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)
77 #define FIELD_PICTURE 0
79 #define IS_INTERLACED(mb_type) 0
81 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
84 #define CABAC h->pps.cabac
87 #define CHROMA444 (h->sps.chroma_format_idc == 3)
89 #define EXTENDED_SAR 255
91 #define MB_TYPE_REF0 MB_TYPE_ACPRED //dirty but it fits in 16 bit
92 #define MB_TYPE_8x8DCT 0x01000000
93 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
94 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
97 * Value of Picture.reference when Picture is not a reference picture, but
98 * is held for delayed output.
100 #define DELAYED_PIC_REF 4
102 #define QP_MAX_NUM (51 + 2*6) // The maximum supported qp
119 NAL_AUXILIARY_SLICE=19
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_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
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
171 unsigned int crop_left; ///< frame_cropping_rect_left_offset
172 unsigned int crop_right; ///< frame_cropping_rect_right_offset
173 unsigned int crop_top; ///< frame_cropping_rect_top_offset
174 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
175 int vui_parameters_present_flag;
177 int video_signal_type_present_flag;
179 int colour_description_present_flag;
180 enum AVColorPrimaries color_primaries;
181 enum AVColorTransferCharacteristic color_trc;
182 enum AVColorSpace colorspace;
183 int timing_info_present_flag;
184 uint32_t num_units_in_tick;
186 int fixed_frame_rate_flag;
187 short offset_for_ref_frame[256]; //FIXME dyn aloc?
188 int bitstream_restriction_flag;
189 int num_reorder_frames;
190 int scaling_matrix_present;
191 uint8_t scaling_matrix4[6][16];
192 uint8_t scaling_matrix8[6][64];
193 int nal_hrd_parameters_present_flag;
194 int vcl_hrd_parameters_present_flag;
195 int pic_struct_present_flag;
196 int time_offset_length;
197 int cpb_cnt; ///< See H.264 E.1.2
198 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1
199 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
200 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
201 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
202 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
203 int residual_color_transform_flag; ///< residual_colour_transform_flag
204 int constraint_set_flags; ///< constraint_set[0-3]_flag
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][64]; ///< 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.
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{
259 H264DSPContext h264dsp;
260 int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
261 int chroma_qp[2]; //QPc
263 int qp_thresh; ///< QP threshold to skip loopfilter
269 int chroma_pred_mode;
270 int intra16x16_pred_mode;
282 const uint8_t * left_block;
283 int topleft_partition;
285 int8_t intra4x4_pred_mode_cache[5*8];
286 int8_t (*intra4x4_pred_mode);
288 unsigned int topleft_samples_available;
289 unsigned int top_samples_available;
290 unsigned int topright_samples_available;
291 unsigned int left_samples_available;
292 uint8_t (*top_borders[2])[(16*3)*2];
295 * non zero coeff count cache.
296 * is 64 if not available.
298 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15*8];
300 uint8_t (*non_zero_count)[48];
303 * Motion vector cache.
305 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5*8][2];
306 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5*8];
307 #define LIST_NOT_USED -1 //FIXME rename?
308 #define PART_NOT_AVAILABLE -2
311 * number of neighbors (top and/or left) that used 8x8 dct
313 int neighbor_transform_size;
316 * block_offset[ 0..23] for frame macroblocks
317 * block_offset[24..47] for field macroblocks
319 int block_offset[2*(16*3)];
321 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
323 int b_stride; //FIXME use s->b4_stride
325 int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff
331 SPS sps; ///< current sps
336 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
338 uint32_t dequant4_buffer[6][QP_MAX_NUM+1][16]; //FIXME should these be moved down?
