2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 * @file libavcodec/h264.h
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"
36 #include "rectangle.h"
38 #define interlaced_dct interlaced_dct_is_a_bad_name
39 #define mb_intra mb_intra_is_not_initialized_see_mb_type
41 #define LUMA_DC_BLOCK_INDEX 25
42 #define CHROMA_DC_BLOCK_INDEX 26
44 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
45 #define COEFF_TOKEN_VLC_BITS 8
46 #define TOTAL_ZEROS_VLC_BITS 9
47 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
48 #define RUN_VLC_BITS 3
49 #define RUN7_VLC_BITS 6
51 #define MAX_SPS_COUNT 32
52 #define MAX_PPS_COUNT 256
54 #define MAX_MMCO_COUNT 66
56 #define MAX_DELAYED_PIC_COUNT 16
58 /* Compiling in interlaced support reduces the speed
59 * of progressive decoding by about 2%. */
60 #define ALLOW_INTERLACE
62 #define ALLOW_NOCHROMA
67 * The maximum number of slices supported by the decoder.
68 * must be a power of 2
72 #ifdef ALLOW_INTERLACE
73 #define MB_MBAFF h->mb_mbaff
74 #define MB_FIELD h->mb_field_decoding_flag
75 #define FRAME_MBAFF h->mb_aff_frame
76 #define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
81 #define FIELD_PICTURE 0
83 #define IS_INTERLACED(mb_type) 0
85 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
88 #define CHROMA h->sps.chroma_format_idc
94 #define CABAC h->pps.cabac
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
126 NAL_AUXILIARY_SLICE=19
133 SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
134 SEI_TYPE_PIC_TIMING = 1, ///< picture timing
135 SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
136 SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync)
140 * pic_struct in picture timing SEI message
143 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
144 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
145 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
146 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
147 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
148 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
149 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
150 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
151 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
155 * Sequence parameter set
161 int chroma_format_idc;
162 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
163 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
164 int poc_type; ///< pic_order_cnt_type
165 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
166 int delta_pic_order_always_zero_flag;
167 int offset_for_non_ref_pic;
168 int offset_for_top_to_bottom_field;
169 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
170 int ref_frame_count; ///< num_ref_frames
171 int gaps_in_frame_num_allowed_flag;
172 int mb_width; ///< pic_width_in_mbs_minus1 + 1
173 int mb_height; ///< pic_height_in_map_units_minus1 + 1
174 int frame_mbs_only_flag;
175 int mb_aff; ///<mb_adaptive_frame_field_flag
176 int direct_8x8_inference_flag;
177 int crop; ///< frame_cropping_flag
178 unsigned int crop_left; ///< frame_cropping_rect_left_offset
179 unsigned int crop_right; ///< frame_cropping_rect_right_offset
180 unsigned int crop_top; ///< frame_cropping_rect_top_offset
181 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
182 int vui_parameters_present_flag;
184 int video_signal_type_present_flag;
186 int colour_description_present_flag;
187 enum AVColorPrimaries color_primaries;
188 enum AVColorTransferCharacteristic color_trc;
189 enum AVColorSpace colorspace;
190 int timing_info_present_flag;
191 uint32_t num_units_in_tick;
193 int fixed_frame_rate_flag;
194 short offset_for_ref_frame[256]; //FIXME dyn aloc?
195 int bitstream_restriction_flag;
196 int num_reorder_frames;
197 int scaling_matrix_present;
198 uint8_t scaling_matrix4[6][16];
199 uint8_t scaling_matrix8[2][64];
200 int nal_hrd_parameters_present_flag;
201 int vcl_hrd_parameters_present_flag;
202 int pic_struct_present_flag;
203 int time_offset_length;
204 int cpb_cnt; ///< See H.264 E.1.2
205 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1
206 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
207 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
208 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
209 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
210 int residual_color_transform_flag; ///< residual_colour_transform_flag
214 * Picture parameter set
218 int cabac; ///< entropy_coding_mode_flag
219 int pic_order_present; ///< pic_order_present_flag
220 int slice_group_count; ///< num_slice_groups_minus1 + 1
221 int mb_slice_group_map_type;
222 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
223 int weighted_pred; ///< weighted_pred_flag
224 int weighted_bipred_idc;
225 int init_qp; ///< pic_init_qp_minus26 + 26
226 int init_qs; ///< pic_init_qs_minus26 + 26
227 int chroma_qp_index_offset[2];
228 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
229 int constrained_intra_pred; ///< constrained_intra_pred_flag
230 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
231 int transform_8x8_mode; ///< transform_8x8_mode_flag
232 uint8_t scaling_matrix4[6][16];
233 uint8_t scaling_matrix8[2][64];
234 uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
239 * Memory management control operation opcode.
