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 * is 1 if the specific list MV&references are set to 0,0,-2.
313 int mv_cache_clean[2];
316 * number of neighbors (top and/or left) that used 8x8 dct
318 int neighbor_transform_size;
321 * block_offset[ 0..23] for frame macroblocks
322 * block_offset[24..47] for field macroblocks
324 int block_offset[2*(16*3)];
326 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
328 int b_stride; //FIXME use s->b4_stride
330 int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff
336 SPS sps; ///< current sps
341 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
343 uint32_t dequant4_buffer[6][QP_MAX_NUM+1][16]; //FIXME should these be moved down?
344 uint32_t dequant8_buffer[6][QP_MAX_NUM+1][64];
345 uint32_t (*dequant4_coeff[6])[16];
346 uint32_t (*dequant8_coeff[6])[64];
349 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
351 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
352 int slice_type_fixed;
354 //interlacing specific flags
356 int mb_field_decoding_flag;
357 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
359 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
361 //Weighted pred stuff
363 int use_weight_chroma;
364 int luma_log2_weight_denom;
365 int chroma_log2_weight_denom;
366 //The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
367 int luma_weight[48][2][2];
368 int chroma_weight[48][2][2][2];
369 int implicit_weight[48][48][2];
371 int direct_spatial_mv_pred;
374 int dist_scale_factor[16];
375 int dist_scale_factor_field[2][32];
376 int map_col_to_list0[2][16+32];
377 int map_col_to_list0_field[2][2][16+32];
380 * num_ref_idx_l0/1_active_minus1 + 1
382 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
383 unsigned int list_count;
384 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
385 Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
386 Reordered version of default_ref_list
387 according to picture reordering in slice header */
388 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
391 GetBitContext intra_gb;
392 GetBitContext inter_gb;
393 GetBitContext *intra_gb_ptr;
394 GetBitContext *inter_gb_ptr;
396 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.
397 DECLARE_ALIGNED(16, DCTELEM, mb_luma_dc)[3][16*2];
398 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
404 uint8_t cabac_state[1024];
406 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
411 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
412 uint8_t *chroma_pred_mode_table;
413 int last_qscale_diff;
414 uint8_t (*mvd_table[2])[2];
415 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5*8][2];
416 uint8_t *direct_table;
417 uint8_t direct_cache[5*8];
419 uint8_t zigzag_scan[16];
420 uint8_t zigzag_scan8x8[64];
421 uint8_t zigzag_scan8x8_cavlc[64];
422 uint8_t field_scan[16];
423 uint8_t field_scan8x8[64];
424 uint8_t field_scan8x8_cavlc[64];
425 const uint8_t *zigzag_scan_q0;
426 const uint8_t *zigzag_scan8x8_q0;
427 const uint8_t *zigzag_scan8x8_cavlc_q0;
428 const uint8_t *field_scan_q0;
429 const uint8_t *field_scan8x8_q0;
430 const uint8_t *field_scan8x8_cavlc_q0;
439 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
440 int slice_alpha_c0_offset;
441 int slice_beta_offset;
443 //=============================================================
444 //Things below are not used in the MB or more inner code
448 uint8_t *rbsp_buffer[2];
449 unsigned int rbsp_buffer_size[2];
452 * Used to parse AVC variant of h264
454 int is_avc; ///< this flag is != 0 if codec is avc1
455 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
456 int got_first; ///< this flag is != 0 if we've parsed a frame
458 SPS *sps_buffers[MAX_SPS_COUNT];
459 PPS *pps_buffers[MAX_PPS_COUNT];
461 int dequant_coeff_pps; ///< reinit tables when pps changes
463 uint16_t *slice_table_base;
469 int delta_poc_bottom;
472 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
473 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
474 int frame_num_offset; ///< for POC type 2
475 int prev_frame_num_offset; ///< for POC type 2
476 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
479 * frame_num for frames or 2*frame_num+1 for field pics.