339 uint32_t dequant8_buffer[6][QP_MAX_NUM+1][64];
340 uint32_t (*dequant4_coeff[6])[16];
341 uint32_t (*dequant8_coeff[6])[64];
344 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
346 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
347 int slice_type_fixed;
349 //interlacing specific flags
351 int mb_field_decoding_flag;
352 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
354 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
356 //Weighted pred stuff
358 int use_weight_chroma;
359 int luma_log2_weight_denom;
360 int chroma_log2_weight_denom;
361 //The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
362 int luma_weight[48][2][2];
363 int chroma_weight[48][2][2][2];
364 int implicit_weight[48][48][2];
366 int direct_spatial_mv_pred;
369 int dist_scale_factor[16];
370 int dist_scale_factor_field[2][32];
371 int map_col_to_list0[2][16+32];
372 int map_col_to_list0_field[2][2][16+32];
375 * num_ref_idx_l0/1_active_minus1 + 1
377 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
378 unsigned int list_count;
379 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
380 Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
381 Reordered version of default_ref_list
382 according to picture reordering in slice header */
383 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
386 GetBitContext intra_gb;
387 GetBitContext inter_gb;
388 GetBitContext *intra_gb_ptr;
389 GetBitContext *inter_gb_ptr;
391 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.
392 DECLARE_ALIGNED(16, DCTELEM, mb_luma_dc)[3][16*2];
393 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
399 uint8_t cabac_state[1024];
401 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
406 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
407 uint8_t *chroma_pred_mode_table;
408 int last_qscale_diff;
409 uint8_t (*mvd_table[2])[2];
410 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5*8][2];
411 uint8_t *direct_table;
412 uint8_t direct_cache[5*8];
414 uint8_t zigzag_scan[16];
415 uint8_t zigzag_scan8x8[64];
416 uint8_t zigzag_scan8x8_cavlc[64];
417 uint8_t field_scan[16];
418 uint8_t field_scan8x8[64];
419 uint8_t field_scan8x8_cavlc[64];
420 const uint8_t *zigzag_scan_q0;
421 const uint8_t *zigzag_scan8x8_q0;
422 const uint8_t *zigzag_scan8x8_cavlc_q0;
423 const uint8_t *field_scan_q0;
424 const uint8_t *field_scan8x8_q0;
425 const uint8_t *field_scan8x8_cavlc_q0;
434 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
435 int slice_alpha_c0_offset;
436 int slice_beta_offset;
438 //=============================================================
439 //Things below are not used in the MB or more inner code
443 uint8_t *rbsp_buffer[2];
444 unsigned int rbsp_buffer_size[2];
447 * Used to parse AVC variant of h264
449 int is_avc; ///< this flag is != 0 if codec is avc1
450 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
451 int got_first; ///< this flag is != 0 if we've parsed a frame
453 SPS *sps_buffers[MAX_SPS_COUNT];
454 PPS *pps_buffers[MAX_PPS_COUNT];
456 int dequant_coeff_pps; ///< reinit tables when pps changes
458 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 Picture *next_output_pic;
491 int next_outputed_poc;
494 * memory management control operations buffer.
496 MMCO mmco[MAX_MMCO_COUNT];
499 int long_ref_count; ///< number of actual long term references
500 int short_ref_count; ///< number of actual short term references
505 * @name Members for slice based multithreading
508 struct H264Context *thread_context[MAX_THREADS];
511 * current slice number, used to initalize slice_num of each thread/context
516 * Max number of threads / contexts.
517 * This is equal to AVCodecContext.thread_count unless
518 * multithreaded decoding is impossible, in which case it is
524 * 1 if the single thread fallback warning has already been
525 * displayed, 0 otherwise.
527 int single_decode_warning;
533 * pic_struct in picture timing SEI message
535 SEI_PicStructType sei_pic_struct;
538 * Complement sei_pic_struct
539 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
540 * However, soft telecined frames may have these values.
541 * This is used in an attempt to flag soft telecine progressive.
543 int prev_interlaced_frame;
546 * Bit set of clock types for fields/frames in picture timing SEI message.
547 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
553 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
555 int sei_dpb_output_delay;
558 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
560 int sei_cpb_removal_delay;
563 * recovery_frame_cnt from SEI message
565 * Set to -1 if no recovery point SEI message found or to number of frames
566 * before playback synchronizes. Frames having recovery point are key
569 int sei_recovery_frame_cnt;
571 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
572 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
575 int sei_buffering_period_present; ///< Buffering period SEI flag
576 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
580 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).
585 int ff_h264_decode_sei(H264Context *h);
590 int ff_h264_decode_seq_parameter_set(H264Context *h);
593 * compute profile from sps
595 int ff_h264_get_profile(SPS *sps);
600 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
603 * Decode a network abstraction layer unit.