241 typedef enum MMCOOpcode{
252 * Memory management control operation.
256 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
257 int long_arg; ///< index, pic_num, or num long refs depending on opcode
263 typedef struct H264Context{
265 int chroma_qp[2]; //QPc
267 int qp_thresh; ///< QP threshold to skip loopfilter
273 int chroma_pred_mode;
274 int intra16x16_pred_mode;
286 const uint8_t * left_block;
287 int topleft_partition;
289 int8_t intra4x4_pred_mode_cache[5*8];
290 int8_t (*intra4x4_pred_mode);
292 unsigned int topleft_samples_available;
293 unsigned int top_samples_available;
294 unsigned int topright_samples_available;
295 unsigned int left_samples_available;
296 uint8_t (*top_borders[2])[16+2*8];
299 * non zero coeff count cache.
300 * is 64 if not available.
302 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[6*8];
310 uint8_t (*non_zero_count)[32];
313 * Motion vector cache.
315 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5*8][2];
316 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5*8];
317 #define LIST_NOT_USED -1 //FIXME rename?
318 #define PART_NOT_AVAILABLE -2
321 * is 1 if the specific list MV&references are set to 0,0,-2.
323 int mv_cache_clean[2];
326 * number of neighbors (top and/or left) that used 8x8 dct
328 int neighbor_transform_size;
331 * block_offset[ 0..23] for frame macroblocks
332 * block_offset[24..47] for field macroblocks
334 int block_offset[2*(16+8)];
336 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
338 int b_stride; //FIXME use s->b4_stride
340 int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff
346 SPS sps; ///< current sps
351 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
353 uint32_t dequant4_buffer[6][52][16]; //FIXME should these be moved down?
354 uint32_t dequant8_buffer[2][52][64];
355 uint32_t (*dequant4_coeff[6])[16];
356 uint32_t (*dequant8_coeff[2])[64];
359 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
361 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
362 int slice_type_fixed;
364 //interlacing specific flags
366 int mb_field_decoding_flag;
367 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
369 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
371 //Weighted pred stuff
373 int use_weight_chroma;
374 int luma_log2_weight_denom;
375 int chroma_log2_weight_denom;
376 //The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
377 int luma_weight[48][2][2];
378 int chroma_weight[48][2][2][2];
379 int implicit_weight[48][48];
381 int direct_spatial_mv_pred;
384 int dist_scale_factor[16];
385 int dist_scale_factor_field[2][32];
386 int map_col_to_list0[2][16+32];
387 int map_col_to_list0_field[2][2][16+32];
390 * num_ref_idx_l0/1_active_minus1 + 1
392 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
393 unsigned int list_count;
394 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
395 Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
396 Reordered version of default_ref_list
397 according to picture reordering in slice header */
398 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
401 GetBitContext intra_gb;
402 GetBitContext inter_gb;
403 GetBitContext *intra_gb_ptr;
404 GetBitContext *inter_gb_ptr;
406 DECLARE_ALIGNED(16, DCTELEM, mb)[16*24];
407 DCTELEM mb_padding[256]; ///< 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
413 uint8_t cabac_state[460];
415 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
420 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
421 uint8_t *chroma_pred_mode_table;
422 int last_qscale_diff;
423 uint8_t (*mvd_table[2])[2];
424 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5*8][2];
425 uint8_t *direct_table;
426 uint8_t direct_cache[5*8];
428 uint8_t zigzag_scan[16];
429 uint8_t zigzag_scan8x8[64];
430 uint8_t zigzag_scan8x8_cavlc[64];
431 uint8_t field_scan[16];
432 uint8_t field_scan8x8[64];
433 uint8_t field_scan8x8_cavlc[64];
434 const uint8_t *zigzag_scan_q0;
435 const uint8_t *zigzag_scan8x8_q0;
436 const uint8_t *zigzag_scan8x8_cavlc_q0;
437 const uint8_t *field_scan_q0;
438 const uint8_t *field_scan8x8_q0;
439 const uint8_t *field_scan8x8_cavlc_q0;
448 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
449 int slice_alpha_c0_offset;
450 int slice_beta_offset;
452 //=============================================================
453 //Things below are not used in the MB or more inner code
457 uint8_t *rbsp_buffer[2];
458 unsigned int rbsp_buffer_size[2];
461 * Used to parse AVC variant of h264
463 int is_avc; ///< this flag is != 0 if codec is avc1
464 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
466 SPS *sps_buffers[MAX_SPS_COUNT];
467 PPS *pps_buffers[MAX_PPS_COUNT];
469 int dequant_coeff_pps; ///< reinit tables when pps changes
471 uint16_t *slice_table_base;
477 int delta_poc_bottom;
480 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
481 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
482 int frame_num_offset; ///< for POC type 2
483 int prev_frame_num_offset; ///< for POC type 2
484 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
487 * frame_num for frames or 2*frame_num+1 for field pics.