484 * max_frame_num or 2*max_frame_num for field pics.
488 int redundant_pic_count;
490 Picture *short_ref[32];
491 Picture *long_ref[32];
492 Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
493 Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
494 Picture *next_output_pic;
496 int next_outputed_poc;
499 * memory management control operations buffer.
501 MMCO mmco[MAX_MMCO_COUNT];
504 int long_ref_count; ///< number of actual long term references
505 int short_ref_count; ///< number of actual short term references
510 * @name Members for slice based multithreading
513 struct H264Context *thread_context[MAX_THREADS];
516 * current slice number, used to initalize slice_num of each thread/context
521 * Max number of threads / contexts.
522 * This is equal to AVCodecContext.thread_count unless
523 * multithreaded decoding is impossible, in which case it is
529 * 1 if the single thread fallback warning has already been
530 * displayed, 0 otherwise.
532 int single_decode_warning;
538 * pic_struct in picture timing SEI message
540 SEI_PicStructType sei_pic_struct;
543 * Complement sei_pic_struct
544 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
545 * However, soft telecined frames may have these values.
546 * This is used in an attempt to flag soft telecine progressive.
548 int prev_interlaced_frame;
551 * Bit set of clock types for fields/frames in picture timing SEI message.
552 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
558 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
560 int sei_dpb_output_delay;
563 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
565 int sei_cpb_removal_delay;
568 * recovery_frame_cnt from SEI message
570 * Set to -1 if no recovery point SEI message found or to number of frames
571 * before playback synchronizes. Frames having recovery point are key
574 int sei_recovery_frame_cnt;
576 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
577 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
580 int sei_buffering_period_present; ///< Buffering period SEI flag
581 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
585 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).
590 int ff_h264_decode_sei(H264Context *h);
595 int ff_h264_decode_seq_parameter_set(H264Context *h);
598 * compute profile from sps
600 int ff_h264_get_profile(SPS *sps);
605 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
608 * Decode a network abstraction layer unit.
609 * @param consumed is the number of bytes used as input
610 * @param length is the length of the array
611 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
612 * @return decoded bytes, might be src+1 if no escapes
614 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
617 * Free any data that may have been allocated in the H264 context like SPS, PPS etc.
619 av_cold void ff_h264_free_context(H264Context *h);
622 * Reconstruct bitstream slice_type.
624 int ff_h264_get_slice_type(const H264Context *h);
630 int ff_h264_alloc_tables(H264Context *h);
633 * Fill the default_ref_list.
635 int ff_h264_fill_default_ref_list(H264Context *h);
637 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
638 void ff_h264_fill_mbaff_ref_list(H264Context *h);
639 void ff_h264_remove_all_refs(H264Context *h);
642 * Execute the reference picture marking (memory management control operations).
644 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
646 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
648 void ff_generate_sliding_window_mmcos(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_intra4x4_pred_mode(H264Context *h);
657 * Check if the top & left blocks are available if needed & change the dc mode so it only uses the available blocks.
659 int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
661 void ff_h264_hl_decode_mb(H264Context *h);
662 int ff_h264_frame_start(H264Context *h);
663 int ff_h264_decode_extradata(H264Context *h);
664 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
665 av_cold int ff_h264_decode_end(AVCodecContext *avctx);
666 av_cold void ff_h264_decode_init_vlc(void);
669 * Decode a macroblock
670 * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
672 int ff_h264_decode_mb_cavlc(H264Context *h);
675 * Decode a CABAC coded macroblock
676 * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
678 int ff_h264_decode_mb_cabac(H264Context *h);
680 void ff_h264_init_cabac_states(H264Context *h);
682 void ff_h264_direct_dist_scale_factor(H264Context * const h);
683 void ff_h264_direct_ref_list_init(H264Context * const h);
684 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
686 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);
687 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);
690 * Reset SEI values at the beginning of the frame.
692 * @param h H.264 context.
