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 CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
43 #define COEFF_TOKEN_VLC_BITS 8
44 #define TOTAL_ZEROS_VLC_BITS 9
45 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
46 #define RUN_VLC_BITS 3
47 #define RUN7_VLC_BITS 6
49 #define MAX_SPS_COUNT 32
50 #define MAX_PPS_COUNT 256
52 #define MAX_MMCO_COUNT 66
54 #define MAX_DELAYED_PIC_COUNT 16
56 /* Compiling in interlaced support reduces the speed
57 * of progressive decoding by about 2%. */
58 #define ALLOW_INTERLACE
63 * The maximum number of slices supported by the decoder.
64 * must be a power of 2
68 #ifdef ALLOW_INTERLACE
69 #define MB_MBAFF h->mb_mbaff
70 #define MB_FIELD h->mb_field_decoding_flag
71 #define FRAME_MBAFF h->mb_aff_frame
72 #define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
81 #define FIELD_PICTURE 0
83 #define IS_INTERLACED(mb_type) 0
89 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
92 #define CABAC h->pps.cabac
95 #define CHROMA444 (h->sps.chroma_format_idc == 3)
97 #define EXTENDED_SAR 255
99 #define MB_TYPE_REF0 MB_TYPE_ACPRED //dirty but it fits in 16 bit
100 #define MB_TYPE_8x8DCT 0x01000000
101 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
102 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
105 * Value of Picture.reference when Picture is not a reference picture, but
106 * is held for delayed output.
108 #define DELAYED_PIC_REF 4
110 #define QP_MAX_NUM (51 + 2*6) // The maximum supported qp
127 NAL_AUXILIARY_SLICE=19
134 SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
135 SEI_TYPE_PIC_TIMING = 1, ///< picture timing
136 SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
137 SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync)
141 * pic_struct in picture timing SEI message
144 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
145 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
146 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
147 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
148 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
149 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
150 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
151 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
152 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
156 * Sequence parameter set
162 int chroma_format_idc;
163 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
164 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
165 int poc_type; ///< pic_order_cnt_type
166 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
167 int delta_pic_order_always_zero_flag;
168 int offset_for_non_ref_pic;
169 int offset_for_top_to_bottom_field;
170 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
171 int ref_frame_count; ///< num_ref_frames
172 int gaps_in_frame_num_allowed_flag;
173 int mb_width; ///< pic_width_in_mbs_minus1 + 1
174 int mb_height; ///< pic_height_in_map_units_minus1 + 1
175 int frame_mbs_only_flag;
176 int mb_aff; ///<mb_adaptive_frame_field_flag
177 int direct_8x8_inference_flag;
178 int crop; ///< frame_cropping_flag
179 unsigned int crop_left; ///< frame_cropping_rect_left_offset
180 unsigned int crop_right; ///< frame_cropping_rect_right_offset
181 unsigned int crop_top; ///< frame_cropping_rect_top_offset
182 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
183 int vui_parameters_present_flag;
185 int video_signal_type_present_flag;
187 int colour_description_present_flag;
188 enum AVColorPrimaries color_primaries;
189 enum AVColorTransferCharacteristic color_trc;
190 enum AVColorSpace colorspace;
191 int timing_info_present_flag;
192 uint32_t num_units_in_tick;
194 int fixed_frame_rate_flag;
195 short offset_for_ref_frame[256]; //FIXME dyn aloc?