492 * max_frame_num or 2*max_frame_num for field pics.
496 int redundant_pic_count;
498 Picture *short_ref[32];
499 Picture *long_ref[32];
500 Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
501 Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
505 * memory management control operations buffer.
507 MMCO mmco[MAX_MMCO_COUNT];
510 int long_ref_count; ///< number of actual long term references
511 int short_ref_count; ///< number of actual short term references
516 * @defgroup multithreading Members for slice based multithreading
519 struct H264Context *thread_context[MAX_THREADS];
522 * current slice number, used to initalize slice_num of each thread/context
527 * Max number of threads / contexts.
528 * This is equal to AVCodecContext.thread_count unless
529 * multithreaded decoding is impossible, in which case it is
535 * 1 if the single thread fallback warning has already been
536 * displayed, 0 otherwise.
538 int single_decode_warning;
544 * pic_struct in picture timing SEI message
546 SEI_PicStructType sei_pic_struct;
549 * Complement sei_pic_struct
550 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
551 * However, soft telecined frames may have these values.
552 * This is used in an attempt to flag soft telecine progressive.
554 int prev_interlaced_frame;
557 * Bit set of clock types for fields/frames in picture timing SEI message.
558 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
564 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
566 int sei_dpb_output_delay;
569 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
571 int sei_cpb_removal_delay;
574 * recovery_frame_cnt from SEI message
576 * Set to -1 if no recovery point SEI message found or to number of frames
577 * before playback synchronizes. Frames having recovery point are key
580 int sei_recovery_frame_cnt;
582 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
583 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
586 int sei_buffering_period_present; ///< Buffering period SEI flag
587 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
589 //SVQ3 specific fields
592 int unknown_svq3_flag;
593 int next_slice_index;
594 uint32_t svq3_watermark_key;
598 extern const uint8_t ff_h264_chroma_qp[52];
600 void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
602 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
607 int ff_h264_decode_sei(H264Context *h);
612 int ff_h264_decode_seq_parameter_set(H264Context *h);
617 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
620 * Decodes a network abstraction layer unit.
621 * @param consumed is the number of bytes used as input
622 * @param length is the length of the array
623 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
624 * @returns decoded bytes, might be src+1 if no escapes
626 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
629 * identifies the exact end of the bitstream
630 * @return the length of the trailing, or 0 if damaged
632 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
635 * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
637 av_cold void ff_h264_free_context(H264Context *h);
640 * reconstructs bitstream slice_type.
642 int ff_h264_get_slice_type(const H264Context *h);
648 int ff_h264_alloc_tables(H264Context *h);
651 * fills the default_ref_list.
653 int ff_h264_fill_default_ref_list(H264Context *h);
655 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
656 void ff_h264_fill_mbaff_ref_list(H264Context *h);
657 void ff_h264_remove_all_refs(H264Context *h);
660 * Executes the reference picture marking (memory management control operations).