694 void ff_h264_reset_sei(H264Context *h);
707 /* Scan8 organization:
724 * DY/DU/DV are for luma/chroma DC.
727 #define LUMA_DC_BLOCK_INDEX 48
728 #define CHROMA_DC_BLOCK_INDEX 49
730 //This table must be here because scan8[constant] must be known at compiletime
731 static const uint8_t scan8[16*3 + 3]={
732 4+ 1*8, 5+ 1*8, 4+ 2*8, 5+ 2*8,
733 6+ 1*8, 7+ 1*8, 6+ 2*8, 7+ 2*8,
734 4+ 3*8, 5+ 3*8, 4+ 4*8, 5+ 4*8,
735 6+ 3*8, 7+ 3*8, 6+ 4*8, 7+ 4*8,
736 4+ 6*8, 5+ 6*8, 4+ 7*8, 5+ 7*8,
737 6+ 6*8, 7+ 6*8, 6+ 7*8, 7+ 7*8,
738 4+ 8*8, 5+ 8*8, 4+ 9*8, 5+ 9*8,
739 6+ 8*8, 7+ 8*8, 6+ 9*8, 7+ 9*8,
740 4+11*8, 5+11*8, 4+12*8, 5+12*8,
741 6+11*8, 7+11*8, 6+12*8, 7+12*8,
742 4+13*8, 5+13*8, 4+14*8, 5+14*8,
743 6+13*8, 7+13*8, 6+14*8, 7+14*8,
744 0+ 0*8, 0+ 5*8, 0+10*8
747 static av_always_inline uint32_t pack16to32(int a, int b){
749 return (b&0xFFFF) + (a<<16);
751 return (a&0xFFFF) + (b<<16);
755 static av_always_inline uint16_t pack8to16(int a, int b){
757 return (b&0xFF) + (a<<8);
759 return (a&0xFF) + (b<<8);
764 * gets the chroma qp.
766 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
767 return h->pps.chroma_qp_table[t][qscale];
770 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
772 static void fill_decode_neighbors(H264Context *h, int mb_type){
773 MpegEncContext * const s = &h->s;
774 const int mb_xy= h->mb_xy;
775 int topleft_xy, top_xy, topright_xy, left_xy[2];
776 static const uint8_t left_block_options[4][32]={
777 {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},
778 {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},
779 {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},
780 {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}
783 h->topleft_partition= -1;
785 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
787 /* Wow, what a mess, why didn't they simplify the interlacing & intra
788 * stuff, I can't imagine that these complex rules are worth it. */
790 topleft_xy = top_xy - 1;
791 topright_xy= top_xy + 1;
792 left_xy[1] = left_xy[0] = mb_xy-1;
793 h->left_block = left_block_options[0];
795 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
796 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
798 if (left_mb_field_flag != curr_mb_field_flag) {
799 left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
800 if (curr_mb_field_flag) {
801 left_xy[1] += s->mb_stride;
802 h->left_block = left_block_options[3];
804 topleft_xy += s->mb_stride;
805 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
806 h->topleft_partition = 0;
807 h->left_block = left_block_options[1];
811 if(curr_mb_field_flag){
812 topleft_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
813 topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
814 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
816 if (left_mb_field_flag != curr_mb_field_flag) {
817 if (curr_mb_field_flag) {
818 left_xy[1] += s->mb_stride;
819 h->left_block = left_block_options[3];
821 h->left_block = left_block_options[2];
827 h->topleft_mb_xy = topleft_xy;
828 h->top_mb_xy = top_xy;
829 h->topright_mb_xy= topright_xy;
830 h->left_mb_xy[0] = left_xy[0];
831 h->left_mb_xy[1] = left_xy[1];
832 //FIXME do we need all in the context?
