number of neighbors (top and/or left) that used 8x8 dct
342 int neighbor_transform_size;
345 * block_offset[ 0..23] for frame macroblocks
346 * block_offset[24..47] for field macroblocks
348 int block_offset[2 * (16 * 3)];
350 uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
352 int b_stride; // FIXME use s->b4_stride
354 int mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
357 unsigned current_sps_id; ///< id of the current SPS
358 SPS sps; ///< current sps
363 PPS pps; // FIXME move to Picture perhaps? (->no) do we need that?
365 uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
366 uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
367 uint32_t(*dequant4_coeff[6])[16];
368 uint32_t(*dequant8_coeff[6])[64];
371 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
373 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
374 int slice_type_fixed;
376 // interlacing specific flags
378 int mb_field_decoding_flag;
379 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
380 int picture_structure;
383 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
385 // Weighted pred stuff
387 int use_weight_chroma;
388 int luma_log2_weight_denom;
389 int chroma_log2_weight_denom;
390 // The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
391 int luma_weight[48][2][2];
392 int chroma_weight[48][2][2][2];
393 int implicit_weight[48][48][2];
395 int direct_spatial_mv_pred;
398 int dist_scale_factor[32];
399 int dist_scale_factor_field[2][32];
400 int map_col_to_list0[2][16 + 32];
401 int map_col_to_list0_field[2][2][16 + 32];
404 * num_ref_idx_l0/1_active_minus1 + 1
406 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
407 unsigned int list_count;
408 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
409 Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
410 * Reordered version of default_ref_list
411 * according to picture reordering in slice header */
412 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
415 GetBitContext intra_gb;
416 GetBitContext inter_gb;
417 GetBitContext *intra_gb_ptr;
418 GetBitContext *inter_gb_ptr;
420 const uint8_t *intra_pcm_ptr;
421 DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2]; ///< as a dct coeffecient is int32_t in high depth, we need to reserve twice the space.
422 DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
423 int16_t mb_padding[256 * 2]; ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either check that i is not too large or ensure that there is some unused stuff after mb
429 uint8_t cabac_state[1024];
431 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
436 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
437 uint8_t *chroma_pred_mode_table;
438 int last_qscale_diff;
439 uint8_t (*mvd_table[2])[2];
440 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
441 uint8_t *direct_table;
442 uint8_t direct_cache[5 * 8];
444 uint8_t zigzag_scan[16];
445 uint8_t zigzag_scan8x8[64];
446 uint8_t zigzag_scan8x8_cavlc[64];
447 uint8_t field_scan[16];
448 uint8_t field_scan8x8[64];
449 uint8_t field_scan8x8_cavlc[64];
450 uint8_t zigzag_scan_q0[16];
451 uint8_t zigzag_scan8x8_q0[64];
452 uint8_t zigzag_scan8x8_cavlc_q0[64];
453 uint8_t field_scan_q0[16];
454 uint8_t field_scan8x8_q0[64];
455 uint8_t field_scan8x8_cavlc_q0[64];
463 int mb_height, mb_width;
471 int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
472 int slice_alpha_c0_offset;
473 int slice_beta_offset;
475 // =============================================================
476 // Things below are not used in the MB or more inner code
480 uint8_t *rbsp_buffer[2];
481 unsigned int rbsp_buffer_size[2];
484 * Used to parse AVC variant of h264
486 int is_avc; ///< this flag is != 0 if codec is avc1
487 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
488 int got_first; ///< this flag is != 0 if we've parsed a frame
490 int bit_depth_luma; ///< luma bit depth from sps to detect changes
491 int chroma_format_idc; ///< chroma format from sps to detect changes
493 SPS *sps_buffers[MAX_SPS_COUNT];
494 PPS *pps_buffers[MAX_PPS_COUNT];
496 int dequant_coeff_pps; ///< reinit tables when pps changes
498 uint16_t *slice_table_base;
503 int delta_poc_bottom;
506 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
507 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
508 int frame_num_offset; ///< for POC type 2
509 int prev_frame_num_offset; ///< for POC type 2
510 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
513 * frame_num for frames or 2 * frame_num + 1 for field pics.
518 * max_frame_num or 2 * max_frame_num for field pics.
522 int redundant_pic_count;
524 Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
525 Picture *short_ref[32];
526 Picture *long_ref[32];
527 Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
528 int last_pocs[MAX_DELAYED_PIC_COUNT];
529 Picture *next_output_pic;
531 int next_outputed_poc;
534 * memory management control operations buffer.