604 * @param consumed is the number of bytes used as input
605 * @param length is the length of the array
606 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
607 * @return decoded bytes, might be src+1 if no escapes
609 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
612 * Free any data that may have been allocated in the H264 context like SPS, PPS etc.
614 av_cold void ff_h264_free_context(H264Context *h);
617 * Reconstruct bitstream slice_type.
619 int ff_h264_get_slice_type(const H264Context *h);
625 int ff_h264_alloc_tables(H264Context *h);
628 * Fill the default_ref_list.
630 int ff_h264_fill_default_ref_list(H264Context *h);
632 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
633 void ff_h264_fill_mbaff_ref_list(H264Context *h);
634 void ff_h264_remove_all_refs(H264Context *h);
637 * Execute the reference picture marking (memory management control operations).
639 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
641 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
643 void ff_generate_sliding_window_mmcos(H264Context *h);
647 * Check if the top & left blocks are available if needed & change the dc mode so it only uses the available blocks.
649 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
652 * Check if the top & left blocks are available if needed & change the dc mode so it only uses the available blocks.
654 int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
656 void ff_h264_hl_decode_mb(H264Context *h);
657 int ff_h264_frame_start(H264Context *h);
658 int ff_h264_decode_extradata(H264Context *h);
659 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
660 av_cold int ff_h264_decode_end(AVCodecContext *avctx);
661 av_cold void ff_h264_decode_init_vlc(void);
664 * Decode a macroblock
665 * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
667 int ff_h264_decode_mb_cavlc(H264Context *h);
670 * Decode a CABAC coded macroblock
671 * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
673 int ff_h264_decode_mb_cabac(H264Context *h);
675 void ff_h264_init_cabac_states(H264Context *h);
677 void ff_h264_direct_dist_scale_factor(H264Context * const h);
678 void ff_h264_direct_ref_list_init(H264Context * const h);
679 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
681 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);
682 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);
685 * Reset SEI values at the beginning of the frame.
687 * @param h H.264 context.
689 void ff_h264_reset_sei(H264Context *h);
702 /* Scan8 organization:
719 * DY/DU/DV are for luma/chroma DC.
722 #define LUMA_DC_BLOCK_INDEX 48
723 #define CHROMA_DC_BLOCK_INDEX 49
725 //This table must be here because scan8[constant] must be known at compiletime
726 static const uint8_t scan8[16*3 + 3]={
727 4+ 1*8, 5+ 1*8, 4+ 2*8, 5+ 2*8,
728 6+ 1*8, 7+ 1*8, 6+ 2*8, 7+ 2*8,
729 4+ 3*8, 5+ 3*8, 4+ 4*8, 5+ 4*8,
730 6+ 3*8, 7+ 3*8, 6+ 4*8, 7+ 4*8,
731 4+ 6*8, 5+ 6*8, 4+ 7*8, 5+ 7*8,
732 6+ 6*8, 7+ 6*8, 6+ 7*8, 7+ 7*8,
733 4+ 8*8, 5+ 8*8, 4+ 9*8, 5+ 9*8,
734 6+ 8*8, 7+ 8*8, 6+ 9*8, 7+ 9*8,
735 4+11*8, 5+11*8, 4+12*8, 5+12*8,
736 6+11*8, 7+11*8, 6+12*8, 7+12*8,
737 4+13*8, 5+13*8, 4+14*8, 5+14*8,
738 6+13*8, 7+13*8, 6+14*8, 7+14*8,
739 0+ 0*8, 0+ 5*8, 0+10*8
742 static av_always_inline uint32_t pack16to32(int a, int b){
744 return (b&0xFFFF) + (a<<16);
746 return (a&0xFFFF) + (b<<16);
750 static av_always_inline uint16_t pack8to16(int a, int b){
752 return (b&0xFF) + (a<<8);
754 return (a&0xFF) + (b<<8);
759 * gets the chroma qp.