834 h->topleft_type = s->current_picture.mb_type[topleft_xy] ;
835 h->top_type = s->current_picture.mb_type[top_xy] ;
836 h->topright_type= s->current_picture.mb_type[topright_xy];
837 h->left_type[0] = s->current_picture.mb_type[left_xy[0]] ;
838 h->left_type[1] = s->current_picture.mb_type[left_xy[1]] ;
841 if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0;
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;
845 if(h->slice_table[topleft_xy ] != h->slice_num){
847 if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0;
848 if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
851 if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
854 static void fill_decode_caches(H264Context *h, int mb_type){
855 MpegEncContext * const s = &h->s;
856 int topleft_xy, top_xy, topright_xy, left_xy[2];
857 int topleft_type, top_type, topright_type, left_type[2];
858 const uint8_t * left_block= h->left_block;
861 topleft_xy = h->topleft_mb_xy ;
862 top_xy = h->top_mb_xy ;
863 topright_xy = h->topright_mb_xy;
864 left_xy[0] = h->left_mb_xy[0] ;
865 left_xy[1] = h->left_mb_xy[1] ;
866 topleft_type = h->topleft_type ;
867 top_type = h->top_type ;
868 topright_type= h->topright_type ;
869 left_type[0] = h->left_type[0] ;
870 left_type[1] = h->left_type[1] ;
872 if(!IS_SKIP(mb_type)){
873 if(IS_INTRA(mb_type)){
874 int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
875 h->topleft_samples_available=
876 h->top_samples_available=
877 h->left_samples_available= 0xFFFF;
878 h->topright_samples_available= 0xEEEA;
880 if(!(top_type & type_mask)){
881 h->topleft_samples_available= 0xB3FF;
882 h->top_samples_available= 0x33FF;
883 h->topright_samples_available= 0x26EA;
885 if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
886 if(IS_INTERLACED(mb_type)){
887 if(!(left_type[0] & type_mask)){
888 h->topleft_samples_available&= 0xDFFF;
889 h->left_samples_available&= 0x5FFF;
891 if(!(left_type[1] & type_mask)){
892 h->topleft_samples_available&= 0xFF5F;
893 h->left_samples_available&= 0xFF5F;
896 int left_typei = s->current_picture.mb_type[left_xy[0] + s->mb_stride];
898 assert(left_xy[0] == left_xy[1]);
899 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
900 h->topleft_samples_available&= 0xDF5F;
901 h->left_samples_available&= 0x5F5F;
905 if(!(left_type[0] & type_mask)){
906 h->topleft_samples_available&= 0xDF5F;
907 h->left_samples_available&= 0x5F5F;
911 if(!(topleft_type & type_mask))
912 h->topleft_samples_available&= 0x7FFF;
914 if(!(topright_type & type_mask))
915 h->topright_samples_available&= 0xFBFF;
917 if(IS_INTRA4x4(mb_type)){
918 if(IS_INTRA4x4(top_type)){
919 AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]);
921 h->intra4x4_pred_mode_cache[4+8*0]=
922 h->intra4x4_pred_mode_cache[5+8*0]=
923 h->intra4x4_pred_mode_cache[6+8*0]=
924 h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask);
927 if(IS_INTRA4x4(left_type[i])){
928 int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]];
929 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]];
930 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]];
932 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
933 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[i] & type_mask);
948 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
950 AV_COPY32(&h->non_zero_count_cache[4+8* 0], &h->non_zero_count[top_xy][4*3]);
952 AV_COPY32(&h->non_zero_count_cache[4+8* 5], &h->non_zero_count[top_xy][4* 7]);
953 AV_COPY32(&h->non_zero_count_cache[4+8*10], &h->non_zero_count[top_xy][4*11]);
955 AV_COPY32(&h->non_zero_count_cache[4+8* 5], &h->non_zero_count[top_xy][4* 5]);
956 AV_COPY32(&h->non_zero_count_cache[4+8*10], &h->non_zero_count[top_xy][4* 9]);
959 uint32_t top_empty = CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040;
960 AV_WN32A(&h->non_zero_count_cache[4+8* 0], top_empty);
961 AV_WN32A(&h->non_zero_count_cache[4+8* 5], top_empty);
962 AV_WN32A(&h->non_zero_count_cache[4+8*10], top_empty);
965 for (i=0; i<2; i++) {
967 h->non_zero_count_cache[3+8* 1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
