2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 * @file libavcodec/h264.h
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #ifndef AVCODEC_H264_H
29 #define AVCODEC_H264_H
31 #include "libavutil/intreadwrite.h"
34 #include "mpegvideo.h"
36 #include "rectangle.h"
38 #define interlaced_dct interlaced_dct_is_a_bad_name
39 #define mb_intra mb_intra_is_not_initialized_see_mb_type
41 #define LUMA_DC_BLOCK_INDEX 25
42 #define CHROMA_DC_BLOCK_INDEX 26
44 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
45 #define COEFF_TOKEN_VLC_BITS 8
46 #define TOTAL_ZEROS_VLC_BITS 9
47 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
48 #define RUN_VLC_BITS 3
49 #define RUN7_VLC_BITS 6
51 #define MAX_SPS_COUNT 32
52 #define MAX_PPS_COUNT 256
54 #define MAX_MMCO_COUNT 66
56 #define MAX_DELAYED_PIC_COUNT 16
58 /* Compiling in interlaced support reduces the speed
59 * of progressive decoding by about 2%. */
60 #define ALLOW_INTERLACE
62 #define ALLOW_NOCHROMA
67 * The maximum number of slices supported by the decoder.
68 * must be a power of 2
72 #ifdef ALLOW_INTERLACE
73 #define MB_MBAFF h->mb_mbaff
74 #define MB_FIELD h->mb_field_decoding_flag
75 #define FRAME_MBAFF h->mb_aff_frame
76 #define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
81 #define FIELD_PICTURE 0
83 #define IS_INTERLACED(mb_type) 0
85 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
88 #define CHROMA h->sps.chroma_format_idc
94 #define CABAC h->pps.cabac
97 #define EXTENDED_SAR 255
99 #define MB_TYPE_REF0 MB_TYPE_ACPRED //dirty but it fits in 16 bit
100 #define MB_TYPE_8x8DCT 0x01000000
101 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
102 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
105 * Value of Picture.reference when Picture is not a reference picture, but
106 * is held for delayed output.
108 #define DELAYED_PIC_REF 4
126 NAL_AUXILIARY_SLICE=19
133 SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
134 SEI_TYPE_PIC_TIMING = 1, ///< picture timing
135 SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
136 SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync)
140 * pic_struct in picture timing SEI message
143 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
144 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
145 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
146 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
147 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
148 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
149 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
150 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
151 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
155 * Sequence parameter set
161 int chroma_format_idc;
162 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
163 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
164 int poc_type; ///< pic_order_cnt_type
165 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
166 int delta_pic_order_always_zero_flag;
167 int offset_for_non_ref_pic;
168 int offset_for_top_to_bottom_field;
169 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
170 int ref_frame_count; ///< num_ref_frames
171 int gaps_in_frame_num_allowed_flag;
172 int mb_width; ///< pic_width_in_mbs_minus1 + 1
173 int mb_height; ///< pic_height_in_map_units_minus1 + 1
174 int frame_mbs_only_flag;
175 int mb_aff; ///<mb_adaptive_frame_field_flag
176 int direct_8x8_inference_flag;
177 int crop; ///< frame_cropping_flag
178 unsigned int crop_left; ///< frame_cropping_rect_left_offset
179 unsigned int crop_right; ///< frame_cropping_rect_right_offset
180 unsigned int crop_top; ///< frame_cropping_rect_top_offset
181 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
182 int vui_parameters_present_flag;
184 int video_signal_type_present_flag;
186 int colour_description_present_flag;
187 enum AVColorPrimaries color_primaries;
188 enum AVColorTransferCharacteristic color_trc;
189 enum AVColorSpace colorspace;
190 int timing_info_present_flag;
191 uint32_t num_units_in_tick;
193 int fixed_frame_rate_flag;
194 short offset_for_ref_frame[256]; //FIXME dyn aloc?