761 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
762 return h->pps.chroma_qp_table[t][qscale];
765 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
767 static void fill_decode_neighbors(H264Context *h, int mb_type){
768 MpegEncContext * const s = &h->s;
769 const int mb_xy= h->mb_xy;
770 int topleft_xy, top_xy, topright_xy, left_xy[2];
771 static const uint8_t left_block_options[4][32]={
772 {0,1,2,3,7,10,8,11,3+0*4, 3+1*4, 3+2*4, 3+3*4, 1+4*4, 1+8*4, 1+5*4, 1+9*4},
773 {2,2,3,3,8,11,8,11,3+2*4, 3+2*4, 3+3*4, 3+3*4, 1+5*4, 1+9*4, 1+5*4, 1+9*4},
774 {0,0,1,1,7,10,7,10,3+0*4, 3+0*4, 3+1*4, 3+1*4, 1+4*4, 1+8*4, 1+4*4, 1+8*4},
775 {0,2,0,2,7,10,7,10,3+0*4, 3+2*4, 3+0*4, 3+2*4, 1+4*4, 1+8*4, 1+4*4, 1+8*4}
778 h->topleft_partition= -1;
780 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
782 /* Wow, what a mess, why didn't they simplify the interlacing & intra
783 * stuff, I can't imagine that these complex rules are worth it. */
785 topleft_xy = top_xy - 1;
786 topright_xy= top_xy + 1;
787 left_xy[1] = left_xy[0] = mb_xy-1;
788 h->left_block = left_block_options[0];
790 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
791 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
793 if (left_mb_field_flag != curr_mb_field_flag) {
794 left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
795 if (curr_mb_field_flag) {
796 left_xy[1] += s->mb_stride;
797 h->left_block = left_block_options[3];
799 topleft_xy += s->mb_stride;
800 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
801 h->topleft_partition = 0;
802 h->left_block = left_block_options[1];
806 if(curr_mb_field_flag){
807 topleft_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
808 topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
809 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
811 if (left_mb_field_flag != curr_mb_field_flag) {
812 if (curr_mb_field_flag) {
813 left_xy[1] += s->mb_stride;
814 h->left_block = left_block_options[3];
816 h->left_block = left_block_options[2];
822 h->topleft_mb_xy = topleft_xy;
823 h->top_mb_xy = top_xy;
824 h->topright_mb_xy= topright_xy;
825 h->left_mb_xy[0] = left_xy[0];
826 h->left_mb_xy[1] = left_xy[1];
827 //FIXME do we need all in the context?
829 h->topleft_type = s->current_picture.mb_type[topleft_xy] ;
830 h->top_type = s->current_picture.mb_type[top_xy] ;
831 h->topright_type= s->current_picture.mb_type[topright_xy];
832 h->left_type[0] = s->current_picture.mb_type[left_xy[0]] ;
833 h->left_type[1] = s->current_picture.mb_type[left_xy[1]] ;
836 if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0;
837 if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0;
838 if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
840 if(h->slice_table[topleft_xy ] != h->slice_num){
842 if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0;
843 if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
846 if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
849 static void fill_decode_caches(H264Context *h, int mb_type){
850 MpegEncContext * const s = &h->s;
851 int topleft_xy, top_xy, topright_xy, left_xy[2];
852 int topleft_type, top_type, topright_type, left_type[2];
853 const uint8_t * left_block= h->left_block;
858 topleft_xy = h->topleft_mb_xy ;
859 top_xy = h->top_mb_xy ;
860 topright_xy = h->topright_mb_xy;
861 left_xy[0] = h->left_mb_xy[0] ;
862 left_xy[1] = h->left_mb_xy[1] ;
863 topleft_type = h->topleft_type ;
864 top_type = h->top_type ;
865 topright_type= h->topright_type ;
866 left_type[0] = h->left_type[0] ;
867 left_type[1] = h->left_type[1] ;
869 if(!IS_SKIP(mb_type)){
870 if(IS_INTRA(mb_type)){
