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"
33 #include "error_resilience.h"
35 #include "mpegvideo.h"
36 #include "h264chroma.h"
40 #include "rectangle.h"
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 #define MAX_MBPAIR_SIZE (256*1024) // a tighter bound could be calculated if someone cares about a few bytes
51 /* Compiling in interlaced support reduces the speed
52 * of progressive decoding by about 2%. */
53 #define ALLOW_INTERLACE
58 * The maximum number of slices supported by the decoder.
59 * must be a power of 2
63 #ifdef ALLOW_INTERLACE
64 #define MB_MBAFF(h) h->mb_mbaff
65 #define MB_FIELD(h) h->mb_field_decoding_flag
66 #define FRAME_MBAFF(h) h->mb_aff_frame
67 #define FIELD_PICTURE(h) (h->picture_structure != PICT_FRAME)
75 #define FRAME_MBAFF(h) 0
76 #define FIELD_PICTURE(h) 0
78 #define IS_INTERLACED(mb_type) 0
84 #define FIELD_OR_MBAFF_PICTURE(h) (FRAME_MBAFF(h) || FIELD_PICTURE(h))
87 #define CABAC(h) h->pps.cabac
90 #define CHROMA(h) (h->sps.chroma_format_idc)
91 #define CHROMA422(h) (h->sps.chroma_format_idc == 2)
92 #define CHROMA444(h) (h->sps.chroma_format_idc == 3)
94 #define EXTENDED_SAR 255
96 #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit
97 #define MB_TYPE_8x8DCT 0x01000000
98 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
99 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
101 #define QP_MAX_NUM (51 + 6*6) // The maximum supported qp
118 NAL_AUXILIARY_SLICE = 19,
119 NAL_FF_IGNORE = 0xff0f001,
126 SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
127 SEI_TYPE_PIC_TIMING = 1, ///< picture timing
128 SEI_TYPE_USER_DATA_ITU_T_T35 = 4, ///< user data registered by ITU-T Recommendation T.35
129 SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
130 SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync)
134 * pic_struct in picture timing SEI message
137 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
138 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
139 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
140 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
141 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
142 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
143 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
144 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
145 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
149 * Sequence parameter set
154 int chroma_format_idc;
155 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
156 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
157 int poc_type; ///< pic_order_cnt_type
158 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
159 int delta_pic_order_always_zero_flag;
160 int offset_for_non_ref_pic;
161 int offset_for_top_to_bottom_field;
162 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
163 int ref_frame_count; ///< num_ref_frames
164 int gaps_in_frame_num_allowed_flag;
165 int mb_width; ///< pic_width_in_mbs_minus1 + 1
166 int mb_height; ///< pic_height_in_map_units_minus1 + 1
167 int frame_mbs_only_flag;
168 int mb_aff; ///< mb_adaptive_frame_field_flag
169 int direct_8x8_inference_flag;
170 int crop; ///< frame_cropping_flag
172 /* those 4 are already in luma samples */
173 unsigned int crop_left; ///< frame_cropping_rect_left_offset
174 unsigned int crop_right; ///< frame_cropping_rect_right_offset
175 unsigned int crop_top; ///< frame_cropping_rect_top_offset
176 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
177 int vui_parameters_present_flag;
179 int video_signal_type_present_flag;
181 int colour_description_present_flag;
182 enum AVColorPrimaries color_primaries;
183 enum AVColorTransferCharacteristic color_trc;
184 enum AVColorSpace colorspace;
185 int timing_info_present_flag;
186 uint32_t num_units_in_tick;
188 int fixed_frame_rate_flag;
189 short offset_for_ref_frame[256]; // FIXME dyn aloc?