196 int bitstream_restriction_flag;
197 int num_reorder_frames;
198 int scaling_matrix_present;
199 uint8_t scaling_matrix4[6][16];
200 uint8_t scaling_matrix8[6][64];
201 int nal_hrd_parameters_present_flag;
202 int vcl_hrd_parameters_present_flag;
203 int pic_struct_present_flag;
204 int time_offset_length;
205 int cpb_cnt; ///< See H.264 E.1.2
206 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1
207 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
208 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
209 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
210 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
211 int residual_color_transform_flag; ///< residual_colour_transform_flag
212 int constraint_set_flags; ///< constraint_set[0-3]_flag
216 * Picture parameter set
220 int cabac; ///< entropy_coding_mode_flag
221 int pic_order_present; ///< pic_order_present_flag
222 int slice_group_count; ///< num_slice_groups_minus1 + 1
223 int mb_slice_group_map_type;
224 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
225 int weighted_pred; ///< weighted_pred_flag
226 int weighted_bipred_idc;
227 int init_qp; ///< pic_init_qp_minus26 + 26
228 int init_qs; ///< pic_init_qs_minus26 + 26
229 int chroma_qp_index_offset[2];
230 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
231 int constrained_intra_pred; ///< constrained_intra_pred_flag
232 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
233 int transform_8x8_mode; ///< transform_8x8_mode_flag
234 uint8_t scaling_matrix4[6][16];
235 uint8_t scaling_matrix8[6][64];
236 uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
241 * Memory management control operation opcode.
243 typedef enum MMCOOpcode{
254 * Memory management control operation.
258 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
259 int long_arg; ///< index, pic_num, or num long refs depending on opcode
265 typedef struct H264Context{
267 H264DSPContext h264dsp;
268 int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
269 int chroma_qp[2]; //QPc
271 int qp_thresh; ///< QP threshold to skip loopfilter
277 int chroma_pred_mode;
278 int intra16x16_pred_mode;
283 int left_mb_xy[LEFT_MBS];
288 int left_type[LEFT_MBS];
290 const uint8_t * left_block;
291 int topleft_partition;
293 int8_t intra4x4_pred_mode_cache[5*8];
294 int8_t (*intra4x4_pred_mode);
296 unsigned int topleft_samples_available;
297 unsigned int top_samples_available;
298 unsigned int topright_samples_available;
299 unsigned int left_samples_available;
300 uint8_t (*top_borders[2])[(16*3)*2];
303 * non zero coeff count cache.
304 * is 64 if not available.
306 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15*8];
308 uint8_t (*non_zero_count)[48];
311 * Motion vector cache.
313 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5*8][2];
314 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5*8];
315 #define LIST_NOT_USED -1 //FIXME rename?
316 #define PART_NOT_AVAILABLE -2
319 * number of neighbors (top and/or left) that used 8x8 dct
321 int neighbor_transform_size;
324 * block_offset[ 0..23] for frame macroblocks
325 * block_offset[24..47] for field macroblocks
327 int block_offset[2*(16*3)];
329 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
331 int b_stride; //FIXME use s->b4_stride
333 int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff
339 SPS sps; ///< current sps
344 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
346 uint32_t dequant4_buffer[6][QP_MAX_NUM+1][16]; //FIXME should these be moved down?
347 uint32_t dequant8_buffer[6][QP_MAX_NUM+1][64];
348 uint32_t (*dequant4_coeff[6])[16];
349 uint32_t (*dequant8_coeff[6])[64];
352 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
354 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
355 int slice_type_fixed;
357 //interlacing specific flags
359 int mb_field_decoding_flag;
360 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
362 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
364 //Weighted pred stuff
366 int use_weight_chroma;
367 int luma_log2_weight_denom;
368 int chroma_log2_weight_denom;
369 //The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
370 int luma_weight[48][2][2];
371 int chroma_weight[48][2][2][2];
372 int implicit_weight[48][48][2];
374 int direct_spatial_mv_pred;
377 int dist_scale_factor[16];
378 int dist_scale_factor_field[2][32];
379 int map_col_to_list0[2][16+32];
380 int map_col_to_list0_field[2][2][16+32];
383 * num_ref_idx_l0/1_active_minus1 + 1
385 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
386 unsigned int list_count;
387 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
388 Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
389 Reordered version of default_ref_list
390 according to picture reordering in slice header */
391 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
394 GetBitContext intra_gb;
395 GetBitContext inter_gb;
396 GetBitContext *intra_gb_ptr;
397 GetBitContext *inter_gb_ptr;
399 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.
