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
33 #include "mpegvideo.h"
35 #include "rectangle.h"
37 #define interlaced_dct interlaced_dct_is_a_bad_name
38 #define mb_intra mb_intra_is_not_initialized_see_mb_type
40 #define LUMA_DC_BLOCK_INDEX 25
41 #define CHROMA_DC_BLOCK_INDEX 26
43 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
44 #define COEFF_TOKEN_VLC_BITS 8
45 #define TOTAL_ZEROS_VLC_BITS 9
46 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
47 #define RUN_VLC_BITS 3
48 #define RUN7_VLC_BITS 6
50 #define MAX_SPS_COUNT 32
51 #define MAX_PPS_COUNT 256
53 #define MAX_MMCO_COUNT 66
55 #define MAX_DELAYED_PIC_COUNT 16
57 /* Compiling in interlaced support reduces the speed
58 * of progressive decoding by about 2%. */
59 #define ALLOW_INTERLACE
61 #define ALLOW_NOCHROMA
64 * The maximum number of slices supported by the decoder.
65 * must be a power of 2
69 #ifdef ALLOW_INTERLACE
70 #define MB_MBAFF h->mb_mbaff
71 #define MB_FIELD h->mb_field_decoding_flag
72 #define FRAME_MBAFF h->mb_aff_frame
73 #define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
78 #define FIELD_PICTURE 0
80 #define IS_INTERLACED(mb_type) 0
82 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
85 #define CHROMA h->sps.chroma_format_idc
91 #define CABAC h->pps.cabac
94 #define EXTENDED_SAR 255
96 #define MB_TYPE_REF0 MB_TYPE_ACPRED //dirty but it fits in 16 bit
97 #define MB_TYPE_8x8DCT 0x01000000
98 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
99 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
102 * Value of Picture.reference when Picture is not a reference picture, but
103 * is held for delayed output.
105 #define DELAYED_PIC_REF 4
123 NAL_AUXILIARY_SLICE=19
130 SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
131 SEI_TYPE_PIC_TIMING = 1, ///< picture timing
132 SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
133 SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync)
137 * pic_struct in picture timing SEI message
140 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
141 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
142 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
143 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
144 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
145 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
146 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
147 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
148 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
152 * Sequence parameter set
158 int chroma_format_idc;
159 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
160 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
161 int poc_type; ///< pic_order_cnt_type
162 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
163 int delta_pic_order_always_zero_flag;
164 int offset_for_non_ref_pic;
165 int offset_for_top_to_bottom_field;
166 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
167 int ref_frame_count; ///< num_ref_frames
168 int gaps_in_frame_num_allowed_flag;
169 int mb_width; ///< pic_width_in_mbs_minus1 + 1
170 int mb_height; ///< pic_height_in_map_units_minus1 + 1
171 int frame_mbs_only_flag;
172 int mb_aff; ///<mb_adaptive_frame_field_flag
173 int direct_8x8_inference_flag;
174 int crop; ///< frame_cropping_flag
175 unsigned int crop_left; ///< frame_cropping_rect_left_offset
176 unsigned int crop_right; ///< frame_cropping_rect_right_offset
177 unsigned int crop_top; ///< frame_cropping_rect_top_offset
178 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
179 int vui_parameters_present_flag;
181 int video_signal_type_present_flag;
183 int colour_description_present_flag;
184 enum AVColorPrimaries color_primaries;
185 enum AVColorTransferCharacteristic color_trc;
186 enum AVColorSpace colorspace;
187 int timing_info_present_flag;
188 uint32_t num_units_in_tick;
190 int fixed_frame_rate_flag;
191 short offset_for_ref_frame[256]; //FIXME dyn aloc?
192 int bitstream_restriction_flag;
193 int num_reorder_frames;
194 int scaling_matrix_present;
195 uint8_t scaling_matrix4[6][16];
196 uint8_t scaling_matrix8[2][64];
197 int nal_hrd_parameters_present_flag;
198 int vcl_hrd_parameters_present_flag;
199 int pic_struct_present_flag;
200 int time_offset_length;
201 int cpb_cnt; ///< See H.264 E.1.2
202 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1
203 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
204 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
205 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
206 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
207 int residual_color_transform_flag; ///< residual_colour_transform_flag
211 * Picture parameter set
215 int cabac; ///< entropy_coding_mode_flag
216 int pic_order_present; ///< pic_order_present_flag
217 int slice_group_count; ///< num_slice_groups_minus1 + 1
218 int mb_slice_group_map_type;
219 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
220 int weighted_pred; ///< weighted_pred_flag
221 int weighted_bipred_idc;
222 int init_qp; ///< pic_init_qp_minus26 + 26
223 int init_qs; ///< pic_init_qs_minus26 + 26
224 int chroma_qp_index_offset[2];
225 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
226 int constrained_intra_pred; ///< constrained_intra_pred_flag
227 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
228 int transform_8x8_mode; ///< transform_8x8_mode_flag
229 uint8_t scaling_matrix4[6][16];
230 uint8_t scaling_matrix8[2][64];
231 uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
236 * Memory management control operation opcode.