190 int bitstream_restriction_flag;
191 int num_reorder_frames;
192 int scaling_matrix_present;
193 uint8_t scaling_matrix4[6][16];
194 uint8_t scaling_matrix8[6][64];
195 int nal_hrd_parameters_present_flag;
196 int vcl_hrd_parameters_present_flag;
197 int pic_struct_present_flag;
198 int time_offset_length;
199 int cpb_cnt; ///< See H.264 E.1.2
200 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
201 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
202 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
203 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
204 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
205 int residual_color_transform_flag; ///< residual_colour_transform_flag
206 int constraint_set_flags; ///< constraint_set[0-3]_flag
207 int new; ///< flag to keep track if the decoder context needs re-init due to changed SPS
211 * Picture parameter set
215 int cabac; ///< entropy_coding_mode_flag
216 int pic_order_present; ///< pic_order_present_flag
217 int slice_group_count; ///< num_slice_groups_minus1 + 1
218 int mb_slice_group_map_type;
219 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
220 int weighted_pred; ///< weighted_pred_flag
221 int weighted_bipred_idc;
222 int init_qp; ///< pic_init_qp_minus26 + 26
223 int init_qs; ///< pic_init_qs_minus26 + 26
224 int chroma_qp_index_offset[2];
225 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
226 int constrained_intra_pred; ///< constrained_intra_pred_flag
227 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
228 int transform_8x8_mode; ///< transform_8x8_mode_flag
229 uint8_t scaling_matrix4[6][16];
230 uint8_t scaling_matrix8[6][64];
231 uint8_t chroma_qp_table[2][QP_MAX_NUM+1]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
236 * Memory management control operation opcode.
238 typedef enum MMCOOpcode {
249 * Memory management control operation.
251 typedef struct MMCO {
253 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
254 int long_arg; ///< index, pic_num, or num long refs depending on opcode
260 typedef struct H264Context {
261 AVCodecContext *avctx;
262 VideoDSPContext vdsp;
263 H264DSPContext h264dsp;
264 H264ChromaContext h264chroma;
265 H264QpelContext h264qpel;
267 ParseContext parse_context;
273 Picture *cur_pic_ptr;
276 int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
277 int chroma_qp[2]; // QPc
279 int qp_thresh; ///< QP threshold to skip loopfilter
281 /* coded dimensions -- 16 * mb w/h */
283 int linesize, uvlinesize;
284 int chroma_x_shift, chroma_y_shift;
288 int data_partitioning;
289 int coded_picture_number;
292 int context_initialized;
300 int chroma_pred_mode;
301 int intra16x16_pred_mode;
306 int left_mb_xy[LEFT_MBS];
311 int left_type[LEFT_MBS];
313 const uint8_t *left_block;
314 int topleft_partition;
316 int8_t intra4x4_pred_mode_cache[5 * 8];
317 int8_t(*intra4x4_pred_mode);
319 unsigned int topleft_samples_available;
320 unsigned int top_samples_available;
321 unsigned int topright_samples_available;
322 unsigned int left_samples_available;
323 uint8_t (*top_borders[2])[(16 * 3) * 2];
326 * non zero coeff count cache.
327 * is 64 if not available.
329 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
331 uint8_t (*non_zero_count)[48];
334 * Motion vector cache.
336 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
337 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
338 #define LIST_NOT_USED -1 // FIXME rename?
339 #define PART_NOT_AVAILABLE -2
342 * number of neighbors (top and/or left) that used 8x8 dct
344 int neighbor_transform_size;
347 * block_offset[ 0..23] for frame macroblocks
348 * block_offset[24..47] for field macroblocks
350 int block_offset[2 * (16 * 3)];
352 uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
354 int b_stride; // FIXME use s->b4_stride
356 int mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
359 unsigned current_sps_id; ///< id of the current SPS
360 SPS sps; ///< current sps
365 PPS pps; // FIXME move to Picture perhaps? (->no) do we need that?
367 uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
368 uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
369 uint32_t(*dequant4_coeff[6])[16];
370 uint32_t(*dequant8_coeff[6])[64];
373 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
375 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
376 int slice_type_fixed;
378 // interlacing specific flags
380 int mb_field_decoding_flag;
381 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
382 int picture_structure;
385 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
387 // Weighted pred stuff
389 int use_weight_chroma;
390 int luma_log2_weight_denom;
391 int chroma_log2_weight_denom;
392 // The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
393 int luma_weight[48][2][2];
394 int chroma_weight[48][2][2][2];
395 int implicit_weight[48][48][2];
397 int direct_spatial_mv_pred;
400 int dist_scale_factor[32];
401 int dist_scale_factor_field[2][32];
402 int map_col_to_list0[2][16 + 32];
403 int map_col_to_list0_field[2][2][16 + 32];
406 * num_ref_idx_l0/1_active_minus1 + 1
408 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
409 unsigned int list_count;
410 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
411 Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
412 * Reordered version of default_ref_list
413 * according to picture reordering in slice header */
414 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
417 GetBitContext intra_gb;
418 GetBitContext inter_gb;
419 GetBitContext *intra_gb_ptr;
420 GetBitContext *inter_gb_ptr;
422 const uint8_t *intra_pcm_ptr;
423 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.