536 MMCO mmco[MAX_MMCO_COUNT];
540 int long_ref_count; ///< number of actual long term references
541 int short_ref_count; ///< number of actual short term references
546 * @name Members for slice based multithreading
549 struct H264Context *thread_context[MAX_THREADS];
552 * current slice number, used to initialize slice_num of each thread/context
557 * Max number of threads / contexts.
558 * This is equal to AVCodecContext.thread_count unless
559 * multithreaded decoding is impossible, in which case it is
564 int slice_context_count;
567 * 1 if the single thread fallback warning has already been
568 * displayed, 0 otherwise.
570 int single_decode_warning;
572 enum AVPictureType pict_type;
578 * pic_struct in picture timing SEI message
580 SEI_PicStructType sei_pic_struct;
583 * Complement sei_pic_struct
584 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
585 * However, soft telecined frames may have these values.
586 * This is used in an attempt to flag soft telecine progressive.
588 int prev_interlaced_frame;
591 * Bit set of clock types for fields/frames in picture timing SEI message.
592 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
598 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
600 int sei_dpb_output_delay;
603 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
605 int sei_cpb_removal_delay;
608 * recovery_frame_cnt from SEI message
610 * Set to -1 if no recovery point SEI message found or to number of frames
611 * before playback synchronizes. Frames having recovery point are key
614 int sei_recovery_frame_cnt;
616 * recovery_frame is the frame_num at which the next frame should
617 * be fully constructed.
619 * Set to -1 when not expecting a recovery point.
624 * Are the SEI recovery points looking valid.
626 int valid_recovery_point;
628 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
629 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
632 int sei_buffering_period_present; ///< Buffering period SEI flag
633 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
635 int cur_chroma_format_idc;
636 uint8_t *bipred_scratchpad;
638 int16_t slice_row[MAX_SLICES]; ///< to detect when MAX_SLICES is too low
640 int sync; ///< did we had a keyframe or recovery point
642 uint8_t parse_history[4];
643 int parse_history_count;
645 uint8_t *edge_emu_buffer;
646 int16_t *dc_val_base;
648 uint8_t *visualization_buffer[3]; ///< temporary buffer vor MV visualization
651 extern const uint8_t ff_h264_chroma_qp[7][QP_MAX_NUM + 1]; ///< One chroma qp table for each possible bit depth (8-14).
652 extern const uint16_t ff_h264_mb_sizes[4];
657 int ff_h264_decode_sei(H264Context *h);
662 int ff_h264_decode_seq_parameter_set(H264Context *h);
665 * compute profile from sps
667 int ff_h264_get_profile(SPS *sps);
672 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
675 * Decode a network abstraction layer unit.
676 * @param consumed is the number of bytes used as input
677 * @param length is the length of the array
678 * @param dst_length is the number of decoded bytes FIXME here
679 * or a decode rbsp tailing?
680 * @return decoded bytes, might be src+1 if no escapes
682 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
683 int *dst_length, int *consumed, int length);
686 * Free any data that may have been allocated in the H264 context
689 av_cold void ff_h264_free_context(H264Context *h);
692 * Reconstruct bitstream slice_type.
694 int ff_h264_get_slice_type(const H264Context *h);
700 int ff_h264_alloc_tables(H264Context *h);
703 * Fill the default_ref_list.
705 int ff_h264_fill_default_ref_list(H264Context *h);
707 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
708 void ff_h264_fill_mbaff_ref_list(H264Context *h);
709 void ff_h264_remove_all_refs(H264Context *h);
712 * Execute the reference picture marking (memory management control operations).
714 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
716 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb,
719 int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
722 * Check if the top & left blocks are available if needed & change the
723 * dc mode so it only uses the available blocks.
725 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
728 * Check if the top & left blocks are available if needed & change the
729 * dc mode so it only uses the available blocks.
731 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma);
733 void ff_h264_hl_decode_mb(H264Context *h);
734 int ff_h264_frame_start(H264Context *h);
735 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size);
736 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
737 av_cold void ff_h264_decode_init_vlc(void);
740 * Decode a macroblock
741 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
743 int ff_h264_decode_mb_cavlc(H264Context *h);
746 * Decode a CABAC coded macroblock
747 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
749 int ff_h264_decode_mb_cabac(H264Context *h);
751 void ff_h264_init_cabac_states(H264Context *h);
753 void ff_h264_direct_dist_scale_factor(H264Context *const h);
754 void ff_h264_direct_ref_list_init(H264Context *const h);
755 void ff_h264_pred_direct_motion(H264Context *const h, int *mb_type);
757 void ff_h264_filter_mb_fast(H264Context *h, int mb_x, int mb_y,
758 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
759 unsigned int linesize, unsigned int uvlinesize);
760 void ff_h264_filter_mb(H264Context *h, int mb_x, int mb_y,
761 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
762 unsigned int linesize, unsigned int uvlinesize);
765 * Reset SEI values at the beginning of the frame.