662 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
664 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
668 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
670 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
673 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
675 int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
677 void ff_h264_write_back_intra_pred_mode(H264Context *h);
678 void ff_h264_hl_decode_mb(H264Context *h);
679 int ff_h264_frame_start(H264Context *h);
680 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
681 av_cold int ff_h264_decode_end(AVCodecContext *avctx);
682 av_cold void ff_h264_decode_init_vlc(void);
685 * decodes a macroblock
686 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
688 int ff_h264_decode_mb_cavlc(H264Context *h);
691 * decodes a CABAC coded macroblock
692 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
694 int ff_h264_decode_mb_cabac(H264Context *h);
696 void ff_h264_init_cabac_states(H264Context *h);
698 void ff_h264_direct_dist_scale_factor(H264Context * const h);
699 void ff_h264_direct_ref_list_init(H264Context * const h);
700 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
702 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);
703 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);
706 * Reset SEI values at the beginning of the frame.
708 * @param h H.264 context.
710 void ff_h264_reset_sei(H264Context *h);
722 //This table must be here because scan8[constant] must be known at compiletime
723 static const uint8_t scan8[16 + 2*4]={
724 4+1*8, 5+1*8, 4+2*8, 5+2*8,
725 6+1*8, 7+1*8, 6+2*8, 7+2*8,
726 4+3*8, 5+3*8, 4+4*8, 5+4*8,
727 6+3*8, 7+3*8, 6+4*8, 7+4*8,
734 static av_always_inline uint32_t pack16to32(int a, int b){
736 return (b&0xFFFF) + (a<<16);
738 return (a&0xFFFF) + (b<<16);
742 static av_always_inline uint16_t pack8to16(int a, int b){
744 return (b&0xFF) + (a<<8);
746 return (a&0xFF) + (b<<8);
751 * gets the chroma qp.
753 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
754 return h->pps.chroma_qp_table[t][qscale];
757 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
759 static void fill_decode_neighbors(H264Context *h, int mb_type){
760 MpegEncContext * const s = &h->s;
761 const int mb_xy= h->mb_xy;
762 int topleft_xy, top_xy, topright_xy, left_xy[2];
763 static const uint8_t left_block_options[4][16]={
764 {0,1,2,3,7,10,8,11,7+0*8, 7+1*8, 7+2*8, 7+3*8, 2+0*8, 2+3*8, 2+1*8, 2+2*8},
765 {2,2,3,3,8,11,8,11,7+2*8, 7+2*8, 7+3*8, 7+3*8, 2+1*8, 2+2*8, 2+1*8, 2+2*8},
766 {0,0,1,1,7,10,7,10,7+0*8, 7+0*8, 7+1*8, 7+1*8, 2+0*8, 2+3*8, 2+0*8, 2+3*8},
767 {0,2,0,2,7,10,7,10,7+0*8, 7+2*8, 7+0*8, 7+2*8, 2+0*8, 2+3*8, 2+0*8, 2+3*8}
770 h->topleft_partition= -1;
772 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
774 /* Wow, what a mess, why didn't they simplify the interlacing & intra
775 * stuff, I can't imagine that these complex rules are worth it. */
777 topleft_xy = top_xy - 1;
778 topright_xy= top_xy + 1;
779 left_xy[1] = left_xy[0] = mb_xy-1;
780 h->left_block = left_block_options[0];
782 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
783 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
785 if (left_mb_field_flag != curr_mb_field_flag) {
786 left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
787 if (curr_mb_field_flag) {
788 left_xy[1] += s->mb_stride;
789 h->left_block = left_block_options[3];
791 topleft_xy += s->mb_stride;
792 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
793 h->topleft_partition = 0;
794 h->left_block = left_block_options[1];
798 if(curr_mb_field_flag){
799 topleft_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
800 topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
801 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
803 if (left_mb_field_flag != curr_mb_field_flag) {
804 if (curr_mb_field_flag) {
805 left_xy[1] += s->mb_stride;
806 h->left_block = left_block_options[3];
808 h->left_block = left_block_options[2];
814 h->topleft_mb_xy = topleft_xy;
815 h->top_mb_xy = top_xy;
816 h->topright_mb_xy= topright_xy;
817 h->left_mb_xy[0] = left_xy[0];
818 h->left_mb_xy[1] = left_xy[1];
819 //FIXME do we need all in the context?