195 int bitstream_restriction_flag;
196 int num_reorder_frames;
197 int scaling_matrix_present;
198 uint8_t scaling_matrix4[6][16];
199 uint8_t scaling_matrix8[2][64];
200 int nal_hrd_parameters_present_flag;
201 int vcl_hrd_parameters_present_flag;
202 int pic_struct_present_flag;
203 int time_offset_length;
204 int cpb_cnt; ///< See H.264 E.1.2
205 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1
206 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
207 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
208 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
209 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
210 int residual_color_transform_flag; ///< residual_colour_transform_flag
214 * Picture parameter set
218 int cabac; ///< entropy_coding_mode_flag
219 int pic_order_present; ///< pic_order_present_flag
220 int slice_group_count; ///< num_slice_groups_minus1 + 1
221 int mb_slice_group_map_type;
222 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
223 int weighted_pred; ///< weighted_pred_flag
224 int weighted_bipred_idc;
225 int init_qp; ///< pic_init_qp_minus26 + 26
226 int init_qs; ///< pic_init_qs_minus26 + 26
227 int chroma_qp_index_offset[2];
228 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
229 int constrained_intra_pred; ///< constrained_intra_pred_flag
230 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
231 int transform_8x8_mode; ///< transform_8x8_mode_flag
232 uint8_t scaling_matrix4[6][16];
233 uint8_t scaling_matrix8[2][64];
234 uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
239 * Memory management control operation opcode.
241 typedef enum MMCOOpcode{
252 * Memory management control operation.
256 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
257 int long_arg; ///< index, pic_num, or num long refs depending on opcode
263 typedef struct H264Context{
267 uint8_t *rbsp_buffer[2];
268 unsigned int rbsp_buffer_size[2];
271 * Used to parse AVC variant of h264
273 int is_avc; ///< this flag is != 0 if codec is avc1
274 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
276 int chroma_qp[2]; //QPc
278 int qp_thresh; ///< QP threshold to skip loopfilter
284 int chroma_pred_mode;
285 int intra16x16_pred_mode;
297 const uint8_t * left_block;
298 int topleft_partition;
300 int8_t intra4x4_pred_mode_cache[5*8];
301 int8_t (*intra4x4_pred_mode);
303 unsigned int topleft_samples_available;
304 unsigned int top_samples_available;
305 unsigned int topright_samples_available;
306 unsigned int left_samples_available;
307 uint8_t (*top_borders[2])[16+2*8];
308 uint8_t left_border[2*(17+2*9)];
311 * non zero coeff count cache.
312 * is 64 if not available.
314 DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache)[6*8];
322 uint8_t (*non_zero_count)[32];
325 * Motion vector cache.
327 DECLARE_ALIGNED_16(int16_t, mv_cache)[2][5*8][2];
328 DECLARE_ALIGNED_8(int8_t, ref_cache)[2][5*8];
329 #define LIST_NOT_USED -1 //FIXME rename?
330 #define PART_NOT_AVAILABLE -2
333 * is 1 if the specific list MV&references are set to 0,0,-2.
335 int mv_cache_clean[2];
338 * number of neighbors (top and/or left) that used 8x8 dct
340 int neighbor_transform_size;
343 * block_offset[ 0..23] for frame macroblocks
344 * block_offset[24..47] for field macroblocks
346 int block_offset[2*(16+8)];
348 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
350 int b_stride; //FIXME use s->b4_stride
352 int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff
358 SPS *sps_buffers[MAX_SPS_COUNT];
359 SPS sps; ///< current sps
361 PPS *pps_buffers[MAX_PPS_COUNT];
365 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
367 uint32_t dequant4_buffer[6][52][16];
368 uint32_t dequant8_buffer[2][52][64];
369 uint32_t (*dequant4_coeff[6])[16];
370 uint32_t (*dequant8_coeff[2])[64];
371 int dequant_coeff_pps; ///< reinit tables when pps changes
374 uint16_t *slice_table_base;
375 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
377 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
378 int slice_type_fixed;
380 //interlacing specific flags
382 int mb_field_decoding_flag;
383 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
385 DECLARE_ALIGNED_8(uint16_t, sub_mb_type)[4];
390 int delta_poc_bottom;
393 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
394 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
395 int frame_num_offset; ///< for POC type 2
396 int prev_frame_num_offset; ///< for POC type 2
397 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
400 * frame_num for frames or 2*frame_num+1 for field pics.
