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"
37 #include "rectangle.h"
39 #define interlaced_dct interlaced_dct_is_a_bad_name
40 #define mb_intra mb_intra_is_not_initialized_see_mb_type
42 #define LUMA_DC_BLOCK_INDEX 25
43 #define CHROMA_DC_BLOCK_INDEX 26
45 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
46 #define COEFF_TOKEN_VLC_BITS 8
47 #define TOTAL_ZEROS_VLC_BITS 9
48 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
49 #define RUN_VLC_BITS 3
50 #define RUN7_VLC_BITS 6
52 #define MAX_SPS_COUNT 32
53 #define MAX_PPS_COUNT 256
55 #define MAX_MMCO_COUNT 66
57 #define MAX_DELAYED_PIC_COUNT 16
59 /* Compiling in interlaced support reduces the speed
60 * of progressive decoding by about 2%. */
61 #define ALLOW_INTERLACE
63 #define ALLOW_NOCHROMA
68 * The maximum number of slices supported by the decoder.
69 * must be a power of 2
73 #ifdef ALLOW_INTERLACE
74 #define MB_MBAFF h->mb_mbaff
75 #define MB_FIELD h->mb_field_decoding_flag
76 #define FRAME_MBAFF h->mb_aff_frame
77 #define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
82 #define FIELD_PICTURE 0
84 #define IS_INTERLACED(mb_type) 0
86 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
89 #define CHROMA h->sps.chroma_format_idc
95 #define CABAC h->pps.cabac
98 #define EXTENDED_SAR 255
100 #define MB_TYPE_REF0 MB_TYPE_ACPRED //dirty but it fits in 16 bit
101 #define MB_TYPE_8x8DCT 0x01000000
102 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
103 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
106 * Value of Picture.reference when Picture is not a reference picture, but
107 * is held for delayed output.
109 #define DELAYED_PIC_REF 4
127 NAL_AUXILIARY_SLICE=19
134 SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
135 SEI_TYPE_PIC_TIMING = 1, ///< picture timing
136 SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
137 SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync)
141 * pic_struct in picture timing SEI message
144 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
145 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
146 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
147 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
148 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
149 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
150 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
151 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
152 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
156 * Sequence parameter set
162 int chroma_format_idc;
163 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
164 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
165 int poc_type; ///< pic_order_cnt_type
166 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
167 int delta_pic_order_always_zero_flag;
168 int offset_for_non_ref_pic;
169 int offset_for_top_to_bottom_field;
170 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
171 int ref_frame_count; ///< num_ref_frames
172 int gaps_in_frame_num_allowed_flag;
173 int mb_width; ///< pic_width_in_mbs_minus1 + 1
174 int mb_height; ///< pic_height_in_map_units_minus1 + 1
175 int frame_mbs_only_flag;
176 int mb_aff; ///<mb_adaptive_frame_field_flag
177 int direct_8x8_inference_flag;
178 int crop; ///< frame_cropping_flag
179 unsigned int crop_left; ///< frame_cropping_rect_left_offset
180 unsigned int crop_right; ///< frame_cropping_rect_right_offset
181 unsigned int crop_top; ///< frame_cropping_rect_top_offset
182 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
183 int vui_parameters_present_flag;
185 int video_signal_type_present_flag;
187 int colour_description_present_flag;
188 enum AVColorPrimaries color_primaries;
189 enum AVColorTransferCharacteristic color_trc;
190 enum AVColorSpace colorspace;
191 int timing_info_present_flag;
192 uint32_t num_units_in_tick;
194 int fixed_frame_rate_flag;
195 short offset_for_ref_frame[256]; //FIXME dyn aloc?