424 DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
425 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
431 uint8_t cabac_state[1024];
433 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
438 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
439 uint8_t *chroma_pred_mode_table;
440 int last_qscale_diff;
441 uint8_t (*mvd_table[2])[2];
442 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
443 uint8_t *direct_table;
444 uint8_t direct_cache[5 * 8];
446 uint8_t zigzag_scan[16];
447 uint8_t zigzag_scan8x8[64];
448 uint8_t zigzag_scan8x8_cavlc[64];
449 uint8_t field_scan[16];
450 uint8_t field_scan8x8[64];
451 uint8_t field_scan8x8_cavlc[64];
452 uint8_t zigzag_scan_q0[16];
453 uint8_t zigzag_scan8x8_q0[64];
454 uint8_t zigzag_scan8x8_cavlc_q0[64];
455 uint8_t field_scan_q0[16];
456 uint8_t field_scan8x8_q0[64];
457 uint8_t field_scan8x8_cavlc_q0[64];
465 int mb_height, mb_width;
473 int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
474 int slice_alpha_c0_offset;
475 int slice_beta_offset;
477 // =============================================================
478 // Things below are not used in the MB or more inner code
482 uint8_t *rbsp_buffer[2];
483 unsigned int rbsp_buffer_size[2];
486 * Used to parse AVC variant of h264
488 int is_avc; ///< this flag is != 0 if codec is avc1
489 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
490 int got_first; ///< this flag is != 0 if we've parsed a frame
492 int bit_depth_luma; ///< luma bit depth from sps to detect changes
493 int chroma_format_idc; ///< chroma format from sps to detect changes
495 SPS *sps_buffers[MAX_SPS_COUNT];
496 PPS *pps_buffers[MAX_PPS_COUNT];
498 int dequant_coeff_pps; ///< reinit tables when pps changes
500 uint16_t *slice_table_base;
505 int delta_poc_bottom;
508 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
509 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
510 int frame_num_offset; ///< for POC type 2
511 int prev_frame_num_offset; ///< for POC type 2
512 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
515 * frame_num for frames or 2 * frame_num + 1 for field pics.
520 * max_frame_num or 2 * max_frame_num for field pics.
524 int redundant_pic_count;
526 Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
527 Picture *short_ref[32];
528 Picture *long_ref[32];
529 Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
530 int last_pocs[MAX_DELAYED_PIC_COUNT];
531 Picture *next_output_pic;
533 int next_outputed_poc;
536 * memory management control operations buffer.
538 MMCO mmco[MAX_MMCO_COUNT];
542 int long_ref_count; ///< number of actual long term references
543 int short_ref_count; ///< number of actual short term references
548 * @name Members for slice based multithreading
551 struct H264Context *thread_context[MAX_THREADS];
554 * current slice number, used to initialize slice_num of each thread/context
559 * Max number of threads / contexts.
560 * This is equal to AVCodecContext.thread_count unless
561 * multithreaded decoding is impossible, in which case it is
566 int slice_context_count;
569 * 1 if the single thread fallback warning has already been
570 * displayed, 0 otherwise.
572 int single_decode_warning;
574 enum AVPictureType pict_type;
577 unsigned int last_ref_count[2];
581 * pic_struct in picture timing SEI message
583 SEI_PicStructType sei_pic_struct;
586 * Complement sei_pic_struct
587 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
588 * However, soft telecined frames may have these values.
589 * This is used in an attempt to flag soft telecine progressive.
591 int prev_interlaced_frame;
594 * Bit set of clock types for fields/frames in picture timing SEI message.