821 h->topleft_type = s->current_picture.mb_type[topleft_xy] ;
822 h->top_type = s->current_picture.mb_type[top_xy] ;
823 h->topright_type= s->current_picture.mb_type[topright_xy];
824 h->left_type[0] = s->current_picture.mb_type[left_xy[0]] ;
825 h->left_type[1] = s->current_picture.mb_type[left_xy[1]] ;
828 if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0;
829 if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0;
830 if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
832 if(h->slice_table[topleft_xy ] != h->slice_num){
834 if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0;
835 if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
838 if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
841 static void fill_decode_caches(H264Context *h, int mb_type){
842 MpegEncContext * const s = &h->s;
843 int topleft_xy, top_xy, topright_xy, left_xy[2];
844 int topleft_type, top_type, topright_type, left_type[2];
845 const uint8_t * left_block= h->left_block;
848 topleft_xy = h->topleft_mb_xy ;
849 top_xy = h->top_mb_xy ;
850 topright_xy = h->topright_mb_xy;
851 left_xy[0] = h->left_mb_xy[0] ;
852 left_xy[1] = h->left_mb_xy[1] ;
853 topleft_type = h->topleft_type ;
854 top_type = h->top_type ;
855 topright_type= h->topright_type ;
856 left_type[0] = h->left_type[0] ;
857 left_type[1] = h->left_type[1] ;
859 if(!IS_SKIP(mb_type)){
860 if(IS_INTRA(mb_type)){
861 int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
862 h->topleft_samples_available=
863 h->top_samples_available=
864 h->left_samples_available= 0xFFFF;
865 h->topright_samples_available= 0xEEEA;
867 if(!(top_type & type_mask)){
868 h->topleft_samples_available= 0xB3FF;
869 h->top_samples_available= 0x33FF;
870 h->topright_samples_available= 0x26EA;
872 if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
873 if(IS_INTERLACED(mb_type)){
874 if(!(left_type[0] & type_mask)){
875 h->topleft_samples_available&= 0xDFFF;
876 h->left_samples_available&= 0x5FFF;
878 if(!(left_type[1] & type_mask)){
879 h->topleft_samples_available&= 0xFF5F;
880 h->left_samples_available&= 0xFF5F;
883 int left_typei = s->current_picture.mb_type[left_xy[0] + s->mb_stride];
885 assert(left_xy[0] == left_xy[1]);
886 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
887 h->topleft_samples_available&= 0xDF5F;
888 h->left_samples_available&= 0x5F5F;
892 if(!(left_type[0] & type_mask)){
893 h->topleft_samples_available&= 0xDF5F;
894 h->left_samples_available&= 0x5F5F;
898 if(!(topleft_type & type_mask))
899 h->topleft_samples_available&= 0x7FFF;
901 if(!(topright_type & type_mask))
902 h->topright_samples_available&= 0xFBFF;
904 if(IS_INTRA4x4(mb_type)){
905 if(IS_INTRA4x4(top_type)){
906 AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]);
908 h->intra4x4_pred_mode_cache[4+8*0]=
909 h->intra4x4_pred_mode_cache[5+8*0]=
910 h->intra4x4_pred_mode_cache[6+8*0]=
911 h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask);
914 if(IS_INTRA4x4(left_type[i])){
915 int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]];
916 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]];
917 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]];
919 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
920 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[i] & type_mask);
935 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
937 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
938 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
939 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
941 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
942 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
944 h->non_zero_count_cache[1+8*0]=
945 h->non_zero_count_cache[2+8*0]=
947 h->non_zero_count_cache[1+8*3]=
948 h->non_zero_count_cache[2+8*3]=
949 AV_WN32A(&h->non_zero_count_cache[4+8*0], CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040);
952 for (i=0; i<2; i++) {
954 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
955 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
956 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
957 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
959 h->non_zero_count_cache[3+8*1 + 2*8*i]=
960 h->non_zero_count_cache[3+8*2 + 2*8*i]=
961 h->non_zero_count_cache[0+8*1 + 8*i]=
962 h->non_zero_count_cache[0+8*4 + 8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
969 h->top_cbp = h->cbp_table[top_xy];
971 h->top_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
975 h->left_cbp = (h->cbp_table[left_xy[0]] & 0x1f0)
976 | ((h->cbp_table[left_xy[0]]>>(left_block[0]&(~1)))&2)
977 | (((h->cbp_table[left_xy[1]]>>(left_block[2]&(~1)))&2) << 2);
979 h->left_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
985 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
987 for(list=0; list<h->list_count; list++){
988 if(!USES_LIST(mb_type, list)){
989 /*if(!h->mv_cache_clean[list]){
990 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
991 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
992 h->mv_cache_clean[list]= 1;
996 assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
998 h->mv_cache_clean[list]= 0;
1000 if(USES_LIST(top_type, list)){
1001 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1002 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1003 h->ref_cache[list][scan8[0] + 0 - 1*8]=
1004 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 2];
1005 h->ref_cache[list][scan8[0] + 2 - 1*8]=
1006 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 3];
1008 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1009 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
1012 if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){
1014 int cache_idx = scan8[0] - 1 + i*2*8;
1015 if(USES_LIST(left_type[i], list)){