400 DECLARE_ALIGNED(16, DCTELEM, mb_luma_dc)[3][16*2];
401 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
407 uint8_t cabac_state[1024];
409 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
414 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
415 uint8_t *chroma_pred_mode_table;
416 int last_qscale_diff;
417 uint8_t (*mvd_table[2])[2];
418 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5*8][2];
419 uint8_t *direct_table;
420 uint8_t direct_cache[5*8];
422 uint8_t zigzag_scan[16];
423 uint8_t zigzag_scan8x8[64];
424 uint8_t zigzag_scan8x8_cavlc[64];
425 uint8_t field_scan[16];
426 uint8_t field_scan8x8[64];
427 uint8_t field_scan8x8_cavlc[64];
428 const uint8_t *zigzag_scan_q0;
429 const uint8_t *zigzag_scan8x8_q0;
430 const uint8_t *zigzag_scan8x8_cavlc_q0;
431 const uint8_t *field_scan_q0;
432 const uint8_t *field_scan8x8_q0;
433 const uint8_t *field_scan8x8_cavlc_q0;
442 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
443 int slice_alpha_c0_offset;
444 int slice_beta_offset;
446 //=============================================================
447 //Things below are not used in the MB or more inner code
451 uint8_t *rbsp_buffer[2];
452 unsigned int rbsp_buffer_size[2];
455 * Used to parse AVC variant of h264
457 int is_avc; ///< this flag is != 0 if codec is avc1
458 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
459 int got_first; ///< this flag is != 0 if we've parsed a frame
461 SPS *sps_buffers[MAX_SPS_COUNT];
462 PPS *pps_buffers[MAX_PPS_COUNT];
464 int dequant_coeff_pps; ///< reinit tables when pps changes
466 uint16_t *slice_table_base;
472 int delta_poc_bottom;
475 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
476 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
477 int frame_num_offset; ///< for POC type 2
478 int prev_frame_num_offset; ///< for POC type 2
479 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
482 * frame_num for frames or 2*frame_num+1 for field pics.
487 * max_frame_num or 2*max_frame_num for field pics.
491 int redundant_pic_count;
493 Picture *short_ref[32];
494 Picture *long_ref[32];
495 Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
496 Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
497 Picture *next_output_pic;
499 int next_outputed_poc;
502 * memory management control operations buffer.
504 MMCO mmco[MAX_MMCO_COUNT];
507 int long_ref_count; ///< number of actual long term references
508 int short_ref_count; ///< number of actual short term references
513 * @name Members for slice based multithreading
516 struct H264Context *thread_context[MAX_THREADS];
519 * current slice number, used to initalize slice_num of each thread/context
524 * Max number of threads / contexts.
525 * This is equal to AVCodecContext.thread_count unless
526 * multithreaded decoding is impossible, in which case it is
532 * 1 if the single thread fallback warning has already been
533 * displayed, 0 otherwise.
535 int single_decode_warning;
541 * pic_struct in picture timing SEI message
543 SEI_PicStructType sei_pic_struct;
546 * Complement sei_pic_struct
547 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
548 * However, soft telecined frames may have these values.
549 * This is used in an attempt to flag soft telecine progressive.
551 int prev_interlaced_frame;
554 * Bit set of clock types for fields/frames in picture timing SEI message.
555 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
561 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
563 int sei_dpb_output_delay;
566 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
568 int sei_cpb_removal_delay;
571 * recovery_frame_cnt from SEI message
573 * Set to -1 if no recovery point SEI message found or to number of frames
574 * before playback synchronizes. Frames having recovery point are key
577 int sei_recovery_frame_cnt;
579 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
580 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
583 int sei_buffering_period_present; ///< Buffering period SEI flag
584 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
588 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).
593 int ff_h264_decode_sei(H264Context *h);
598 int ff_h264_decode_seq_parameter_set(H264Context *h);
601 * compute profile from sps
603 int ff_h264_get_profile(SPS *sps);
608 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
611 * Decode a network abstraction layer unit.