871 int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
872 h->topleft_samples_available=
873 h->top_samples_available=
874 h->left_samples_available= 0xFFFF;
875 h->topright_samples_available= 0xEEEA;
877 if(!(top_type & type_mask)){
878 h->topleft_samples_available= 0xB3FF;
879 h->top_samples_available= 0x33FF;
880 h->topright_samples_available= 0x26EA;
882 if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
883 if(IS_INTERLACED(mb_type)){
884 if(!(left_type[0] & type_mask)){
885 h->topleft_samples_available&= 0xDFFF;
886 h->left_samples_available&= 0x5FFF;
888 if(!(left_type[1] & type_mask)){
889 h->topleft_samples_available&= 0xFF5F;
890 h->left_samples_available&= 0xFF5F;
893 int left_typei = s->current_picture.mb_type[left_xy[0] + s->mb_stride];
895 assert(left_xy[0] == left_xy[1]);
896 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
897 h->topleft_samples_available&= 0xDF5F;
898 h->left_samples_available&= 0x5F5F;
902 if(!(left_type[0] & type_mask)){
903 h->topleft_samples_available&= 0xDF5F;
904 h->left_samples_available&= 0x5F5F;
908 if(!(topleft_type & type_mask))
909 h->topleft_samples_available&= 0x7FFF;
911 if(!(topright_type & type_mask))
912 h->topright_samples_available&= 0xFBFF;
914 if(IS_INTRA4x4(mb_type)){
915 if(IS_INTRA4x4(top_type)){
916 AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]);
918 h->intra4x4_pred_mode_cache[4+8*0]=
919 h->intra4x4_pred_mode_cache[5+8*0]=
920 h->intra4x4_pred_mode_cache[6+8*0]=
921 h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask);
924 if(IS_INTRA4x4(left_type[i])){
925 int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]];
926 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]];
927 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]];
929 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
930 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[i] & type_mask);
945 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
946 nnz_cache = h->non_zero_count_cache;
948 nnz = h->non_zero_count[top_xy];
949 AV_COPY32(&nnz_cache[4+8* 0], &nnz[4*3]);
951 AV_COPY32(&nnz_cache[4+8* 5], &nnz[4* 7]);
952 AV_COPY32(&nnz_cache[4+8*10], &nnz[4*11]);
954 AV_COPY32(&nnz_cache[4+8* 5], &nnz[4* 5]);
955 AV_COPY32(&nnz_cache[4+8*10], &nnz[4* 9]);
958 uint32_t top_empty = CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040;
959 AV_WN32A(&nnz_cache[4+8* 0], top_empty);
960 AV_WN32A(&nnz_cache[4+8* 5], top_empty);
961 AV_WN32A(&nnz_cache[4+8*10], top_empty);
964 for (i=0; i<2; i++) {
966 nnz = h->non_zero_count[left_xy[i]];
967 nnz_cache[3+8* 1 + 2*8*i]= nnz[left_block[8+0+2*i]];
968 nnz_cache[3+8* 2 + 2*8*i]= nnz[left_block[8+1+2*i]];
970 nnz_cache[3+8* 6 + 2*8*i]= nnz[left_block[8+0+2*i]+4*4];
971 nnz_cache[3+8* 7 + 2*8*i]= nnz[left_block[8+1+2*i]+4*4];
972 nnz_cache[3+8*11 + 2*8*i]= nnz[left_block[8+0+2*i]+8*4];
973 nnz_cache[3+8*12 + 2*8*i]= nnz[left_block[8+1+2*i]+8*4];
975 nnz_cache[3+8* 6 + 8*i]= nnz[left_block[8+4+2*i]];
976 nnz_cache[3+8*11 + 8*i]= nnz[left_block[8+5+2*i]];
979 nnz_cache[3+8* 1 + 2*8*i]=
980 nnz_cache[3+8* 2 + 2*8*i]=
981 nnz_cache[3+8* 6 + 2*8*i]=
982 nnz_cache[3+8* 7 + 2*8*i]=
983 nnz_cache[3+8*11 + 2*8*i]=
984 nnz_cache[3+8*12 + 2*8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
991 h->top_cbp = h->cbp_table[top_xy];
993 h->top_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F;
997 h->left_cbp = (h->cbp_table[left_xy[0]] & 0x7F0)
998 | ((h->cbp_table[left_xy[0]]>>(left_block[0]&(~1)))&2)
999 | (((h->cbp_table[left_xy[1]]>>(left_block[2]&(~1)))&2) << 2);
1001 h->left_cbp = IS_INTRA(mb_type) ? 0x7CF : 0x00F;