1016 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1017 const int b8_xy= 4*left_xy[i] + 1;
1018 AV_COPY32(h->mv_cache[list][cache_idx ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]);
1019 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]]);
1020 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + (left_block[0+i*2]&~1)];
1021 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + (left_block[1+i*2]&~1)];
1023 AV_ZERO32(h->mv_cache [list][cache_idx ]);
1024 AV_ZERO32(h->mv_cache [list][cache_idx+8]);
1025 h->ref_cache[list][cache_idx ]=
1026 h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1030 if(USES_LIST(left_type[0], list)){
1031 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1032 const int b8_xy= 4*left_xy[0] + 1;
1033 AV_COPY32(h->mv_cache[list][scan8[0] - 1], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]]);
1034 h->ref_cache[list][scan8[0] - 1]= s->current_picture.ref_index[list][b8_xy + (left_block[0]&~1)];
1036 AV_ZERO32(h->mv_cache [list][scan8[0] - 1]);
1037 h->ref_cache[list][scan8[0] - 1]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1041 if(USES_LIST(topright_type, list)){
1042 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1043 AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]);
1044 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][4*topright_xy + 2];
1046 AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]);
1047 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1049 if(h->ref_cache[list][scan8[0] + 4 - 1*8] < 0){
1050 if(USES_LIST(topleft_type, list)){
1051 const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
1052 const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2);
1053 AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]);
1054 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1056 AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]);
1057 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1061 if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
1064 if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
1065 h->ref_cache[list][scan8[4 ]] =
1066 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1067 AV_ZERO32(h->mv_cache [list][scan8[4 ]]);
1068 AV_ZERO32(h->mv_cache [list][scan8[12]]);
1071 /* XXX beurk, Load mvd */
1072 if(USES_LIST(top_type, list)){
1073 const int b_xy= h->mb2br_xy[top_xy];
1074 AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1076 AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1078 if(USES_LIST(left_type[0], list)){
1079 const int b_xy= h->mb2br_xy[left_xy[0]] + 6;
1080 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy - left_block[0]]);
1081 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy - left_block[1]]);
1083 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]);
1084 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]);
1086 if(USES_LIST(left_type[1], list)){
1087 const int b_xy= h->mb2br_xy[left_xy[1]] + 6;
1088 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy - left_block[2]]);
1089 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy - left_block[3]]);
1091 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]);
1092 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]);
1094 AV_ZERO16(h->mvd_cache [list][scan8[4 ]]);
1095 AV_ZERO16(h->mvd_cache [list][scan8[12]]);
1096 if(h->slice_type_nos == FF_B_TYPE){
1097 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
1099 if(IS_DIRECT(top_type)){
1100 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_DIRECT2>>1));
1101 }else if(IS_8X8(top_type)){
1102 int b8_xy = 4*top_xy;
1103 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy + 2];
1104 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 3];
1106 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1));
1109 if(IS_DIRECT(left_type[0]))
1110 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
1111 else if(IS_8X8(left_type[0]))
1112 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[4*left_xy[0] + 1 + (left_block[0]&~1)];
1114 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
1116 if(IS_DIRECT(left_type[1]))
1117 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
1118 else if(IS_8X8(left_type[1]))
1119 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[4*left_xy[1] + 1 + (left_block[2]&~1)];
1121 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
1127 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1128 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1129 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1130 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1131 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1132 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1133 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1134 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1135 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1136 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1138 #define MAP_F2F(idx, mb_type)\
1139 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1140 h->ref_cache[list][idx] <<= 1;\
1141 h->mv_cache[list][idx][1] /= 2;\
1142 h->mvd_cache[list][idx][1] >>=1;\
1147 #define MAP_F2F(idx, mb_type)\
1148 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1149 h->ref_cache[list][idx] >>= 1;\
1150 h->mv_cache[list][idx][1] <<= 1;\
1151 h->mvd_cache[list][idx][1] <<= 1;\
1161 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1166 * @returns non zero if the loop filter can be skiped
1168 static int fill_filter_caches(H264Context *h, int mb_type){
1169 MpegEncContext * const s = &h->s;
1170 const int mb_xy= h->mb_xy;
1171 int top_xy, left_xy[2];
1172 int top_type, left_type[2];
1174 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
1176 //FIXME deblocking could skip the intra and nnz parts.