612 * @param consumed is the number of bytes used as input
613 * @param length is the length of the array
614 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
615 * @return decoded bytes, might be src+1 if no escapes
617 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
620 * Free any data that may have been allocated in the H264 context like SPS, PPS etc.
622 av_cold void ff_h264_free_context(H264Context *h);
625 * Reconstruct bitstream slice_type.
627 int ff_h264_get_slice_type(const H264Context *h);
633 int ff_h264_alloc_tables(H264Context *h);
636 * Fill the default_ref_list.
638 int ff_h264_fill_default_ref_list(H264Context *h);
640 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
641 void ff_h264_fill_mbaff_ref_list(H264Context *h);
642 void ff_h264_remove_all_refs(H264Context *h);
645 * Execute the reference picture marking (memory management control operations).
647 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
649 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
651 void ff_generate_sliding_window_mmcos(H264Context *h);
655 * Check if the top & left blocks are available if needed & change the dc mode so it only uses the available blocks.
657 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
660 * Check if the top & left blocks are available if needed & change the dc mode so it only uses the available blocks.
662 int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
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 int ff_h264_decode_end(AVCodecContext *avctx);
669 av_cold void ff_h264_decode_init_vlc(void);
672 * Decode a macroblock
673 * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
675 int ff_h264_decode_mb_cavlc(H264Context *h);
678 * Decode a CABAC coded macroblock
679 * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
681 int ff_h264_decode_mb_cabac(H264Context *h);
683 void ff_h264_init_cabac_states(H264Context *h);
685 void ff_h264_direct_dist_scale_factor(H264Context * const h);
686 void ff_h264_direct_ref_list_init(H264Context * const h);
687 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
689 void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
690 void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
693 * Reset SEI values at the beginning of the frame.
695 * @param h H.264 context.
697 void ff_h264_reset_sei(H264Context *h);
710 /* Scan8 organization:
727 * DY/DU/DV are for luma/chroma DC.
730 #define LUMA_DC_BLOCK_INDEX 48
731 #define CHROMA_DC_BLOCK_INDEX 49
733 //This table must be here because scan8[constant] must be known at compiletime
734 static const uint8_t scan8[16*3 + 3]={
735 4+ 1*8, 5+ 1*8, 4+ 2*8, 5+ 2*8,
736 6+ 1*8, 7+ 1*8, 6+ 2*8, 7+ 2*8,
737 4+ 3*8, 5+ 3*8, 4+ 4*8, 5+ 4*8,
738 6+ 3*8, 7+ 3*8, 6+ 4*8, 7+ 4*8,
739 4+ 6*8, 5+ 6*8, 4+ 7*8, 5+ 7*8,
740 6+ 6*8, 7+ 6*8, 6+ 7*8, 7+ 7*8,
741 4+ 8*8, 5+ 8*8, 4+ 9*8, 5+ 9*8,
742 6+ 8*8, 7+ 8*8, 6+ 9*8, 7+ 9*8,
743 4+11*8, 5+11*8, 4+12*8, 5+12*8,
744 6+11*8, 7+11*8, 6+12*8, 7+12*8,
745 4+13*8, 5+13*8, 4+14*8, 5+14*8,
746 6+13*8, 7+13*8, 6+14*8, 7+14*8,
747 0+ 0*8, 0+ 5*8, 0+10*8
750 static av_always_inline uint32_t pack16to32(int a, int b){
752 return (b&0xFFFF) + (a<<16);
754 return (a&0xFFFF) + (b<<16);
758 static av_always_inline uint16_t pack8to16(int a, int b){
760 return (b&0xFF) + (a<<8);
762 return (a&0xFF) + (b<<8);
767 * gets the chroma qp.
769 static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale){
770 return h->pps.chroma_qp_table[t][qscale];
774 * gets the predicted intra4x4 prediction mode.