595 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
601 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
603 int sei_dpb_output_delay;
606 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
608 int sei_cpb_removal_delay;
611 * recovery_frame_cnt from SEI message
613 * Set to -1 if no recovery point SEI message found or to number of frames
614 * before playback synchronizes. Frames having recovery point are key
617 int sei_recovery_frame_cnt;
619 * recovery_frame is the frame_num at which the next frame should
620 * be fully constructed.
622 * Set to -1 when not expecting a recovery point.
627 * Are the SEI recovery points looking valid.
629 int valid_recovery_point;
631 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
632 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
635 int sei_buffering_period_present; ///< Buffering period SEI flag
636 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
638 int cur_chroma_format_idc;
639 uint8_t *bipred_scratchpad;
641 int16_t slice_row[MAX_SLICES]; ///< to detect when MAX_SLICES is too low
643 int sync; ///< did we had a keyframe or recovery point
645 uint8_t parse_history[4];
646 int parse_history_count;
648 uint8_t *edge_emu_buffer;
649 int16_t *dc_val_base;
651 uint8_t *visualization_buffer[3]; ///< temporary buffer vor MV visualization
653 AVBufferPool *qscale_table_pool;
654 AVBufferPool *mb_type_pool;
655 AVBufferPool *motion_val_pool;
656 AVBufferPool *ref_index_pool;
659 extern const uint8_t ff_h264_chroma_qp[7][QP_MAX_NUM + 1]; ///< One chroma qp table for each possible bit depth (8-14).
660 extern const uint16_t ff_h264_mb_sizes[4];
665 int ff_h264_decode_sei(H264Context *h);
670 int ff_h264_decode_seq_parameter_set(H264Context *h);
673 * compute profile from sps
675 int ff_h264_get_profile(SPS *sps);
680 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
683 * Decode a network abstraction layer unit.
684 * @param consumed is the number of bytes used as input
685 * @param length is the length of the array
686 * @param dst_length is the number of decoded bytes FIXME here
687 * or a decode rbsp tailing?
688 * @return decoded bytes, might be src+1 if no escapes
690 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
691 int *dst_length, int *consumed, int length);
694 * Free any data that may have been allocated in the H264 context
697 void ff_h264_free_context(H264Context *h);
700 * Reconstruct bitstream slice_type.
702 int ff_h264_get_slice_type(const H264Context *h);
708 int ff_h264_alloc_tables(H264Context *h);
711 * Fill the default_ref_list.
713 int ff_h264_fill_default_ref_list(H264Context *h);
715 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
716 void ff_h264_fill_mbaff_ref_list(H264Context *h);
717 void ff_h264_remove_all_refs(H264Context *h);
720 * Execute the reference picture marking (memory management control operations).
722 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
724 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb,
727 int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
730 * Check if the top & left blocks are available if needed & change the
731 * dc mode so it only uses the available blocks.
733 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
736 * Check if the top & left blocks are available if needed & change the
737 * dc mode so it only uses the available blocks.
739 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma);
741 void ff_h264_hl_decode_mb(H264Context *h);
742 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size);
743 int ff_h264_decode_init(AVCodecContext *avctx);
744 void ff_h264_decode_init_vlc(void);
747 * Decode a macroblock
748 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
750 int ff_h264_decode_mb_cavlc(H264Context *h);
753 * Decode a CABAC coded macroblock
754 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
756 int ff_h264_decode_mb_cabac(H264Context *h);
758 void ff_h264_init_cabac_states(H264Context *h);
760 void ff_h264_direct_dist_scale_factor(H264Context *const h);
761 void ff_h264_direct_ref_list_init(H264Context *const h);
762 void ff_h264_pred_direct_motion(H264Context *const h, int *mb_type);
764 void ff_h264_filter_mb_fast(H264Context *h, int mb_x, int mb_y,
765 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
766 unsigned int linesize, unsigned int uvlinesize);
767 void ff_h264_filter_mb(H264Context *h, int mb_x, int mb_y,
768 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
769 unsigned int linesize, unsigned int uvlinesize);
772 * Reset SEI values at the beginning of the frame.
774 * @param h H.264 context.