968 h->non_zero_count_cache[3+8* 2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
970 h->non_zero_count_cache[3+8* 6 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]+4*4];
971 h->non_zero_count_cache[3+8* 7 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]+4*4];
972 h->non_zero_count_cache[3+8*11 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]+8*4];
973 h->non_zero_count_cache[3+8*12 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]+8*4];
975 h->non_zero_count_cache[3+8* 6 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
976 h->non_zero_count_cache[3+8*11 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
979 h->non_zero_count_cache[3+8* 1 + 2*8*i]=
980 h->non_zero_count_cache[3+8* 2 + 2*8*i]=
981 h->non_zero_count_cache[3+8* 6 + 2*8*i]=
982 h->non_zero_count_cache[3+8* 7 + 2*8*i]=
983 h->non_zero_count_cache[3+8*11 + 2*8*i]=
984 h->non_zero_count_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 for(list=0; list<h->list_count; list++){
1009 if(!USES_LIST(mb_type, list)){
1010 /*if(!h->mv_cache_clean[list]){
1011 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
1012 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
1013 h->mv_cache_clean[list]= 1;
1017 assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
1019 h->mv_cache_clean[list]= 0;
1021 if(USES_LIST(top_type, list)){
1022 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1023 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1024 h->ref_cache[list][scan8[0] + 0 - 1*8]=
1025 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 2];
1026 h->ref_cache[list][scan8[0] + 2 - 1*8]=
1027 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 3];
1029 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1030 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
1033 if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){
1035 int cache_idx = scan8[0] - 1 + i*2*8;
1036 if(USES_LIST(left_type[i], list)){
1037 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1038 const int b8_xy= 4*left_xy[i] + 1;
1039 AV_COPY32(h->mv_cache[list][cache_idx ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]);
1040 AV_COPY32(h->mv_cache[list][cache_idx+8], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]]);
1041 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + (left_block[0+i*2]&~1)];
1042 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + (left_block[1+i*2]&~1)];
1044 AV_ZERO32(h->mv_cache [list][cache_idx ]);
1045 AV_ZERO32(h->mv_cache [list][cache_idx+8]);
1046 h->ref_cache[list][cache_idx ]=
1047 h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1051 if(USES_LIST(left_type[0], list)){
1052 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1053 const int b8_xy= 4*left_xy[0] + 1;
1054 AV_COPY32(h->mv_cache[list][scan8[0] - 1], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]]);
1055 h->ref_cache[list][scan8[0] - 1]= s->current_picture.ref_index[list][b8_xy + (left_block[0]&~1)];
1057 AV_ZERO32(h->mv_cache [list][scan8[0] - 1]);
1058 h->ref_cache[list][scan8[0] - 1]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1062 if(USES_LIST(topright_type, list)){
1063 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1064 AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]);
1065 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][4*topright_xy + 2];
1067 AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]);
1068 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1070 if(h->ref_cache[list][scan8[0] + 4 - 1*8] < 0){
1071 if(USES_LIST(topleft_type, list)){
1072 const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
1073 const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2);
1074 AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]);
1075 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1077 AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]);
1078 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1082 if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
1085 if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
1086 h->ref_cache[list][scan8[4 ]] =
1087 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1088 AV_ZERO32(h->mv_cache [list][scan8[4 ]]);