238 typedef enum MMCOOpcode{
249 * Memory management control operation.
253 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
254 int long_arg; ///< index, pic_num, or num long refs depending on opcode
260 typedef struct H264Context{
264 uint8_t *rbsp_buffer[2];
265 unsigned int rbsp_buffer_size[2];
268 * Used to parse AVC variant of h264
270 int is_avc; ///< this flag is != 0 if codec is avc1
271 int got_avcC; ///< flag used to parse avcC data only once
272 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
274 int chroma_qp[2]; //QPc
276 int qp_thresh; ///< QP threshold to skip loopfilter
282 int chroma_pred_mode;
283 int intra16x16_pred_mode;
291 int8_t intra4x4_pred_mode_cache[5*8];
292 int8_t (*intra4x4_pred_mode)[8];
294 unsigned int topleft_samples_available;
295 unsigned int top_samples_available;
296 unsigned int topright_samples_available;
297 unsigned int left_samples_available;
298 uint8_t (*top_borders[2])[16+2*8];
299 uint8_t left_border[2*(17+2*9)];
302 * non zero coeff count cache.
303 * is 64 if not available.
305 DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache)[6*8];
313 uint8_t (*non_zero_count)[32];
316 * Motion vector cache.
318 DECLARE_ALIGNED_16(int16_t, mv_cache)[2][5*8][2];
319 DECLARE_ALIGNED_8(int8_t, ref_cache)[2][5*8];
320 #define LIST_NOT_USED -1 //FIXME rename?
321 #define PART_NOT_AVAILABLE -2
324 * is 1 if the specific list MV&references are set to 0,0,-2.
326 int mv_cache_clean[2];
329 * number of neighbors (top and/or left) that used 8x8 dct
331 int neighbor_transform_size;
334 * block_offset[ 0..23] for frame macroblocks
335 * block_offset[24..47] for field macroblocks
337 int block_offset[2*(16+8)];
339 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
341 int b_stride; //FIXME use s->b4_stride
344 int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff
353 int unknown_svq3_flag;
354 int next_slice_index;
356 SPS *sps_buffers[MAX_SPS_COUNT];
357 SPS sps; ///< current sps
359 PPS *pps_buffers[MAX_PPS_COUNT];
363 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
365 uint32_t dequant4_buffer[6][52][16];
366 uint32_t dequant8_buffer[2][52][64];
367 uint32_t (*dequant4_coeff[6])[16];
368 uint32_t (*dequant8_coeff[2])[64];
369 int dequant_coeff_pps; ///< reinit tables when pps changes
372 uint16_t *slice_table_base;
373 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
375 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
376 int slice_type_fixed;
378 //interlacing specific flags
380 int mb_field_decoding_flag;
381 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
383 DECLARE_ALIGNED_8(uint16_t, sub_mb_type)[4];
388 int delta_poc_bottom;
391 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
392 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
393 int frame_num_offset; ///< for POC type 2
394 int prev_frame_num_offset; ///< for POC type 2
395 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
398 * frame_num for frames or 2*frame_num+1 for field pics.
403 * max_frame_num or 2*max_frame_num for field pics.
407 //Weighted pred stuff
409 int use_weight_chroma;
410 int luma_log2_weight_denom;
411 int chroma_log2_weight_denom;
412 int luma_weight[2][48];
413 int luma_offset[2][48];
414 int chroma_weight[2][48][2];
415 int chroma_offset[2][48][2];
416 int implicit_weight[48][48];
419 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
420 int slice_alpha_c0_offset;
421 int slice_beta_offset;
423 int redundant_pic_count;
425 int direct_spatial_mv_pred;
426 int dist_scale_factor[16];
427 int dist_scale_factor_field[2][32];
428 int map_col_to_list0[2][16+32];
429 int map_col_to_list0_field[2][2][16+32];
432 * num_ref_idx_l0/1_active_minus1 + 1
434 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
435 unsigned int list_count;
436 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
437 Picture *short_ref[32];
438 Picture *long_ref[32];
439 Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
440 Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
441 Reordered version of default_ref_list
442 according to picture reordering in slice header */
443 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
444 Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
448 * memory management control operations buffer.