1178 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1179 * stuff, I can't imagine that these complex rules are worth it. */
1181 left_xy[1] = left_xy[0] = mb_xy-1;
1183 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
1184 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1186 if (left_mb_field_flag != curr_mb_field_flag) {
1187 left_xy[0] -= s->mb_stride;
1190 if(curr_mb_field_flag){
1191 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
1193 if (left_mb_field_flag != curr_mb_field_flag) {
1194 left_xy[1] += s->mb_stride;
1199 h->top_mb_xy = top_xy;
1200 h->left_mb_xy[0] = left_xy[0];
1201 h->left_mb_xy[1] = left_xy[1];
1203 //for sufficiently low qp, filtering wouldn't do anything
1204 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
1205 int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
1206 int qp = s->current_picture.qscale_table[mb_xy];
1208 && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
1209 && (top_xy < 0 || ((qp + s->current_picture.qscale_table[top_xy ] + 1)>>1) <= qp_thresh)){
1212 if( (left_xy[0]< 0 || ((qp + s->current_picture.qscale_table[left_xy[1] ] + 1)>>1) <= qp_thresh)
1213 && (top_xy < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy -s->mb_stride] + 1)>>1) <= qp_thresh))
1218 top_type = s->current_picture.mb_type[top_xy] ;
1219 left_type[0] = s->current_picture.mb_type[left_xy[0]];
1220 left_type[1] = s->current_picture.mb_type[left_xy[1]];
1221 if(h->deblocking_filter == 2){
1222 if(h->slice_table[top_xy ] != h->slice_num) top_type= 0;
1223 if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
1225 if(h->slice_table[top_xy ] == 0xFFFF) top_type= 0;
1226 if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
1228 h->top_type = top_type ;
1229 h->left_type[0]= left_type[0];
1230 h->left_type[1]= left_type[1];
1232 if(IS_INTRA(mb_type))
1235 AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
1236 AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
1237 AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
1238 AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
1239 AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
1241 h->cbp= h->cbp_table[mb_xy];
1245 for(list=0; list<h->list_count; list++){
1248 int16_t (*mv_dst)[2];
1249 int16_t (*mv_src)[2];
1251 if(!USES_LIST(mb_type, list)){
1252 fill_rectangle( h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
1253 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1254 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1255 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1256 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1260 ref = &s->current_picture.ref_index[list][4*mb_xy];
1262 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1263 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1264 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1266 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1267 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1270 b_stride = h->b_stride;
1271 mv_dst = &h->mv_cache[list][scan8[0]];
1272 mv_src = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
1274 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
1289 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
1291 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
1295 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
1296 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
1297 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
1298 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
1301 // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
1302 if(!CABAC && h->pps.transform_8x8_mode){
1303 if(IS_8x8DCT(top_type)){
1304 h->non_zero_count_cache[4+8*0]=
1305 h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
1306 h->non_zero_count_cache[6+8*0]=
1307 h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
1309 if(IS_8x8DCT(left_type[0])){
1310 h->non_zero_count_cache[3+8*1]=
1311 h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
1313 if(IS_8x8DCT(left_type[1])){
1314 h->non_zero_count_cache[3+8*3]=
1315 h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
1318 if(IS_8x8DCT(mb_type)){
1319 h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]=
1320 h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= h->cbp & 1;
1322 h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
1323 h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
1325 h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
1326 h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
1328 h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
1329 h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
1333 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
1335 for(list=0; list<h->list_count; list++){
1336 if(USES_LIST(top_type, list)){
1337 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1338 const int b8_xy= 4*top_xy + 2;
1339 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1340 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1341 h->ref_cache[list][scan8[0] + 0 - 1*8]=
1342 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
1343 h->ref_cache[list][scan8[0] + 2 - 1*8]=
1344 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
1346 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1347 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1350 if(!IS_INTERLACED(mb_type^left_type[0])){
1351 if(USES_LIST(left_type[0], list)){
1352 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1353 const int b8_xy= 4*left_xy[0] + 1;
1354 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1355 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
1356 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
1357 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
1358 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
1359 h->ref_cache[list][scan8[0] - 1 + 0 ]=
1360 h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
1361 h->ref_cache[list][scan8[0] - 1 +16 ]=
1362 h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
1364 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
1365 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
1366 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
1367 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
1368 h->ref_cache[list][scan8[0] - 1 + 0 ]=
1369 h->ref_cache[list][scan8[0] - 1 + 8 ]=
1370 h->ref_cache[list][scan8[0] - 1 + 16 ]=
1371 h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
1381 * gets the predicted intra4x4 prediction mode.