1089 AV_ZERO32(h->mv_cache [list][scan8[12]]);
1092 /* XXX beurk, Load mvd */
1093 if(USES_LIST(top_type, list)){
1094 const int b_xy= h->mb2br_xy[top_xy];
1095 AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1097 AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1099 if(USES_LIST(left_type[0], list)){
1100 const int b_xy= h->mb2br_xy[left_xy[0]] + 6;
1101 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy - left_block[0]]);
1102 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy - left_block[1]]);
1104 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]);
1105 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]);
1107 if(USES_LIST(left_type[1], list)){
1108 const int b_xy= h->mb2br_xy[left_xy[1]] + 6;
1109 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy - left_block[2]]);
1110 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy - left_block[3]]);
1112 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]);
1113 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]);
1115 AV_ZERO16(h->mvd_cache [list][scan8[4 ]]);
1116 AV_ZERO16(h->mvd_cache [list][scan8[12]]);
1117 if(h->slice_type_nos == AV_PICTURE_TYPE_B){
1118 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
1120 if(IS_DIRECT(top_type)){
1121 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101u*(MB_TYPE_DIRECT2>>1));
1122 }else if(IS_8X8(top_type)){
1123 int b8_xy = 4*top_xy;
1124 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy + 2];
1125 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 3];
1127 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1));
1130 if(IS_DIRECT(left_type[0]))
1131 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
1132 else if(IS_8X8(left_type[0]))
1133 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[4*left_xy[0] + 1 + (left_block[0]&~1)];
1135 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
1137 if(IS_DIRECT(left_type[1]))
1138 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
1139 else if(IS_8X8(left_type[1]))
1140 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[4*left_xy[1] + 1 + (left_block[2]&~1)];
1142 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
1148 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1149 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1150 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1151 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1152 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1153 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1154 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1155 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1156 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1157 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1159 #define MAP_F2F(idx, mb_type)\
1160 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1161 h->ref_cache[list][idx] <<= 1;\
1162 h->mv_cache[list][idx][1] /= 2;\
1163 h->mvd_cache[list][idx][1] >>=1;\
1168 #define MAP_F2F(idx, mb_type)\
1169 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1170 h->ref_cache[list][idx] >>= 1;\
1171 h->mv_cache[list][idx][1] <<= 1;\
1172 h->mvd_cache[list][idx][1] <<= 1;\
1181 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1185 * gets the predicted intra4x4 prediction mode.
1187 static av_always_inline int pred_intra_mode(H264Context *h, int n){
1188 const int index8= scan8[n];
1189 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1190 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1191 const int min= FFMIN(left, top);
1193 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1195 if(min<0) return DC_PRED;
1199 static av_always_inline void write_back_intra_pred_mode(H264Context *h){
1200 int8_t *i4x4= h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
1201 int8_t *i4x4_cache= h->intra4x4_pred_mode_cache;
1203 AV_COPY32(i4x4, i4x4_cache + 4 + 8*4);
1204 i4x4[4]= i4x4_cache[7+8*3];
1205 i4x4[5]= i4x4_cache[7+8*2];
1206 i4x4[6]= i4x4_cache[7+8*1];
1209 static av_always_inline void write_back_non_zero_count(H264Context *h){
1210 const int mb_xy= h->mb_xy;