450 MMCO mmco[MAX_MMCO_COUNT];
453 int long_ref_count; ///< number of actual long term references
454 int short_ref_count; ///< number of actual short term references
457 GetBitContext intra_gb;
458 GetBitContext inter_gb;
459 GetBitContext *intra_gb_ptr;
460 GetBitContext *inter_gb_ptr;
462 DECLARE_ALIGNED_16(DCTELEM, mb)[16*24];
463 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
469 uint8_t cabac_state[460];
472 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
477 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
478 uint8_t *chroma_pred_mode_table;
479 int last_qscale_diff;
480 int16_t (*mvd_table[2])[2];
481 DECLARE_ALIGNED_16(int16_t, mvd_cache)[2][5*8][2];
482 uint8_t *direct_table;
483 uint8_t direct_cache[5*8];
485 uint8_t zigzag_scan[16];
486 uint8_t zigzag_scan8x8[64];
487 uint8_t zigzag_scan8x8_cavlc[64];
488 uint8_t field_scan[16];
489 uint8_t field_scan8x8[64];
490 uint8_t field_scan8x8_cavlc[64];
491 const uint8_t *zigzag_scan_q0;
492 const uint8_t *zigzag_scan8x8_q0;
493 const uint8_t *zigzag_scan8x8_cavlc_q0;
494 const uint8_t *field_scan_q0;
495 const uint8_t *field_scan8x8_q0;
496 const uint8_t *field_scan8x8_cavlc_q0;
501 * @defgroup multithreading Members for slice based multithreading
504 struct H264Context *thread_context[MAX_THREADS];
507 * current slice number, used to initalize slice_num of each thread/context
512 * Max number of threads / contexts.
513 * This is equal to AVCodecContext.thread_count unless
514 * multithreaded decoding is impossible, in which case it is
520 * 1 if the single thread fallback warning has already been
521 * displayed, 0 otherwise.
523 int single_decode_warning;
530 uint32_t svq3_watermark_key;
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;
573 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
574 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
577 int sei_buffering_period_present; ///< Buffering period SEI flag
578 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
582 extern const uint8_t ff_h264_chroma_qp[52];
584 void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
586 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
591 int ff_h264_decode_sei(H264Context *h);
596 int ff_h264_decode_seq_parameter_set(H264Context *h);
601 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
604 * Decodes a network abstraction layer unit.
605 * @param consumed is the number of bytes used as input
606 * @param length is the length of the array
607 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
608 * @returns decoded bytes, might be src+1 if no escapes
610 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
613 * identifies the exact end of the bitstream
614 * @return the length of the trailing, or 0 if damaged
616 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
619 * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
621 av_cold void ff_h264_free_context(H264Context *h);
624 * reconstructs bitstream slice_type.
626 int ff_h264_get_slice_type(const H264Context *h);
632 int ff_h264_alloc_tables(H264Context *h);
635 * fills the default_ref_list.
637 int ff_h264_fill_default_ref_list(H264Context *h);
639 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
640 void ff_h264_fill_mbaff_ref_list(H264Context *h);
641 void ff_h264_remove_all_refs(H264Context *h);
644 * Executes the reference picture marking (memory management control operations).
646 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
648 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
652 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
654 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
657 * checks if the top & left blocks are available if needed & changes 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_write_back_intra_pred_mode(H264Context *h);
662 void ff_h264_hl_decode_mb(H264Context *h);
663 int ff_h264_frame_start(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 * decodes a macroblock
670 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
672 int ff_h264_decode_mb_cavlc(H264Context *h);
675 * decodes a CABAC coded macroblock
676 * @returns 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);
706 //This table must be here because scan8[constant] must be known at compiletime
707 static const uint8_t scan8[16 + 2*4]={
708 4+1*8, 5+1*8, 4+2*8, 5+2*8,
709 6+1*8, 7+1*8, 6+2*8, 7+2*8,
710 4+3*8, 5+3*8, 4+4*8, 5+4*8,
711 6+3*8, 7+3*8, 6+4*8, 7+4*8,
718 static av_always_inline uint32_t pack16to32(int a, int b){
720 return (b&0xFFFF) + (a<<16);
722 return (a&0xFFFF) + (b<<16);
727 * gets the chroma qp.