196 int bitstream_restriction_flag;
197 int num_reorder_frames;
198 int scaling_matrix_present;
199 uint8_t scaling_matrix4[6][16];
200 uint8_t scaling_matrix8[2][64];
201 int nal_hrd_parameters_present_flag;
202 int vcl_hrd_parameters_present_flag;
203 int pic_struct_present_flag;
204 int time_offset_length;
205 int cpb_cnt; ///< See H.264 E.1.2
206 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1
207 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
208 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
209 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
210 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
211 int residual_color_transform_flag; ///< residual_colour_transform_flag
215 * Picture parameter set
219 int cabac; ///< entropy_coding_mode_flag
220 int pic_order_present; ///< pic_order_present_flag
221 int slice_group_count; ///< num_slice_groups_minus1 + 1
222 int mb_slice_group_map_type;
223 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
224 int weighted_pred; ///< weighted_pred_flag
225 int weighted_bipred_idc;
226 int init_qp; ///< pic_init_qp_minus26 + 26
227 int init_qs; ///< pic_init_qs_minus26 + 26
228 int chroma_qp_index_offset[2];
229 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
230 int constrained_intra_pred; ///< constrained_intra_pred_flag
231 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
232 int transform_8x8_mode; ///< transform_8x8_mode_flag
233 uint8_t scaling_matrix4[6][16];
234 uint8_t scaling_matrix8[2][64];
235 uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
240 * Memory management control operation opcode.
242 typedef enum MMCOOpcode{
253 * Memory management control operation.
257 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
258 int long_arg; ///< index, pic_num, or num long refs depending on opcode
264 typedef struct H264Context{
266 H264DSPContext h264dsp;
267 int chroma_qp[2]; //QPc
269 int qp_thresh; ///< QP threshold to skip loopfilter
275 int chroma_pred_mode;
276 int intra16x16_pred_mode;
288 const uint8_t * left_block;
289 int topleft_partition;
291 int8_t intra4x4_pred_mode_cache[5*8];
292 int8_t (*intra4x4_pred_mode);
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];
301 * non zero coeff count cache.
302 * is 64 if not available.
304 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[6*8];
312 uint8_t (*non_zero_count)[32];
315 * Motion vector cache.
317 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5*8][2];
318 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5*8];
319 #define LIST_NOT_USED -1 //FIXME rename?
320 #define PART_NOT_AVAILABLE -2
323 * is 1 if the specific list MV&references are set to 0,0,-2.
325 int mv_cache_clean[2];
328 * number of neighbors (top and/or left) that used 8x8 dct
330 int neighbor_transform_size;
333 * block_offset[ 0..23] for frame macroblocks
334 * block_offset[24..47] for field macroblocks
336 int block_offset[2*(16+8)];
338 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
340 int b_stride; //FIXME use s->b4_stride
342 int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff
348 SPS sps; ///< current sps
353 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
355 uint32_t dequant4_buffer[6][52][16]; //FIXME should these be moved down?
356 uint32_t dequant8_buffer[2][52][64];
357 uint32_t (*dequant4_coeff[6])[16];
358 uint32_t (*dequant8_coeff[2])[64];
361 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
363 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
364 int slice_type_fixed;
366 //interlacing specific flags
368 int mb_field_decoding_flag;
369 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
371 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
373 //Weighted pred stuff
375 int use_weight_chroma;
376 int luma_log2_weight_denom;
377 int chroma_log2_weight_denom;
378 //The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
379 int luma_weight[48][2][2];
380 int chroma_weight[48][2][2][2];
381 int implicit_weight[48][48][2];
383 int direct_spatial_mv_pred;
386 int dist_scale_factor[16];
387 int dist_scale_factor_field[2][32];
388 int map_col_to_list0[2][16+32];
389 int map_col_to_list0_field[2][2][16+32];
392 * num_ref_idx_l0/1_active_minus1 + 1
394 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
395 unsigned int list_count;
396 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
397 Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
398 Reordered version of default_ref_list
399 according to picture reordering in slice header */
400 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
403 GetBitContext intra_gb;
404 GetBitContext inter_gb;
405 GetBitContext *intra_gb_ptr;
406 GetBitContext *inter_gb_ptr;
408 DECLARE_ALIGNED(16, DCTELEM, mb)[16*24];
409 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
415 uint8_t cabac_state[460];
417 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
422 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
423 uint8_t *chroma_pred_mode_table;
424 int last_qscale_diff;
425 uint8_t (*mvd_table[2])[2];
426 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5*8][2];
427 uint8_t *direct_table;
428 uint8_t direct_cache[5*8];
430 uint8_t zigzag_scan[16];
431 uint8_t zigzag_scan8x8[64];
432 uint8_t zigzag_scan8x8_cavlc[64];
433 uint8_t field_scan[16];
434 uint8_t field_scan8x8[64];
435 uint8_t field_scan8x8_cavlc[64];
436 const uint8_t *zigzag_scan_q0;
437 const uint8_t *zigzag_scan8x8_q0;
438 const uint8_t *zigzag_scan8x8_cavlc_q0;
439 const uint8_t *field_scan_q0;
440 const uint8_t *field_scan8x8_q0;
441 const uint8_t *field_scan8x8_cavlc_q0;
450 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
451 int slice_alpha_c0_offset;
452 int slice_beta_offset;
454 //=============================================================
455 //Things below are not used in the MB or more inner code
459 uint8_t *rbsp_buffer[2];
460 unsigned int rbsp_buffer_size[2];
463 * Used to parse AVC variant of h264
465 int is_avc; ///< this flag is != 0 if codec is avc1
466 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
468 SPS *sps_buffers[MAX_SPS_COUNT];
469 PPS *pps_buffers[MAX_PPS_COUNT];
471 int dequant_coeff_pps; ///< reinit tables when pps changes
473 uint16_t *slice_table_base;
479 int delta_poc_bottom;
482 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
483 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
484 int frame_num_offset; ///< for POC type 2
485 int prev_frame_num_offset; ///< for POC type 2
486 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
489 * frame_num for frames or 2*frame_num+1 for field pics.