1006 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
1008 int b_stride = h->b_stride;
1009 for(list=0; list<h->list_count; list++){
1010 int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
1011 int8_t *ref = s->current_picture.ref_index[list];
1012 int16_t (*mv_cache)[2] = &h->mv_cache[list][scan8[0]];
1013 int16_t (*mv)[2] = s->current_picture.motion_val[list];
1014 if(!USES_LIST(mb_type, list)){
1017 assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
1019 if(USES_LIST(top_type, list)){
1020 const int b_xy= h->mb2b_xy[top_xy] + 3*b_stride;
1021 AV_COPY128(mv_cache[0 - 1*8], mv[b_xy + 0]);
1023 ref_cache[1 - 1*8]= ref[4*top_xy + 2];
1025 ref_cache[3 - 1*8]= ref[4*top_xy + 3];
1027 AV_ZERO128(mv_cache[0 - 1*8]);
1028 AV_WN32A(&ref_cache[0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
1031 if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){
1033 int cache_idx = -1 + i*2*8;
1034 if(USES_LIST(left_type[i], list)){
1035 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1036 const int b8_xy= 4*left_xy[i] + 1;
1037 AV_COPY32(mv_cache[cache_idx ], mv[b_xy + b_stride*left_block[0+i*2]]);
1038 AV_COPY32(mv_cache[cache_idx+8], mv[b_xy + b_stride*left_block[1+i*2]]);
1039 ref_cache[cache_idx ]= ref[b8_xy + (left_block[0+i*2]&~1)];
1040 ref_cache[cache_idx+8]= ref[b8_xy + (left_block[1+i*2]&~1)];
1042 AV_ZERO32(mv_cache[cache_idx ]);
1043 AV_ZERO32(mv_cache[cache_idx+8]);
1044 ref_cache[cache_idx ]=
1045 ref_cache[cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1049 if(USES_LIST(left_type[0], list)){
1050 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1051 const int b8_xy= 4*left_xy[0] + 1;
1052 AV_COPY32(mv_cache[-1], mv[b_xy + b_stride*left_block[0]]);
1053 ref_cache[-1]= ref[b8_xy + (left_block[0]&~1)];
1055 AV_ZERO32(mv_cache[-1]);
1056 ref_cache[-1]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1060 if(USES_LIST(topright_type, list)){
1061 const int b_xy= h->mb2b_xy[topright_xy] + 3*b_stride;
1062 AV_COPY32(mv_cache[4 - 1*8], mv[b_xy]);
1063 ref_cache[4 - 1*8]= ref[4*topright_xy + 2];
1065 AV_ZERO32(mv_cache[4 - 1*8]);
1066 ref_cache[4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1068 if(ref_cache[4 - 1*8] < 0){
1069 if(USES_LIST(topleft_type, list)){
1070 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + b_stride + (h->topleft_partition & 2*b_stride);
1071 const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2);
1072 AV_COPY32(mv_cache[-1 - 1*8], mv[b_xy]);
1073 ref_cache[-1 - 1*8]= ref[b8_xy];
1075 AV_ZERO32(mv_cache[-1 - 1*8]);
1076 ref_cache[-1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1080 if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
1083 if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))){
1084 uint8_t (*mvd_cache)[2] = &h->mvd_cache[list][scan8[0]];
1085 uint8_t (*mvd)[2] = h->mvd_table[list];
1087 ref_cache[2+8*2] = PART_NOT_AVAILABLE;
1088 AV_ZERO32(mv_cache[2+8*0]);
1089 AV_ZERO32(mv_cache[2+8*2]);
1092 if(USES_LIST(top_type, list)){
1093 const int b_xy= h->mb2br_xy[top_xy];
1094 AV_COPY64(mvd_cache[0 - 1*8], mvd[b_xy + 0]);
1096 AV_ZERO64(mvd_cache[0 - 1*8]);
1098 if(USES_LIST(left_type[0], list)){
1099 const int b_xy= h->mb2br_xy[left_xy[0]] + 6;
1100 AV_COPY16(mvd_cache[-1 + 0*8], mvd[b_xy - left_block[0]]);
1101 AV_COPY16(mvd_cache[-1 + 1*8], mvd[b_xy - left_block[1]]);
1103 AV_ZERO16(mvd_cache[-1 + 0*8]);
1104 AV_ZERO16(mvd_cache[-1 + 1*8]);
1106 if(USES_LIST(left_type[1], list)){
1107 const int b_xy= h->mb2br_xy[left_xy[1]] + 6;
1108 AV_COPY16(mvd_cache[-1 + 2*8], mvd[b_xy - left_block[2]]);
1109 AV_COPY16(mvd_cache[-1 + 3*8], mvd[b_xy - left_block[3]]);