776 void ff_h264_reset_sei(H264Context *h);
788 /* Scan8 organization:
805 * DY/DU/DV are for luma/chroma DC.
808 #define LUMA_DC_BLOCK_INDEX 48
809 #define CHROMA_DC_BLOCK_INDEX 49
811 // This table must be here because scan8[constant] must be known at compiletime
812 static const uint8_t scan8[16 * 3 + 3] = {
813 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
814 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
815 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
816 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
817 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
818 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
819 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
820 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
821 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
822 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
823 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
824 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
825 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
828 static av_always_inline uint32_t pack16to32(int a, int b)
831 return (b & 0xFFFF) + (a << 16);
833 return (a & 0xFFFF) + (b << 16);
837 static av_always_inline uint16_t pack8to16(int a, int b)
840 return (b & 0xFF) + (a << 8);
842 return (a & 0xFF) + (b << 8);
849 static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale)
851 return h->pps.chroma_qp_table[t][qscale];
855 * Get the predicted intra4x4 prediction mode.
857 static av_always_inline int pred_intra_mode(H264Context *h, int n)
859 const int index8 = scan8[n];
860 const int left = h->intra4x4_pred_mode_cache[index8 - 1];
861 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
862 const int min = FFMIN(left, top);
864 tprintf(h->avctx, "mode:%d %d min:%d\n", left, top, min);
872 static av_always_inline void write_back_intra_pred_mode(H264Context *h)
874 int8_t *i4x4 = h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
875 int8_t *i4x4_cache = h->intra4x4_pred_mode_cache;
877 AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
878 i4x4[4] = i4x4_cache[7 + 8 * 3];
879 i4x4[5] = i4x4_cache[7 + 8 * 2];
880 i4x4[6] = i4x4_cache[7 + 8 * 1];
883 static av_always_inline void write_back_non_zero_count(H264Context *h)
885 const int mb_xy = h->mb_xy;
886 uint8_t *nnz = h->non_zero_count[mb_xy];
887 uint8_t *nnz_cache = h->non_zero_count_cache;
889 AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
890 AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
891 AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
892 AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
893 AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
894 AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
895 AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
896 AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
898 if (!h->chroma_y_shift) {
899 AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
900 AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
901 AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
902 AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
906 static av_always_inline void write_back_motion_list(H264Context *h,
909 int mb_type, int list)
911 int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
912 int16_t(*mv_src)[2] = &h->mv_cache[list][scan8[0]];
913 AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
914 AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
915 AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
916 AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
918 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8 * h->mb_xy
919 : h->mb2br_xy[h->mb_xy]];
920 uint8_t(*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
921 if (IS_SKIP(mb_type)) {
924 AV_COPY64(mvd_dst, mvd_src + 8 * 3);
925 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
926 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
927 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
932 int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
933 int8_t *ref_cache = h->ref_cache[list];
934 ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
935 ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
936 ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
937 ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
941 static av_always_inline void write_back_motion(H264Context *h, int mb_type)
943 const int b_stride = h->b_stride;
944 const int b_xy = 4 * h->mb_x + 4 * h->mb_y * h->b_stride; // try mb2b(8)_xy
945 const int b8_xy = 4 * h->mb_xy;
947 if (USES_LIST(mb_type, 0)) {
948 write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 0);
950 fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
951 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
953 if (USES_LIST(mb_type, 1))
954 write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 1);
956 if (h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) {
957 if (IS_8X8(mb_type)) {
958 uint8_t *direct_table = &h->direct_table[4 * h->mb_xy];
959 direct_table[1] = h->sub_mb_type[1] >> 1;
960 direct_table[2] = h->sub_mb_type[2] >> 1;
961 direct_table[3] = h->sub_mb_type[3] >> 1;
966 static av_always_inline int get_dct8x8_allowed(H264Context *h)
968 if (h->sps.direct_8x8_inference_flag)
969 return !(AV_RN64A(h->sub_mb_type) &
970 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
971 0x0001000100010001ULL));
973 return !(AV_RN64A(h->sub_mb_type) &
974 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
975 0x0001000100010001ULL));
978 void ff_h264_draw_horiz_band(H264Context *h, int y, int height);
979 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc);
981 #endif /* AVCODEC_H264_H */