767 * @param h H.264 context.
769 void ff_h264_reset_sei(H264Context *h);
781 /* Scan8 organization:
798 * DY/DU/DV are for luma/chroma DC.
801 #define LUMA_DC_BLOCK_INDEX 48
802 #define CHROMA_DC_BLOCK_INDEX 49
804 // This table must be here because scan8[constant] must be known at compiletime
805 static const uint8_t scan8[16 * 3 + 3] = {
806 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
807 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
808 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
809 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
810 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
811 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
812 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
813 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
814 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
815 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
816 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
817 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
818 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
821 static av_always_inline uint32_t pack16to32(int a, int b)
824 return (b & 0xFFFF) + (a << 16);
826 return (a & 0xFFFF) + (b << 16);
830 static av_always_inline uint16_t pack8to16(int a, int b)
833 return (b & 0xFF) + (a << 8);
835 return (a & 0xFF) + (b << 8);
842 static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale)
844 return h->pps.chroma_qp_table[t][qscale];
848 * Get the predicted intra4x4 prediction mode.
850 static av_always_inline int pred_intra_mode(H264Context *h, int n)
852 const int index8 = scan8[n];
853 const int left = h->intra4x4_pred_mode_cache[index8 - 1];
854 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
855 const int min = FFMIN(left, top);
857 tprintf(h->avctx, "mode:%d %d min:%d\n", left, top, min);
865 static av_always_inline void write_back_intra_pred_mode(H264Context *h)
867 int8_t *i4x4 = h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
868 int8_t *i4x4_cache = h->intra4x4_pred_mode_cache;
870 AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
871 i4x4[4] = i4x4_cache[7 + 8 * 3];
872 i4x4[5] = i4x4_cache[7 + 8 * 2];
873 i4x4[6] = i4x4_cache[7 + 8 * 1];
876 static av_always_inline void write_back_non_zero_count(H264Context *h)
878 const int mb_xy = h->mb_xy;
879 uint8_t *nnz = h->non_zero_count[mb_xy];
880 uint8_t *nnz_cache = h->non_zero_count_cache;
882 AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
883 AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
884 AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
885 AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
886 AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
887 AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
888 AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
889 AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
891 if (!h->chroma_y_shift) {
892 AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
893 AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
894 AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
895 AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
899 static av_always_inline void write_back_motion_list(H264Context *h,
902 int mb_type, int list)
904 int16_t(*mv_dst)[2] = &h->cur_pic.f.motion_val[list][b_xy];
905 int16_t(*mv_src)[2] = &h->mv_cache[list][scan8[0]];
906 AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
907 AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
908 AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
909 AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
911 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8 * h->mb_xy
912 : h->mb2br_xy[h->mb_xy]];
913 uint8_t(*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
914 if (IS_SKIP(mb_type)) {
917 AV_COPY64(mvd_dst, mvd_src + 8 * 3);
918 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
919 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
920 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
925 int8_t *ref_index = &h->cur_pic.f.ref_index[list][b8_xy];
926 int8_t *ref_cache = h->ref_cache[list];
927 ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
928 ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
929 ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
930 ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
934 static av_always_inline void write_back_motion(H264Context *h, int mb_type)
936 const int b_stride = h->b_stride;
937 const int b_xy = 4 * h->mb_x + 4 * h->mb_y * h->b_stride; // try mb2b(8)_xy
938 const int b8_xy = 4 * h->mb_xy;
940 if (USES_LIST(mb_type, 0)) {
941 write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 0);
943 fill_rectangle(&h->cur_pic.f.ref_index[0][b8_xy],
944 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
946 if (USES_LIST(mb_type, 1))
947 write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 1);
949 if (h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC) {
950 if (IS_8X8(mb_type)) {
951 uint8_t *direct_table = &h->direct_table[4 * h->mb_xy];
952 direct_table[1] = h->sub_mb_type[1] >> 1;
953 direct_table[2] = h->sub_mb_type[2] >> 1;
954 direct_table[3] = h->sub_mb_type[3] >> 1;
959 static av_always_inline int get_dct8x8_allowed(H264Context *h)
961 if (h->sps.direct_8x8_inference_flag)
962 return !(AV_RN64A(h->sub_mb_type) &
963 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
964 0x0001000100010001ULL));
966 return !(AV_RN64A(h->sub_mb_type) &
967 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
968 0x0001000100010001ULL));
971 void ff_h264_draw_horiz_band(H264Context *h, int y, int height);
973 #endif /* AVCODEC_H264_H */