1211 uint8_t *nnz = h->non_zero_count[mb_xy];
1212 uint8_t *nnz_cache = h->non_zero_count_cache;
1214 AV_COPY32(&nnz[ 0], &nnz_cache[4+8* 1]);
1215 AV_COPY32(&nnz[ 4], &nnz_cache[4+8* 2]);
1216 AV_COPY32(&nnz[ 8], &nnz_cache[4+8* 3]);
1217 AV_COPY32(&nnz[12], &nnz_cache[4+8* 4]);
1218 AV_COPY32(&nnz[16], &nnz_cache[4+8* 6]);
1219 AV_COPY32(&nnz[20], &nnz_cache[4+8* 7]);
1220 AV_COPY32(&nnz[32], &nnz_cache[4+8*11]);
1221 AV_COPY32(&nnz[36], &nnz_cache[4+8*12]);
1224 AV_COPY32(&nnz[24], &nnz_cache[4+8* 8]);
1225 AV_COPY32(&nnz[28], &nnz_cache[4+8* 9]);
1226 AV_COPY32(&nnz[40], &nnz_cache[4+8*13]);
1227 AV_COPY32(&nnz[44], &nnz_cache[4+8*14]);
1231 static av_always_inline void write_back_motion_list(H264Context *h, MpegEncContext * const s, int b_stride,
1232 int b_xy, int b8_xy, int mb_type, int list )
1234 int16_t (*mv_dst)[2] = &s->current_picture.motion_val[list][b_xy];
1235 int16_t (*mv_src)[2] = &h->mv_cache[list][scan8[0]];
1236 AV_COPY128(mv_dst + 0*b_stride, mv_src + 8*0);
1237 AV_COPY128(mv_dst + 1*b_stride, mv_src + 8*1);
1238 AV_COPY128(mv_dst + 2*b_stride, mv_src + 8*2);
1239 AV_COPY128(mv_dst + 3*b_stride, mv_src + 8*3);
1241 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
1242 uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1243 if(IS_SKIP(mb_type))
1244 AV_ZERO128(mvd_dst);
1246 AV_COPY64(mvd_dst, mvd_src + 8*3);
1247 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
1248 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
1249 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
1254 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1255 int8_t *ref_cache = h->ref_cache[list];
1256 ref_index[0+0*2]= ref_cache[scan8[0]];
1257 ref_index[1+0*2]= ref_cache[scan8[4]];
1258 ref_index[0+1*2]= ref_cache[scan8[8]];
1259 ref_index[1+1*2]= ref_cache[scan8[12]];
1263 static av_always_inline void write_back_motion(H264Context *h, int mb_type){
1264 MpegEncContext * const s = &h->s;
1265 const int b_stride = h->b_stride;
1266 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
1267 const int b8_xy= 4*h->mb_xy;
1269 if(USES_LIST(mb_type, 0)){
1270 write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 0);
1272 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1274 if(USES_LIST(mb_type, 1)){
1275 write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 1);
1278 if(h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC){
1279 if(IS_8X8(mb_type)){
1280 uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
1281 direct_table[1] = h->sub_mb_type[1]>>1;
1282 direct_table[2] = h->sub_mb_type[2]>>1;
1283 direct_table[3] = h->sub_mb_type[3]>>1;
1288 static av_always_inline int get_dct8x8_allowed(H264Context *h){
1289 if(h->sps.direct_8x8_inference_flag)
1290 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL));
1292 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1296 * decodes a P_SKIP or B_SKIP macroblock
1298 static void av_unused decode_mb_skip(H264Context *h){
1299 MpegEncContext * const s = &h->s;
1300 const int mb_xy= h->mb_xy;
1303 memset(h->non_zero_count[mb_xy], 0, 48);
1306 mb_type|= MB_TYPE_INTERLACED;
1308 if( h->slice_type_nos == AV_PICTURE_TYPE_B )
1310 // just for fill_caches. pred_direct_motion will set the real mb_type
1311 mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1312 if(h->direct_spatial_mv_pred){
1313 fill_decode_neighbors(h, mb_type);
1314 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1316 ff_h264_pred_direct_motion(h, &mb_type);
1317 mb_type|= MB_TYPE_SKIP;
1322 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1324 fill_decode_neighbors(h, mb_type);
1325 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1326 pred_pskip_motion(h, &mx, &my);
1327 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1328 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1331 write_back_motion(h, mb_type);
1332 s->current_picture.mb_type[mb_xy]= mb_type;
1333 s->current_picture.qscale_table[mb_xy]= s->qscale;
1334 h->slice_table[ mb_xy ]= h->slice_num;
1335 h->prev_mb_skipped= 1;
1338 #include "h264_mvpred.h" //For pred_pskip_motion()
1340 #endif /* AVCODEC_H264_H */