776 static av_always_inline int pred_intra_mode(H264Context *h, int n){
777 const int index8= scan8[n];
778 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
779 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
780 const int min= FFMIN(left, top);
782 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
784 if(min<0) return DC_PRED;
788 static av_always_inline void write_back_intra_pred_mode(H264Context *h){
789 int8_t *i4x4= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
790 int8_t *i4x4_cache= h->intra4x4_pred_mode_cache;
792 AV_COPY32(i4x4, i4x4_cache + 4 + 8*4);
793 i4x4[4]= i4x4_cache[7+8*3];
794 i4x4[5]= i4x4_cache[7+8*2];
795 i4x4[6]= i4x4_cache[7+8*1];
798 static av_always_inline void write_back_non_zero_count(H264Context *h){
799 const int mb_xy= h->mb_xy;
800 uint8_t *nnz = h->non_zero_count[mb_xy];
801 uint8_t *nnz_cache = h->non_zero_count_cache;
803 AV_COPY32(&nnz[ 0], &nnz_cache[4+8* 1]);
804 AV_COPY32(&nnz[ 4], &nnz_cache[4+8* 2]);
805 AV_COPY32(&nnz[ 8], &nnz_cache[4+8* 3]);
806 AV_COPY32(&nnz[12], &nnz_cache[4+8* 4]);
807 AV_COPY32(&nnz[16], &nnz_cache[4+8* 6]);
808 AV_COPY32(&nnz[20], &nnz_cache[4+8* 7]);
809 AV_COPY32(&nnz[32], &nnz_cache[4+8*11]);
810 AV_COPY32(&nnz[36], &nnz_cache[4+8*12]);
813 AV_COPY32(&nnz[24], &nnz_cache[4+8* 8]);
814 AV_COPY32(&nnz[28], &nnz_cache[4+8* 9]);
815 AV_COPY32(&nnz[40], &nnz_cache[4+8*13]);
816 AV_COPY32(&nnz[44], &nnz_cache[4+8*14]);
820 static av_always_inline void write_back_motion_list(H264Context *h, MpegEncContext * const s, int b_stride,
821 int b_xy, int b8_xy, int mb_type, int list )
823 int16_t (*mv_dst)[2] = &s->current_picture.f.motion_val[list][b_xy];
824 int16_t (*mv_src)[2] = &h->mv_cache[list][scan8[0]];
825 AV_COPY128(mv_dst + 0*b_stride, mv_src + 8*0);
826 AV_COPY128(mv_dst + 1*b_stride, mv_src + 8*1);
827 AV_COPY128(mv_dst + 2*b_stride, mv_src + 8*2);
828 AV_COPY128(mv_dst + 3*b_stride, mv_src + 8*3);
830 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
831 uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
835 AV_COPY64(mvd_dst, mvd_src + 8*3);
836 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
837 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
838 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
843 int8_t *ref_index = &s->current_picture.f.ref_index[list][b8_xy];
844 int8_t *ref_cache = h->ref_cache[list];
845 ref_index[0+0*2]= ref_cache[scan8[0]];
846 ref_index[1+0*2]= ref_cache[scan8[4]];
847 ref_index[0+1*2]= ref_cache[scan8[8]];
848 ref_index[1+1*2]= ref_cache[scan8[12]];
852 static av_always_inline void write_back_motion(H264Context *h, int mb_type){
853 MpegEncContext * const s = &h->s;
854 const int b_stride = h->b_stride;
855 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
856 const int b8_xy= 4*h->mb_xy;
858 if(USES_LIST(mb_type, 0)){
859 write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 0);
861 fill_rectangle(&s->current_picture.f.ref_index[0][b8_xy],
862 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
864 if(USES_LIST(mb_type, 1)){
865 write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 1);
868 if(h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC){
870 uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
871 direct_table[1] = h->sub_mb_type[1]>>1;
872 direct_table[2] = h->sub_mb_type[2]>>1;
873 direct_table[3] = h->sub_mb_type[3]>>1;
878 static av_always_inline int get_dct8x8_allowed(H264Context *h){
879 if(h->sps.direct_8x8_inference_flag)
880 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL));
882 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
885 #endif /* AVCODEC_H264_H */