494 * max_frame_num or 2*max_frame_num for field pics.
498 int redundant_pic_count;
500 Picture *short_ref[32];
501 Picture *long_ref[32];
502 Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
503 Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
507 * memory management control operations buffer.
509 MMCO mmco[MAX_MMCO_COUNT];
512 int long_ref_count; ///< number of actual long term references
513 int short_ref_count; ///< number of actual short term references
518 * @defgroup multithreading Members for slice based multithreading
521 struct H264Context *thread_context[MAX_THREADS];
524 * current slice number, used to initalize slice_num of each thread/context
529 * Max number of threads / contexts.
530 * This is equal to AVCodecContext.thread_count unless
531 * multithreaded decoding is impossible, in which case it is
537 * 1 if the single thread fallback warning has already been
538 * displayed, 0 otherwise.
540 int single_decode_warning;
546 * pic_struct in picture timing SEI message
548 SEI_PicStructType sei_pic_struct;
551 * Complement sei_pic_struct
552 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
553 * However, soft telecined frames may have these values.
554 * This is used in an attempt to flag soft telecine progressive.
556 int prev_interlaced_frame;
559 * Bit set of clock types for fields/frames in picture timing SEI message.
560 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
566 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
568 int sei_dpb_output_delay;
571 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
573 int sei_cpb_removal_delay;
576 * recovery_frame_cnt from SEI message
578 * Set to -1 if no recovery point SEI message found or to number of frames
579 * before playback synchronizes. Frames having recovery point are key
582 int sei_recovery_frame_cnt;
584 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
585 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
588 int sei_buffering_period_present; ///< Buffering period SEI flag
589 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
591 //SVQ3 specific fields
594 int unknown_svq3_flag;
595 int next_slice_index;
596 uint32_t svq3_watermark_key;
600 extern const uint8_t ff_h264_chroma_qp[52];
602 void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
604 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
609 int ff_h264_decode_sei(H264Context *h);
614 int ff_h264_decode_seq_parameter_set(H264Context *h);
619 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
622 * Decodes a network abstraction layer unit.
623 * @param consumed is the number of bytes used as input
624 * @param length is the length of the array
625 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
626 * @return decoded bytes, might be src+1 if no escapes
628 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
631 * identifies the exact end of the bitstream
632 * @return the length of the trailing, or 0 if damaged
634 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
637 * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
639 av_cold void ff_h264_free_context(H264Context *h);
642 * reconstructs bitstream slice_type.
644 int ff_h264_get_slice_type(const H264Context *h);
650 int ff_h264_alloc_tables(H264Context *h);
653 * fills the default_ref_list.
655 int ff_h264_fill_default_ref_list(H264Context *h);
657 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
658 void ff_h264_fill_mbaff_ref_list(H264Context *h);
659 void ff_h264_remove_all_refs(H264Context *h);
662 * Executes the reference picture marking (memory management control operations).