1383 static inline int pred_intra_mode(H264Context *h, int n){
1384 const int index8= scan8[n];
1385 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1386 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1387 const int min= FFMIN(left, top);
1389 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1391 if(min<0) return DC_PRED;
1395 static inline void write_back_non_zero_count(H264Context *h){
1396 const int mb_xy= h->mb_xy;
1398 AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1399 AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1400 AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[0+8*5]);
1401 AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8*3]);
1402 AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1405 static inline void write_back_motion(H264Context *h, int mb_type){
1406 MpegEncContext * const s = &h->s;
1407 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
1408 const int b8_xy= 4*h->mb_xy;
1411 if(!USES_LIST(mb_type, 0))
1412 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1414 for(list=0; list<h->list_count; list++){
1416 int16_t (*mv_dst)[2];
1417 int16_t (*mv_src)[2];
1419 if(!USES_LIST(mb_type, list))
1422 b_stride = h->b_stride;
1423 mv_dst = &s->current_picture.motion_val[list][b_xy];
1424 mv_src = &h->mv_cache[list][scan8[0]];
1426 AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1429 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
1430 uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1431 if(IS_SKIP(mb_type))
1432 AV_ZERO128(mvd_dst);
1434 AV_COPY64(mvd_dst, mvd_src + 8*3);
1435 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
1436 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
1437 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
1442 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1443 ref_index[0+0*2]= h->ref_cache[list][scan8[0]];
1444 ref_index[1+0*2]= h->ref_cache[list][scan8[4]];
1445 ref_index[0+1*2]= h->ref_cache[list][scan8[8]];
1446 ref_index[1+1*2]= h->ref_cache[list][scan8[12]];
1450 if(h->slice_type_nos == FF_B_TYPE && CABAC){
1451 if(IS_8X8(mb_type)){
1452 uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
1453 direct_table[1] = h->sub_mb_type[1]>>1;
1454 direct_table[2] = h->sub_mb_type[2]>>1;
1455 direct_table[3] = h->sub_mb_type[3]>>1;
1460 static inline int get_dct8x8_allowed(H264Context *h){
1461 if(h->sps.direct_8x8_inference_flag)
1462 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL));
1464 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1468 * decodes a P_SKIP or B_SKIP macroblock
1470 static void decode_mb_skip(H264Context *h){
1471 MpegEncContext * const s = &h->s;
1472 const int mb_xy= h->mb_xy;
1475 memset(h->non_zero_count[mb_xy], 0, 32);
1476 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1479 mb_type|= MB_TYPE_INTERLACED;
1481 if( h->slice_type_nos == FF_B_TYPE )
1483 // just for fill_caches. pred_direct_motion will set the real mb_type
1484 mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1485 if(h->direct_spatial_mv_pred){
1486 fill_decode_neighbors(h, mb_type);
1487 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1489 ff_h264_pred_direct_motion(h, &mb_type);
1490 mb_type|= MB_TYPE_SKIP;
1495 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1497 fill_decode_neighbors(h, mb_type);
1498 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1499 pred_pskip_motion(h, &mx, &my);
1500 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1501 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1504 write_back_motion(h, mb_type);
1505 s->current_picture.mb_type[mb_xy]= mb_type;
1506 s->current_picture.qscale_table[mb_xy]= s->qscale;
1507 h->slice_table[ mb_xy ]= h->slice_num;
1508 h->prev_mb_skipped= 1;
1511 #include "h264_mvpred.h" //For pred_pskip_motion()
1513 #endif /* AVCODEC_H264_H */