1111 AV_ZERO16(mvd_cache[-1 + 2*8]);
1112 AV_ZERO16(mvd_cache[-1 + 3*8]);
1114 AV_ZERO16(mvd_cache[2+8*0]);
1115 AV_ZERO16(mvd_cache[2+8*2]);
1116 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
1117 uint8_t *direct_cache = &h->direct_cache[scan8[0]];
1118 uint8_t *direct_table = h->direct_table;
1119 fill_rectangle(direct_cache, 4, 4, 8, MB_TYPE_16x16>>1, 1);
1121 if(IS_DIRECT(top_type)){
1122 AV_WN32A(&direct_cache[-1*8], 0x01010101u*(MB_TYPE_DIRECT2>>1));
1123 }else if(IS_8X8(top_type)){
1124 int b8_xy = 4*top_xy;
1125 direct_cache[0 - 1*8]= direct_table[b8_xy + 2];
1126 direct_cache[2 - 1*8]= direct_table[b8_xy + 3];
1128 AV_WN32A(&direct_cache[-1*8], 0x01010101*(MB_TYPE_16x16>>1));
1131 if(IS_DIRECT(left_type[0]))
1132 direct_cache[-1 + 0*8]= MB_TYPE_DIRECT2>>1;
1133 else if(IS_8X8(left_type[0]))
1134 direct_cache[-1 + 0*8]= direct_table[4*left_xy[0] + 1 + (left_block[0]&~1)];
1136 direct_cache[-1 + 0*8]= MB_TYPE_16x16>>1;
1138 if(IS_DIRECT(left_type[1]))
1139 direct_cache[-1 + 2*8]= MB_TYPE_DIRECT2>>1;
1140 else if(IS_8X8(left_type[1]))
1141 direct_cache[-1 + 2*8]= direct_table[4*left_xy[1] + 1 + (left_block[2]&~1)];
1143 direct_cache[-1 + 2*8]= MB_TYPE_16x16>>1;
1149 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1150 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1151 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1152 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1153 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1154 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1155 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1156 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1157 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1158 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1160 #define MAP_F2F(idx, mb_type)\
1161 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1162 h->ref_cache[list][idx] <<= 1;\
1163 h->mv_cache[list][idx][1] /= 2;\
1164 h->mvd_cache[list][idx][1] >>=1;\
1169 #define MAP_F2F(idx, mb_type)\
1170 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1171 h->ref_cache[list][idx] >>= 1;\
1172 h->mv_cache[list][idx][1] <<= 1;\
1173 h->mvd_cache[list][idx][1] <<= 1;\
1182 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1186 * gets the predicted intra4x4 prediction mode.
1188 static av_always_inline int pred_intra_mode(H264Context *h, int n){
1189 const int index8= scan8[n];
1190 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1191 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1192 const int min= FFMIN(left, top);
1194 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1196 if(min<0) return DC_PRED;
1200 static av_always_inline void write_back_intra_pred_mode(H264Context *h){
1201 int8_t *i4x4= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
1202 int8_t *i4x4_cache= h->intra4x4_pred_mode_cache;
1204 AV_COPY32(i4x4, i4x4_cache + 4 + 8*4);
1205 i4x4[4]= i4x4_cache[7+8*3];
1206 i4x4[5]= i4x4_cache[7+8*2];
1207 i4x4[6]= i4x4_cache[7+8*1];
1210 static av_always_inline void write_back_non_zero_count(H264Context *h){
1211 const int mb_xy= h->mb_xy;
1212 uint8_t *nnz = h->non_zero_count[mb_xy];
1213 uint8_t *nnz_cache = h->non_zero_count_cache;
1215 AV_COPY32(&nnz[ 0], &nnz_cache[4+8* 1]);
1216 AV_COPY32(&nnz[ 4], &nnz_cache[4+8* 2]);
1217 AV_COPY32(&nnz[ 8], &nnz_cache[4+8* 3]);
1218 AV_COPY32(&nnz[12], &nnz_cache[4+8* 4]);
1219 AV_COPY32(&nnz[16], &nnz_cache[4+8* 6]);
1220 AV_COPY32(&nnz[20], &nnz_cache[4+8* 7]);
1221 AV_COPY32(&nnz[32], &nnz_cache[4+8*11]);
1222 AV_COPY32(&nnz[36], &nnz_cache[4+8*12]);
1225 AV_COPY32(&nnz[24], &nnz_cache[4+8* 8]);