664 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
666 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
670 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
672 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
675 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
677 int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
679 void ff_h264_write_back_intra_pred_mode(H264Context *h);
680 void ff_h264_hl_decode_mb(H264Context *h);
681 int ff_h264_frame_start(H264Context *h);
682 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
683 av_cold int ff_h264_decode_end(AVCodecContext *avctx);
684 av_cold void ff_h264_decode_init_vlc(void);
687 * decodes a macroblock
688 * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
690 int ff_h264_decode_mb_cavlc(H264Context *h);
693 * decodes a CABAC coded macroblock
694 * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
696 int ff_h264_decode_mb_cabac(H264Context *h);
698 void ff_h264_init_cabac_states(H264Context *h);
700 void ff_h264_direct_dist_scale_factor(H264Context * const h);
701 void ff_h264_direct_ref_list_init(H264Context * const h);
702 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
704 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);
705 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);
708 * Reset SEI values at the beginning of the frame.
710 * @param h H.264 context.
712 void ff_h264_reset_sei(H264Context *h);
724 //This table must be here because scan8[constant] must be known at compiletime
725 static const uint8_t scan8[16 + 2*4]={
726 4+1*8, 5+1*8, 4+2*8, 5+2*8,
727 6+1*8, 7+1*8, 6+2*8, 7+2*8,
728 4+3*8, 5+3*8, 4+4*8, 5+4*8,
729 6+3*8, 7+3*8, 6+4*8, 7+4*8,
736 static av_always_inline uint32_t pack16to32(int a, int b){
738 return (b&0xFFFF) + (a<<16);
740 return (a&0xFFFF) + (b<<16);
744 static av_always_inline uint16_t pack8to16(int a, int b){
746 return (b&0xFF) + (a<<8);
748 return (a&0xFF) + (b<<8);
753 * gets the chroma qp.
755 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
756 return h->pps.chroma_qp_table[t][qscale];
759 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
761 static void fill_decode_neighbors(H264Context *h, int mb_type){
762 MpegEncContext * const s = &h->s;
763 const int mb_xy= h->mb_xy;
764 int topleft_xy, top_xy, topright_xy, left_xy[2];
765 static const uint8_t left_block_options[4][16]={
766 {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},
767 {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},
768 {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},
769 {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}
772 h->topleft_partition= -1;
774 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
776 /* Wow, what a mess, why didn't they simplify the interlacing & intra
777 * stuff, I can't imagine that these complex rules are worth it. */
779 topleft_xy = top_xy - 1;
780 topright_xy= top_xy + 1;
781 left_xy[1] = left_xy[0] = mb_xy-1;
782 h->left_block = left_block_options[0];
784 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
785 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
787 if (left_mb_field_flag != curr_mb_field_flag) {
788 left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
789 if (curr_mb_field_flag) {
790 left_xy[1] += s->mb_stride;
791 h->left_block = left_block_options[3];
793 topleft_xy += s->mb_stride;
794 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
795 h->topleft_partition = 0;
796 h->left_block = left_block_options[1];
800 if(curr_mb_field_flag){
801 topleft_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
802 topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
803 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
805 if (left_mb_field_flag != curr_mb_field_flag) {
806 if (curr_mb_field_flag) {
807 left_xy[1] += s->mb_stride;
808 h->left_block = left_block_options[3];
810 h->left_block = left_block_options[2];
816 h->topleft_mb_xy = topleft_xy;
817 h->top_mb_xy = top_xy;
818 h->topright_mb_xy= topright_xy;
819 h->left_mb_xy[0] = left_xy[0];
820 h->left_mb_xy[1] = left_xy[1];
821 //FIXME do we need all in the context?