729 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
730 return h->pps.chroma_qp_table[t][qscale];
733 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
735 static void fill_decode_caches(H264Context *h, int mb_type){
736 MpegEncContext * const s = &h->s;
737 const int mb_xy= h->mb_xy;
738 int topleft_xy, top_xy, topright_xy, left_xy[2];
739 int topleft_type, top_type, topright_type, left_type[2];
740 const uint8_t * left_block;
741 int topleft_partition= -1;
743 static const uint8_t left_block_options[4][16]={
744 {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},
745 {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},
746 {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},
747 {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}
750 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
752 /* Wow, what a mess, why didn't they simplify the interlacing & intra
753 * stuff, I can't imagine that these complex rules are worth it. */
755 topleft_xy = top_xy - 1;
756 topright_xy= top_xy + 1;
757 left_xy[1] = left_xy[0] = mb_xy-1;
758 left_block = left_block_options[0];
760 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
761 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
763 if (left_mb_field_flag != curr_mb_field_flag) {
764 left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
765 if (curr_mb_field_flag) {
766 left_xy[1] += s->mb_stride;
767 left_block = left_block_options[3];
769 topleft_xy += s->mb_stride;
770 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
771 topleft_partition = 0;
772 left_block = left_block_options[1];
776 if(curr_mb_field_flag){
777 topleft_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
778 topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
779 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
781 if (left_mb_field_flag != curr_mb_field_flag) {
782 left_xy[1] = left_xy[0] = mb_xy - 1;
783 if (curr_mb_field_flag) {
784 left_xy[1] += s->mb_stride;
785 left_block = left_block_options[3];
787 left_block = left_block_options[2];
793 h->top_mb_xy = top_xy;
794 h->left_mb_xy[0] = left_xy[0];
795 h->left_mb_xy[1] = left_xy[1];
796 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
797 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
798 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
799 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
800 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
802 if(IS_INTRA(mb_type)){
803 int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
804 h->topleft_samples_available=
805 h->top_samples_available=
806 h->left_samples_available= 0xFFFF;
807 h->topright_samples_available= 0xEEEA;
809 if(!(top_type & type_mask)){
810 h->topleft_samples_available= 0xB3FF;
811 h->top_samples_available= 0x33FF;
812 h->topright_samples_available= 0x26EA;
814 if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
815 if(IS_INTERLACED(mb_type)){
816 if(!(left_type[0] & type_mask)){
817 h->topleft_samples_available&= 0xDFFF;
818 h->left_samples_available&= 0x5FFF;
820 if(!(left_type[1] & type_mask)){
821 h->topleft_samples_available&= 0xFF5F;
822 h->left_samples_available&= 0xFF5F;
825 int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
826 ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
827 assert(left_xy[0] == left_xy[1]);
828 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
829 h->topleft_samples_available&= 0xDF5F;
830 h->left_samples_available&= 0x5F5F;
834 if(!(left_type[0] & type_mask)){
835 h->topleft_samples_available&= 0xDF5F;
836 h->left_samples_available&= 0x5F5F;
840 if(!(topleft_type & type_mask))
841 h->topleft_samples_available&= 0x7FFF;
843 if(!(topright_type & type_mask))
844 h->topright_samples_available&= 0xFBFF;
846 if(IS_INTRA4x4(mb_type)){
847 if(IS_INTRA4x4(top_type)){
848 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
849 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
850 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
851 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
854 if(!(top_type & type_mask))
859 h->intra4x4_pred_mode_cache[4+8*0]=
860 h->intra4x4_pred_mode_cache[5+8*0]=
861 h->intra4x4_pred_mode_cache[6+8*0]=
862 h->intra4x4_pred_mode_cache[7+8*0]= pred;
865 if(IS_INTRA4x4(left_type[i])){
866 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
867 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
870 if(!(left_type[i] & type_mask))
875 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
876 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
891 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
893 *(uint32_t*)&h->non_zero_count_cache[4+8*0]= *(uint32_t*)&h->non_zero_count[top_xy][4+3*8];
894 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
895 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
897 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
898 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
900 h->non_zero_count_cache[1+8*0]=
901 h->non_zero_count_cache[2+8*0]=
903 h->non_zero_count_cache[1+8*3]=
904 h->non_zero_count_cache[2+8*3]=
905 *(uint32_t*)&h->non_zero_count_cache[4+8*0]= CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040;
908 for (i=0; i<2; i++) {
910 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
911 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
912 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
913 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
915 h->non_zero_count_cache[3+8*1 + 2*8*i]=
916 h->non_zero_count_cache[3+8*2 + 2*8*i]=
917 h->non_zero_count_cache[0+8*1 + 8*i]=
918 h->non_zero_count_cache[0+8*4 + 8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
925 h->top_cbp = h->cbp_table[top_xy];
926 } else if(IS_INTRA(mb_type)) {
933 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
934 } else if(IS_INTRA(mb_type)) {
940 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
943 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
948 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
950 for(list=0; list<h->list_count; list++){
951 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type)){
952 /*if(!h->mv_cache_clean[list]){
953 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
954 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
955 h->mv_cache_clean[list]= 1;