405 * max_frame_num or 2*max_frame_num for field pics.
409 //Weighted pred stuff
411 int use_weight_chroma;
412 int luma_log2_weight_denom;
413 int chroma_log2_weight_denom;
414 //The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
415 int luma_weight[2][48][2];
416 int chroma_weight[2][48][2][2];
417 int implicit_weight[48][48];
420 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
421 int slice_alpha_c0_offset;
422 int slice_beta_offset;
424 int redundant_pic_count;
426 int direct_spatial_mv_pred;
429 int dist_scale_factor[16];
430 int dist_scale_factor_field[2][32];
431 int map_col_to_list0[2][16+32];
432 int map_col_to_list0_field[2][2][16+32];
435 * num_ref_idx_l0/1_active_minus1 + 1
437 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
438 unsigned int list_count;
439 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
440 Picture *short_ref[32];
441 Picture *long_ref[32];
442 Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
443 Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
444 Reordered version of default_ref_list
445 according to picture reordering in slice header */
446 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
447 Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
451 * memory management control operations buffer.
453 MMCO mmco[MAX_MMCO_COUNT];
456 int long_ref_count; ///< number of actual long term references
457 int short_ref_count; ///< number of actual short term references
460 GetBitContext intra_gb;
461 GetBitContext inter_gb;
462 GetBitContext *intra_gb_ptr;
463 GetBitContext *inter_gb_ptr;
465 DECLARE_ALIGNED_16(DCTELEM, mb)[16*24];
466 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
472 uint8_t cabac_state[460];
475 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
480 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
481 uint8_t *chroma_pred_mode_table;
482 int last_qscale_diff;
483 uint8_t (*mvd_table[2])[2];
484 DECLARE_ALIGNED_16(uint8_t, mvd_cache)[2][5*8][2];
485 uint8_t *direct_table;
486 uint8_t direct_cache[5*8];
488 uint8_t zigzag_scan[16];
489 uint8_t zigzag_scan8x8[64];
490 uint8_t zigzag_scan8x8_cavlc[64];
491 uint8_t field_scan[16];
492 uint8_t field_scan8x8[64];
493 uint8_t field_scan8x8_cavlc[64];
494 const uint8_t *zigzag_scan_q0;
495 const uint8_t *zigzag_scan8x8_q0;
496 const uint8_t *zigzag_scan8x8_cavlc_q0;
497 const uint8_t *field_scan_q0;
498 const uint8_t *field_scan8x8_q0;
499 const uint8_t *field_scan8x8_cavlc_q0;
504 * @defgroup multithreading Members for slice based multithreading
507 struct H264Context *thread_context[MAX_THREADS];
510 * current slice number, used to initalize slice_num of each thread/context
515 * Max number of threads / contexts.
516 * This is equal to AVCodecContext.thread_count unless
517 * multithreaded decoding is impossible, in which case it is
523 * 1 if the single thread fallback warning has already been
524 * displayed, 0 otherwise.
526 int single_decode_warning;
534 * pic_struct in picture timing SEI message
536 SEI_PicStructType sei_pic_struct;
539 * Complement sei_pic_struct
540 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
541 * However, soft telecined frames may have these values.
542 * This is used in an attempt to flag soft telecine progressive.
544 int prev_interlaced_frame;
547 * Bit set of clock types for fields/frames in picture timing SEI message.