823 h->topleft_type = s->current_picture.mb_type[topleft_xy] ;
824 h->top_type = s->current_picture.mb_type[top_xy] ;
825 h->topright_type= s->current_picture.mb_type[topright_xy];
826 h->left_type[0] = s->current_picture.mb_type[left_xy[0]] ;
827 h->left_type[1] = s->current_picture.mb_type[left_xy[1]] ;
830 if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0;
831 if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0;
832 if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
834 if(h->slice_table[topleft_xy ] != h->slice_num){
836 if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0;
837 if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
840 if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
843 static void fill_decode_caches(H264Context *h, int mb_type){
844 MpegEncContext * const s = &h->s;
845 int topleft_xy, top_xy, topright_xy, left_xy[2];
846 int topleft_type, top_type, topright_type, left_type[2];
847 const uint8_t * left_block= h->left_block;
850 topleft_xy = h->topleft_mb_xy ;
851 top_xy = h->top_mb_xy ;
852 topright_xy = h->topright_mb_xy;
853 left_xy[0] = h->left_mb_xy[0] ;
854 left_xy[1] = h->left_mb_xy[1] ;
855 topleft_type = h->topleft_type ;
856 top_type = h->top_type ;
857 topright_type= h->topright_type ;
858 left_type[0] = h->left_type[0] ;
859 left_type[1] = h->left_type[1] ;
861 if(!IS_SKIP(mb_type)){
862 if(IS_INTRA(mb_type)){
863 int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
864 h->topleft_samples_available=
865 h->top_samples_available=
866 h->left_samples_available= 0xFFFF;
867 h->topright_samples_available= 0xEEEA;
869 if(!(top_type & type_mask)){
870 h->topleft_samples_available= 0xB3FF;
871 h->top_samples_available= 0x33FF;
872 h->topright_samples_available= 0x26EA;
874 if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
875 if(IS_INTERLACED(mb_type)){
876 if(!(left_type[0] & type_mask)){
877 h->topleft_samples_available&= 0xDFFF;
878 h->left_samples_available&= 0x5FFF;
880 if(!(left_type[1] & type_mask)){
881 h->topleft_samples_available&= 0xFF5F;
882 h->left_samples_available&= 0xFF5F;
885 int left_typei = s->current_picture.mb_type[left_xy[0] + s->mb_stride];
887 assert(left_xy[0] == left_xy[1]);
888 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
889 h->topleft_samples_available&= 0xDF5F;
890 h->left_samples_available&= 0x5F5F;
894 if(!(left_type[0] & type_mask)){
895 h->topleft_samples_available&= 0xDF5F;
896 h->left_samples_available&= 0x5F5F;
900 if(!(topleft_type & type_mask))
901 h->topleft_samples_available&= 0x7FFF;
903 if(!(topright_type & type_mask))
904 h->topright_samples_available&= 0xFBFF;
906 if(IS_INTRA4x4(mb_type)){
907 if(IS_INTRA4x4(top_type)){
908 AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]);
910 h->intra4x4_pred_mode_cache[4+8*0]=
911 h->intra4x4_pred_mode_cache[5+8*0]=
912 h->intra4x4_pred_mode_cache[6+8*0]=
913 h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask);
916 if(IS_INTRA4x4(left_type[i])){
917 int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]];
918 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]];
919 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]];
921 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
922 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[i] & type_mask);
937 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
939 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
940 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
941 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
943 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
944 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
946 h->non_zero_count_cache[1+8*0]=
947 h->non_zero_count_cache[2+8*0]=
949 h->non_zero_count_cache[1+8*3]=
950 h->non_zero_count_cache[2+8*3]=
951 AV_WN32A(&h->non_zero_count_cache[4+8*0], CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040);
954 for (i=0; i<2; i++) {
956 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
957 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
958 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
959 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
961 h->non_zero_count_cache[3+8*1 + 2*8*i]=
962 h->non_zero_count_cache[3+8*2 + 2*8*i]=
963 h->non_zero_count_cache[0+8*1 + 8*i]=
964 h->non_zero_count_cache[0+8*4 + 8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
971 h->top_cbp = h->cbp_table[top_xy];
973 h->top_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
977 h->left_cbp = (h->cbp_table[left_xy[0]] & 0x1f0)
978 | ((h->cbp_table[left_xy[0]]>>(left_block[0]&(~1)))&2)
979 | (((h->cbp_table[left_xy[1]]>>(left_block[2]&(~1)))&2) << 2);
981 h->left_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
987 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
989 for(list=0; list<h->list_count; list++){
990 if(!USES_LIST(mb_type, list)){
991 /*if(!h->mv_cache_clean[list]){
992 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
993 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
994 h->mv_cache_clean[list]= 1;
998 assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
1000 h->mv_cache_clean[list]= 0;
1002 if(USES_LIST(top_type, list)){
1003 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1004 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1005 h->ref_cache[list][scan8[0] + 0 - 1*8]=
1006 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 2];
1007 h->ref_cache[list][scan8[0] + 2 - 1*8]=
1008 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 3];
1010 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1011 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
1014 if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){
1016 int cache_idx = scan8[0] - 1 + i*2*8;
1017 if(USES_LIST(left_type[i], list)){