959 h->mv_cache_clean[list]= 0;
961 if(USES_LIST(top_type, list)){
962 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
963 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
964 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
965 h->ref_cache[list][scan8[0] + 0 - 1*8]=
966 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
967 h->ref_cache[list][scan8[0] + 2 - 1*8]=
968 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
970 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
971 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
975 int cache_idx = scan8[0] - 1 + i*2*8;
976 if(USES_LIST(left_type[i], list)){
977 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
978 const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
979 *(uint32_t*)h->mv_cache[list][cache_idx ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]];
980 *(uint32_t*)h->mv_cache[list][cache_idx+8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]];
981 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
982 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
984 *(uint32_t*)h->mv_cache [list][cache_idx ]=
985 *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
986 h->ref_cache[list][cache_idx ]=
987 h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
991 if((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF)
994 if(USES_LIST(topleft_type, list)){
995 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
996 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
997 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
998 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1000 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
1001 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1004 if(USES_LIST(topright_type, list)){
1005 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1006 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
1007 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
1008 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1010 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
1011 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1014 if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
1017 h->ref_cache[list][scan8[5 ]+1] =
1018 h->ref_cache[list][scan8[7 ]+1] =
1019 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
1020 h->ref_cache[list][scan8[4 ]] =
1021 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1022 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
1023 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
1024 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
1025 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
1026 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
1029 /* XXX beurk, Load mvd */
1030 if(USES_LIST(top_type, list)){
1031 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1032 AV_COPY128(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1034 AV_ZERO128(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1036 if(USES_LIST(left_type[0], list)){
1037 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1038 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
1039 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
1041 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
1042 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
1044 if(USES_LIST(left_type[1], list)){
1045 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
1046 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
1047 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
1049 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
1050 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
1052 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
1053 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
1054 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
1055 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
1056 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
1058 if(h->slice_type_nos == FF_B_TYPE){
1059 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
1061 if(IS_DIRECT(top_type)){
1062 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
1063 }else if(IS_8X8(top_type)){
1064 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
1065 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
1066 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
1068 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
1071 if(IS_DIRECT(left_type[0]))
1072 h->direct_cache[scan8[0] - 1 + 0*8]= 1;
1073 else if(IS_8X8(left_type[0]))
1074 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
1076 h->direct_cache[scan8[0] - 1 + 0*8]= 0;
1078 if(IS_DIRECT(left_type[1]))
1079 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
1080 else if(IS_8X8(left_type[1]))
1081 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
1083 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
1089 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1090 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1091 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1092 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1093 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1094 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1095 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1096 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1097 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1098 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1100 #define MAP_F2F(idx, mb_type)\
1101 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1102 h->ref_cache[list][idx] <<= 1;\
1103 h->mv_cache[list][idx][1] /= 2;\
1104 h->mvd_cache[list][idx][1] /= 2;\
1109 #define MAP_F2F(idx, mb_type)\
1110 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1111 h->ref_cache[list][idx] >>= 1;\
1112 h->mv_cache[list][idx][1] <<= 1;\
1113 h->mvd_cache[list][idx][1] <<= 1;\
1123 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1128 * @returns non zero if the loop filter can be skiped
1130 static int fill_filter_caches(H264Context *h, int mb_type){
1131 MpegEncContext * const s = &h->s;
1132 const int mb_xy= h->mb_xy;
1133 int top_xy, left_xy[2];
1134 int top_type, left_type[2];
1137 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
1139 //FIXME deblocking could skip the intra and nnz parts.