548 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
554 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
556 int sei_dpb_output_delay;
559 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
561 int sei_cpb_removal_delay;
564 * recovery_frame_cnt from SEI message
566 * Set to -1 if no recovery point SEI message found or to number of frames
567 * before playback synchronizes. Frames having recovery point are key
570 int sei_recovery_frame_cnt;
574 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
575 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
578 int sei_buffering_period_present; ///< Buffering period SEI flag
579 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
581 //SVQ3 specific fields
584 int unknown_svq3_flag;
585 int next_slice_index;
586 uint32_t svq3_watermark_key;
590 extern const uint8_t ff_h264_chroma_qp[52];
592 void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
594 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
599 int ff_h264_decode_sei(H264Context *h);
604 int ff_h264_decode_seq_parameter_set(H264Context *h);
609 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
612 * Decodes a network abstraction layer unit.
613 * @param consumed is the number of bytes used as input
614 * @param length is the length of the array
615 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
616 * @returns decoded bytes, might be src+1 if no escapes
618 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
621 * identifies the exact end of the bitstream
622 * @return the length of the trailing, or 0 if damaged
624 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
627 * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
629 av_cold void ff_h264_free_context(H264Context *h);
632 * reconstructs bitstream slice_type.
634 int ff_h264_get_slice_type(const H264Context *h);
640 int ff_h264_alloc_tables(H264Context *h);
643 * fills the default_ref_list.
645 int ff_h264_fill_default_ref_list(H264Context *h);
647 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
648 void ff_h264_fill_mbaff_ref_list(H264Context *h);
649 void ff_h264_remove_all_refs(H264Context *h);
652 * Executes the reference picture marking (memory management control operations).
654 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
656 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
660 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
662 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
665 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
667 int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
669 void ff_h264_write_back_intra_pred_mode(H264Context *h);
670 void ff_h264_hl_decode_mb(H264Context *h);
671 int ff_h264_frame_start(H264Context *h);
672 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
673 av_cold int ff_h264_decode_end(AVCodecContext *avctx);
674 av_cold void ff_h264_decode_init_vlc(void);
677 * decodes a macroblock
678 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
680 int ff_h264_decode_mb_cavlc(H264Context *h);
683 * decodes a CABAC coded macroblock
684 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
686 int ff_h264_decode_mb_cabac(H264Context *h);
688 void ff_h264_init_cabac_states(H264Context *h);
690 void ff_h264_direct_dist_scale_factor(H264Context * const h);
691 void ff_h264_direct_ref_list_init(H264Context * const h);
692 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
694 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);
695 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);
698 * Reset SEI values at the beginning of the frame.
700 * @param h H.264 context.
702 void ff_h264_reset_sei(H264Context *h);
714 //This table must be here because scan8[constant] must be known at compiletime
715 static const uint8_t scan8[16 + 2*4]={
716 4+1*8, 5+1*8, 4+2*8, 5+2*8,
717 6+1*8, 7+1*8, 6+2*8, 7+2*8,
718 4+3*8, 5+3*8, 4+4*8, 5+4*8,
719 6+3*8, 7+3*8, 6+4*8, 7+4*8,
726 static av_always_inline uint32_t pack16to32(int a, int b){
728 return (b&0xFFFF) + (a<<16);
730 return (a&0xFFFF) + (b<<16);
734 static av_always_inline uint16_t pack8to16(int a, int b){
736 return (b&0xFF) + (a<<8);
738 return (a&0xFF) + (b<<8);
743 * gets the chroma qp.