1018 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1019 const int b8_xy= 4*left_xy[i] + 1;
1020 AV_COPY32(h->mv_cache[list][cache_idx ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]);
1021 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]]);
1022 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + (left_block[0+i*2]&~1)];
1023 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + (left_block[1+i*2]&~1)];
1025 AV_ZERO32(h->mv_cache [list][cache_idx ]);
1026 AV_ZERO32(h->mv_cache [list][cache_idx+8]);
1027 h->ref_cache[list][cache_idx ]=
1028 h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1032 if(USES_LIST(left_type[0], list)){
1033 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1034 const int b8_xy= 4*left_xy[0] + 1;
1035 AV_COPY32(h->mv_cache[list][scan8[0] - 1], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]]);
1036 h->ref_cache[list][scan8[0] - 1]= s->current_picture.ref_index[list][b8_xy + (left_block[0]&~1)];
1038 AV_ZERO32(h->mv_cache [list][scan8[0] - 1]);
1039 h->ref_cache[list][scan8[0] - 1]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1043 if(USES_LIST(topright_type, list)){
1044 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1045 AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]);
1046 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][4*topright_xy + 2];
1048 AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]);
1049 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1051 if(h->ref_cache[list][scan8[0] + 4 - 1*8] < 0){
1052 if(USES_LIST(topleft_type, list)){
1053 const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
1054 const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2);
1055 AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]);
1056 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1058 AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]);
1059 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1063 if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
1066 if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
1067 h->ref_cache[list][scan8[4 ]] =
1068 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1069 AV_ZERO32(h->mv_cache [list][scan8[4 ]]);
1070 AV_ZERO32(h->mv_cache [list][scan8[12]]);
1073 /* XXX beurk, Load mvd */
1074 if(USES_LIST(top_type, list)){
1075 const int b_xy= h->mb2br_xy[top_xy];
1076 AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1078 AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1080 if(USES_LIST(left_type[0], list)){
1081 const int b_xy= h->mb2br_xy[left_xy[0]] + 6;
1082 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy - left_block[0]]);
1083 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy - left_block[1]]);
1085 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]);
1086 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]);
1088 if(USES_LIST(left_type[1], list)){
1089 const int b_xy= h->mb2br_xy[left_xy[1]] + 6;
1090 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy - left_block[2]]);
1091 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy - left_block[3]]);
1093 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]);
1094 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]);
1096 AV_ZERO16(h->mvd_cache [list][scan8[4 ]]);
1097 AV_ZERO16(h->mvd_cache [list][scan8[12]]);
1098 if(h->slice_type_nos == FF_B_TYPE){
1099 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
1101 if(IS_DIRECT(top_type)){
1102 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101u*(MB_TYPE_DIRECT2>>1));
1103 }else if(IS_8X8(top_type)){
1104 int b8_xy = 4*top_xy;
1105 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy + 2];
1106 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 3];
1108 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1));
1111 if(IS_DIRECT(left_type[0]))
1112 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
1113 else if(IS_8X8(left_type[0]))
1114 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[4*left_xy[0] + 1 + (left_block[0]&~1)];
1116 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
1118 if(IS_DIRECT(left_type[1]))
1119 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
1120 else if(IS_8X8(left_type[1]))
1121 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[4*left_xy[1] + 1 + (left_block[2]&~1)];
1123 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
1129 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1130 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1131 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1132 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1133 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1134 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1135 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1136 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1137 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1138 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1140 #define MAP_F2F(idx, mb_type)\
1141 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1142 h->ref_cache[list][idx] <<= 1;\
1143 h->mv_cache[list][idx][1] /= 2;\
1144 h->mvd_cache[list][idx][1] >>=1;\
1149 #define MAP_F2F(idx, mb_type)\
1150 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1151 h->ref_cache[list][idx] >>= 1;\
1152 h->mv_cache[list][idx][1] <<= 1;\
1153 h->mvd_cache[list][idx][1] <<= 1;\
1163 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1168 * @return non zero if the loop filter can be skiped
1170 static int fill_filter_caches(H264Context *h, int mb_type){
1171 MpegEncContext * const s = &h->s;
1172 const int mb_xy= h->mb_xy;
1173 int top_xy, left_xy[2];
1174 int top_type, left_type[2];
1176 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
1178 //FIXME deblocking could skip the intra and nnz parts.