1141 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1142 * stuff, I can't imagine that these complex rules are worth it. */
1144 left_xy[1] = left_xy[0] = mb_xy-1;
1146 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
1147 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1149 if (left_mb_field_flag != curr_mb_field_flag) {
1150 left_xy[0] -= s->mb_stride;
1153 if(curr_mb_field_flag){
1154 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
1156 if (left_mb_field_flag != curr_mb_field_flag) {
1157 left_xy[1] += s->mb_stride;
1162 h->top_mb_xy = top_xy;
1163 h->left_mb_xy[0] = left_xy[0];
1164 h->left_mb_xy[1] = left_xy[1];
1166 //for sufficiently low qp, filtering wouldn't do anything
1167 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
1168 int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
1169 int qp = s->current_picture.qscale_table[mb_xy];
1171 && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
1172 && (top_xy < 0 || ((qp + s->current_picture.qscale_table[top_xy ] + 1)>>1) <= qp_thresh)){
1175 if( (left_xy[0]< 0 || ((qp + s->current_picture.qscale_table[left_xy[1] ] + 1)>>1) <= qp_thresh)
1176 && (top_xy < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy -s->mb_stride] + 1)>>1) <= qp_thresh))
1181 if(h->deblocking_filter == 2){
1182 h->top_type = top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
1183 h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
1184 h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
1186 h->top_type = top_type = h->slice_table[top_xy ] < 0xFFFF ? s->current_picture.mb_type[top_xy] : 0;
1187 h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
1188 h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
1190 if(IS_INTRA(mb_type))
1193 AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
1194 AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
1195 *((uint32_t*)&h->non_zero_count_cache[0+8*5])= *((uint32_t*)&h->non_zero_count[mb_xy][16]);
1196 *((uint32_t*)&h->non_zero_count_cache[4+8*3])= *((uint32_t*)&h->non_zero_count[mb_xy][20]);
1197 AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
1199 h->cbp= h->cbp_table[mb_xy];
1203 for(list=0; list<h->list_count; list++){
1206 int16_t (*mv_dst)[2];
1207 int16_t (*mv_src)[2];
1209 if(!USES_LIST(mb_type, list)){
1210 fill_rectangle( h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
1211 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
1212 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] =
1213 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
1214 *(uint32_t*)&h->ref_cache[list][scan8[10]] = ((LIST_NOT_USED)&0xFF)*0x01010101;
1218 ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
1220 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1221 *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
1222 *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
1223 ref += h->b8_stride;
1224 *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
1225 *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
1228 b_stride = h->b_stride;
1229 mv_dst = &h->mv_cache[list][scan8[0]];
1230 mv_src = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
1232 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
1247 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
1249 *(uint32_t*)&h->non_zero_count_cache[4+8*0]= *(uint32_t*)&h->non_zero_count[top_xy][4+3*8];
1253 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
1254 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
1255 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
1256 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
1259 // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
1260 if(!CABAC && h->pps.transform_8x8_mode){
1261 if(IS_8x8DCT(top_type)){
1262 h->non_zero_count_cache[4+8*0]=
1263 h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
1264 h->non_zero_count_cache[6+8*0]=
1265 h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
1267 if(IS_8x8DCT(left_type[0])){
1268 h->non_zero_count_cache[3+8*1]=
1269 h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
1271 if(IS_8x8DCT(left_type[1])){
1272 h->non_zero_count_cache[3+8*3]=
1273 h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
1276 if(IS_8x8DCT(mb_type)){
1277 h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]=
1278 h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= h->cbp & 1;
1280 h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
1281 h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
1283 h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
1284 h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
1286 h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
1287 h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
1291 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
1293 for(list=0; list<h->list_count; list++){
1294 if(USES_LIST(top_type, list)){
1295 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1296 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
1297 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1298 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1299 h->ref_cache[list][scan8[0] + 0 - 1*8]=
1300 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
1301 h->ref_cache[list][scan8[0] + 2 - 1*8]=
1302 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
1304 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1305 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((LIST_NOT_USED)&0xFF)*0x01010101;
1308 if(!IS_INTERLACED(mb_type^left_type[0])){
1309 if(USES_LIST(left_type[0], list)){
1310 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1311 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
1312 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1313 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 0 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*0];
1314 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 8 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*1];
1315 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 +16 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*2];
1316 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 +24 ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*3];
1317 h->ref_cache[list][scan8[0] - 1 + 0 ]=
1318 h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + h->b8_stride*0]];
1319 h->ref_cache[list][scan8[0] - 1 +16 ]=
1320 h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + h->b8_stride*1]];
1322 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0 ]=
1323 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 8 ]=
1324 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 +16 ]=
1325 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 +24 ]= 0;
1326 h->ref_cache[list][scan8[0] - 1 + 0 ]=
1327 h->ref_cache[list][scan8[0] - 1 + 8 ]=
1328 h->ref_cache[list][scan8[0] - 1 + 16 ]=
1329 h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
1339 * gets the predicted intra4x4 prediction mode.