1226 AV_COPY32(&nnz[28], &nnz_cache[4+8* 9]);
1227 AV_COPY32(&nnz[40], &nnz_cache[4+8*13]);
1228 AV_COPY32(&nnz[44], &nnz_cache[4+8*14]);
1232 static av_always_inline void write_back_motion_list(H264Context *h, MpegEncContext * const s, int b_stride,
1233 int b_xy, int b8_xy, int mb_type, int list )
1235 int16_t (*mv_dst)[2] = &s->current_picture.motion_val[list][b_xy];
1236 int16_t (*mv_src)[2] = &h->mv_cache[list][scan8[0]];
1237 AV_COPY128(mv_dst + 0*b_stride, mv_src + 8*0);
1238 AV_COPY128(mv_dst + 1*b_stride, mv_src + 8*1);
1239 AV_COPY128(mv_dst + 2*b_stride, mv_src + 8*2);
1240 AV_COPY128(mv_dst + 3*b_stride, mv_src + 8*3);
1242 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
1243 uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1244 if(IS_SKIP(mb_type))
1245 AV_ZERO128(mvd_dst);
1247 AV_COPY64(mvd_dst, mvd_src + 8*3);
1248 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
1249 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
1250 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
1255 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1256 int8_t *ref_cache = h->ref_cache[list];
1257 ref_index[0+0*2]= ref_cache[scan8[0]];
1258 ref_index[1+0*2]= ref_cache[scan8[4]];
1259 ref_index[0+1*2]= ref_cache[scan8[8]];
1260 ref_index[1+1*2]= ref_cache[scan8[12]];
1264 static av_always_inline void write_back_motion(H264Context *h, int mb_type){
1265 MpegEncContext * const s = &h->s;
1266 const int b_stride = h->b_stride;
1267 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
1268 const int b8_xy= 4*h->mb_xy;
1270 if(USES_LIST(mb_type, 0)){
1271 write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 0);
1273 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1275 if(USES_LIST(mb_type, 1)){
1276 write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 1);
1279 if(h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC){
1280 if(IS_8X8(mb_type)){
1281 uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
1282 direct_table[1] = h->sub_mb_type[1]>>1;
1283 direct_table[2] = h->sub_mb_type[2]>>1;
1284 direct_table[3] = h->sub_mb_type[3]>>1;
1289 static av_always_inline int get_dct8x8_allowed(H264Context *h){
1290 if(h->sps.direct_8x8_inference_flag)
1291 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL));
1293 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1297 * decodes a P_SKIP or B_SKIP macroblock
1299 static void av_unused decode_mb_skip(H264Context *h){
1300 MpegEncContext * const s = &h->s;
1301 const int mb_xy= h->mb_xy;
1304 memset(h->non_zero_count[mb_xy], 0, 48);
1307 mb_type|= MB_TYPE_INTERLACED;
1309 if( h->slice_type_nos == AV_PICTURE_TYPE_B )
1311 // just for fill_caches. pred_direct_motion will set the real mb_type
1312 mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1313 if(h->direct_spatial_mv_pred){
1314 fill_decode_neighbors(h, mb_type);
1315 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1317 ff_h264_pred_direct_motion(h, &mb_type);
1318 mb_type|= MB_TYPE_SKIP;
1323 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1325 fill_decode_neighbors(h, mb_type);
1326 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1327 pred_pskip_motion(h, &mx, &my);
1328 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1329 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1332 write_back_motion(h, mb_type);
1333 s->current_picture.mb_type[mb_xy]= mb_type;
1334 s->current_picture.qscale_table[mb_xy]= s->qscale;
1335 h->slice_table[ mb_xy ]= h->slice_num;
1336 h->prev_mb_skipped= 1;
1339 #include "h264_mvpred.h" //For pred_pskip_motion()
1341 #endif /* AVCODEC_H264_H */