1180 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1181 * stuff, I can't imagine that these complex rules are worth it. */
1183 left_xy[1] = left_xy[0] = mb_xy-1;
1185 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
1186 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1188 if (left_mb_field_flag != curr_mb_field_flag) {
1189 left_xy[0] -= s->mb_stride;
1192 if(curr_mb_field_flag){
1193 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
1195 if (left_mb_field_flag != curr_mb_field_flag) {
1196 left_xy[1] += s->mb_stride;
1201 h->top_mb_xy = top_xy;
1202 h->left_mb_xy[0] = left_xy[0];
1203 h->left_mb_xy[1] = left_xy[1];
1205 //for sufficiently low qp, filtering wouldn't do anything
1206 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
1207 int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
1208 int qp = s->current_picture.qscale_table[mb_xy];
1210 && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
1211 && (top_xy < 0 || ((qp + s->current_picture.qscale_table[top_xy ] + 1)>>1) <= qp_thresh)){
1214 if( (left_xy[0]< 0 || ((qp + s->current_picture.qscale_table[left_xy[1] ] + 1)>>1) <= qp_thresh)
1215 && (top_xy < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy -s->mb_stride] + 1)>>1) <= qp_thresh))
1220 top_type = s->current_picture.mb_type[top_xy] ;
1221 left_type[0] = s->current_picture.mb_type[left_xy[0]];
1222 left_type[1] = s->current_picture.mb_type[left_xy[1]];
1223 if(h->deblocking_filter == 2){
1224 if(h->slice_table[top_xy ] != h->slice_num) top_type= 0;
1225 if(h->slice_table[left_xy[0] ] != h->slice_num) left_type[0]= left_type[1]= 0;
1227 if(h->slice_table[top_xy ] == 0xFFFF) top_type= 0;
1228 if(h->slice_table[left_xy[0] ] == 0xFFFF) left_type[0]= left_type[1] =0;
1230 h->top_type = top_type ;
1231 h->left_type[0]= left_type[0];
1232 h->left_type[1]= left_type[1];
1234 if(IS_INTRA(mb_type))
1237 AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
1238 AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
1239 AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
1240 AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
1241 AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
1243 h->cbp= h->cbp_table[mb_xy];
1247 for(list=0; list<h->list_count; list++){
1250 int16_t (*mv_dst)[2];
1251 int16_t (*mv_src)[2];
1253 if(!USES_LIST(mb_type, list)){
1254 fill_rectangle( h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
1255 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1256 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1257 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1258 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1262 ref = &s->current_picture.ref_index[list][4*mb_xy];
1264 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1265 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1266 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1268 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1269 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1272 b_stride = h->b_stride;
1273 mv_dst = &h->mv_cache[list][scan8[0]];
1274 mv_src = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
1276 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
1291 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
1293 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
1297 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
1298 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
1299 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
1300 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
1303 // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
1304 if(!CABAC && h->pps.transform_8x8_mode){
1305 if(IS_8x8DCT(top_type)){
1306 h->non_zero_count_cache[4+8*0]=
1307 h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
1308 h->non_zero_count_cache[6+8*0]=
1309 h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
1311 if(IS_8x8DCT(left_type[0])){
1312 h->non_zero_count_cache[3+8*1]=
1313 h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
1315 if(IS_8x8DCT(left_type[1])){
1316 h->non_zero_count_cache[3+8*3]=
1317 h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
1320 if(IS_8x8DCT(mb_type)){
1321 h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]=
1322 h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= h->cbp & 1;
1324 h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
1325 h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
1327 h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
1328 h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
1330 h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
1331 h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
1335 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
1337 for(list=0; list<h->list_count; list++){
1338 if(USES_LIST(top_type, list)){
1339 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1340 const int b8_xy= 4*top_xy + 2;
1341 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1342 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1343 h->ref_cache[list][scan8[0] + 0 - 1*8]=
1344 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
1345 h->ref_cache[list][scan8[0] + 2 - 1*8]=
1346 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
1348 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1349 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1352 if(!IS_INTERLACED(mb_type^left_type[0])){
1353 if(USES_LIST(left_type[0], list)){
1354 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1355 const int b8_xy= 4*left_xy[0] + 1;
1356 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1357 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
1358 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
1359 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
1360 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
1361 h->ref_cache[list][scan8[0] - 1 + 0 ]=
1362 h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
1363 h->ref_cache[list][scan8[0] - 1 +16 ]=
1364 h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
1366 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
1367 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
1368 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
1369 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
1370 h->ref_cache[list][scan8[0] - 1 + 0 ]=
1371 h->ref_cache[list][scan8[0] - 1 + 8 ]=
1372 h->ref_cache[list][scan8[0] - 1 + 16 ]=
1373 h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
1383 * gets the predicted intra4x4 prediction mode.