1341 static inline int pred_intra_mode(H264Context *h, int n){
1342 const int index8= scan8[n];
1343 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1344 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1345 const int min= FFMIN(left, top);
1347 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1349 if(min<0) return DC_PRED;
1353 static inline void write_back_non_zero_count(H264Context *h){
1354 const int mb_xy= h->mb_xy;
1356 AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1357 AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1358 *((uint32_t*)&h->non_zero_count[mb_xy][16]) = *((uint32_t*)&h->non_zero_count_cache[0+8*5]);
1359 *((uint32_t*)&h->non_zero_count[mb_xy][20]) = *((uint32_t*)&h->non_zero_count_cache[4+8*3]);
1360 AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1363 static inline void write_back_motion(H264Context *h, int mb_type){
1364 MpegEncContext * const s = &h->s;
1365 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1366 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1369 if(!USES_LIST(mb_type, 0))
1370 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1372 for(list=0; list<h->list_count; list++){
1374 int16_t (*mv_dst)[2];
1375 int16_t (*mv_src)[2];
1377 if(!USES_LIST(mb_type, list))
1380 b_stride = h->b_stride;
1381 mv_dst = &s->current_picture.motion_val[list][b_xy];
1382 mv_src = &h->mv_cache[list][scan8[0]];
1384 AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1387 int16_t (*mvd_dst)[2] = &h->mvd_table[list][b_xy];
1388 int16_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1389 if(IS_SKIP(mb_type))
1390 fill_rectangle(mvd_dst, 4, 4, h->b_stride, 0, 4);
1393 AV_COPY128(mvd_dst + y*b_stride, mvd_src + 8*y);
1398 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1399 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1400 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1401 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1402 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1406 if(h->slice_type_nos == FF_B_TYPE && CABAC){
1407 if(IS_8X8(mb_type)){
1408 uint8_t *direct_table = &h->direct_table[b8_xy];
1409 direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1410 direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1411 direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1416 static inline int get_dct8x8_allowed(H264Context *h){
1417 if(h->sps.direct_8x8_inference_flag)
1418 return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL));
1420 return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1423 static void predict_field_decoding_flag(H264Context *h){
1424 MpegEncContext * const s = &h->s;
1425 const int mb_xy= h->mb_xy;
1426 int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
1427 ? s->current_picture.mb_type[mb_xy-1]
1428 : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
1429 ? s->current_picture.mb_type[mb_xy-s->mb_stride]
1431 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
1435 * decodes a P_SKIP or B_SKIP macroblock
1437 static void decode_mb_skip(H264Context *h){
1438 MpegEncContext * const s = &h->s;
1439 const int mb_xy= h->mb_xy;
1442 memset(h->non_zero_count[mb_xy], 0, 32);
1443 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1446 mb_type|= MB_TYPE_INTERLACED;
1448 if( h->slice_type_nos == FF_B_TYPE )
1450 // just for fill_caches. pred_direct_motion will set the real mb_type
1451 mb_type|= MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1453 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1454 ff_h264_pred_direct_motion(h, &mb_type);
1455 mb_type|= MB_TYPE_SKIP;
1460 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1462 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1463 pred_pskip_motion(h, &mx, &my);
1464 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1465 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1468 write_back_motion(h, mb_type);
1469 s->current_picture.mb_type[mb_xy]= mb_type;
1470 s->current_picture.qscale_table[mb_xy]= s->qscale;
1471 h->slice_table[ mb_xy ]= h->slice_num;
1472 h->prev_mb_skipped= 1;
1475 #include "h264_mvpred.h" //For pred_pskip_motion()
1477 #endif /* AVCODEC_H264_H */