1385 static inline int pred_intra_mode(H264Context *h, int n){
1386 const int index8= scan8[n];
1387 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1388 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1389 const int min= FFMIN(left, top);
1391 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1393 if(min<0) return DC_PRED;
1397 static inline void write_back_non_zero_count(H264Context *h){
1398 const int mb_xy= h->mb_xy;
1400 AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1401 AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1402 AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[0+8*5]);
1403 AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8*3]);
1404 AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1407 static inline void write_back_motion(H264Context *h, int mb_type){
1408 MpegEncContext * const s = &h->s;
1409 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
1410 const int b8_xy= 4*h->mb_xy;
1413 if(!USES_LIST(mb_type, 0))
1414 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1416 for(list=0; list<h->list_count; list++){
1418 int16_t (*mv_dst)[2];
1419 int16_t (*mv_src)[2];
1421 if(!USES_LIST(mb_type, list))
1424 b_stride = h->b_stride;
1425 mv_dst = &s->current_picture.motion_val[list][b_xy];
1426 mv_src = &h->mv_cache[list][scan8[0]];
1428 AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1431 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
1432 uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1433 if(IS_SKIP(mb_type))
1434 AV_ZERO128(mvd_dst);
1436 AV_COPY64(mvd_dst, mvd_src + 8*3);
1437 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
1438 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
1439 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
1444 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1445 ref_index[0+0*2]= h->ref_cache[list][scan8[0]];
1446 ref_index[1+0*2]= h->ref_cache[list][scan8[4]];
1447 ref_index[0+1*2]= h->ref_cache[list][scan8[8]];
1448 ref_index[1+1*2]= h->ref_cache[list][scan8[12]];
1452 if(h->slice_type_nos == FF_B_TYPE && CABAC){
1453 if(IS_8X8(mb_type)){
1454 uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
1455 direct_table[1] = h->sub_mb_type[1]>>1;
1456 direct_table[2] = h->sub_mb_type[2]>>1;
1457 direct_table[3] = h->sub_mb_type[3]>>1;
1462 static inline int get_dct8x8_allowed(H264Context *h){
1463 if(h->sps.direct_8x8_inference_flag)
1464 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL));
1466 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1470 * decodes a P_SKIP or B_SKIP macroblock
1472 static void decode_mb_skip(H264Context *h){
1473 MpegEncContext * const s = &h->s;
1474 const int mb_xy= h->mb_xy;
1477 memset(h->non_zero_count[mb_xy], 0, 32);
1478 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1481 mb_type|= MB_TYPE_INTERLACED;
1483 if( h->slice_type_nos == FF_B_TYPE )
1485 // just for fill_caches. pred_direct_motion will set the real mb_type
1486 mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1487 if(h->direct_spatial_mv_pred){
1488 fill_decode_neighbors(h, mb_type);
1489 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1491 ff_h264_pred_direct_motion(h, &mb_type);
1492 mb_type|= MB_TYPE_SKIP;
1497 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1499 fill_decode_neighbors(h, mb_type);
1500 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1501 pred_pskip_motion(h, &mx, &my);
1502 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1503 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1506 write_back_motion(h, mb_type);
1507 s->current_picture.mb_type[mb_xy]= mb_type;
1508 s->current_picture.qscale_table[mb_xy]= s->qscale;
1509 h->slice_table[ mb_xy ]= h->slice_num;
1510 h->prev_mb_skipped= 1;
1513 #include "h264_mvpred.h" //For pred_pskip_motion()
1515 #endif /* AVCODEC_H264_H */