745 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
746 return h->pps.chroma_qp_table[t][qscale];
749 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
751 static void fill_decode_neighbors(H264Context *h, int mb_type){
752 MpegEncContext * const s = &h->s;
753 const int mb_xy= h->mb_xy;
754 int topleft_xy, top_xy, topright_xy, left_xy[2];
755 static const uint8_t left_block_options[4][16]={
756 {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},
757 {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},
758 {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},
759 {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}
762 h->topleft_partition= -1;
764 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
766 /* Wow, what a mess, why didn't they simplify the interlacing & intra
767 * stuff, I can't imagine that these complex rules are worth it. */
769 topleft_xy = top_xy - 1;
770 topright_xy= top_xy + 1;
771 left_xy[1] = left_xy[0] = mb_xy-1;
772 h->left_block = left_block_options[0];
774 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
775 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
777 if (left_mb_field_flag != curr_mb_field_flag) {
778 left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
779 if (curr_mb_field_flag) {
780 left_xy[1] += s->mb_stride;
781 h->left_block = left_block_options[3];
783 topleft_xy += s->mb_stride;
784 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
785 h->topleft_partition = 0;
786 h->left_block = left_block_options[1];
790 if(curr_mb_field_flag){
791 topleft_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
792 topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
793 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
795 if (left_mb_field_flag != curr_mb_field_flag) {
796 if (curr_mb_field_flag) {
797 left_xy[1] += s->mb_stride;
798 h->left_block = left_block_options[3];
800 h->left_block = left_block_options[2];
806 h->topleft_mb_xy = topleft_xy;
807 h->top_mb_xy = top_xy;
808 h->topright_mb_xy= topright_xy;
809 h->left_mb_xy[0] = left_xy[0];
810 h->left_mb_xy[1] = left_xy[1];
811 //FIXME do we need all in the context?
813 h->topleft_type = s->current_picture.mb_type[topleft_xy] ;
814 h->top_type = s->current_picture.mb_type[top_xy] ;
815 h->topright_type= s->current_picture.mb_type[topright_xy];
816 h->left_type[0] = s->current_picture.mb_type[left_xy[0]] ;
817 h->left_type[1] = s->current_picture.mb_type[left_xy[1]] ;
819 if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0;
820 if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0;
821 if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
822 if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
825 static void fill_decode_caches(H264Context *h, int mb_type){
826 MpegEncContext * const s = &h->s;
827 int topleft_xy, top_xy, topright_xy, left_xy[2];
828 int topleft_type, top_type, topright_type, left_type[2];
829 const uint8_t * left_block= h->left_block;
832 topleft_xy = h->topleft_mb_xy ;
833 top_xy = h->top_mb_xy ;
834 topright_xy = h->topright_mb_xy;
835 left_xy[0] = h->left_mb_xy[0] ;
836 left_xy[1] = h->left_mb_xy[1] ;
837 topleft_type = h->topleft_type ;
838 top_type = h->top_type ;
839 topright_type= h->topright_type ;
840 left_type[0] = h->left_type[0] ;
841 left_type[1] = h->left_type[1] ;
843 if(!IS_SKIP(mb_type)){
844 if(IS_INTRA(mb_type)){
845 int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
846 h->topleft_samples_available=
847 h->top_samples_available=
848 h->left_samples_available= 0xFFFF;
849 h->topright_samples_available= 0xEEEA;
851 if(!(top_type & type_mask)){
852 h->topleft_samples_available= 0xB3FF;
853 h->top_samples_available= 0x33FF;
854 h->topright_samples_available= 0x26EA;
856 if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
857 if(IS_INTERLACED(mb_type)){
858 if(!(left_type[0] & type_mask)){
859 h->topleft_samples_available&= 0xDFFF;
860 h->left_samples_available&= 0x5FFF;
862 if(!(left_type[1] & type_mask)){
863 h->topleft_samples_available&= 0xFF5F;
864 h->left_samples_available&= 0xFF5F;
867 int left_typei = s->current_picture.mb_type[left_xy[0] + s->mb_stride];
869 assert(left_xy[0] == left_xy[1]);
870 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
871 h->topleft_samples_available&= 0xDF5F;
872 h->left_samples_available&= 0x5F5F;
876 if(!(left_type[0] & type_mask)){
877 h->topleft_samples_available&= 0xDF5F;
878 h->left_samples_available&= 0x5F5F;
882 if(!(topleft_type & type_mask))
883 h->topleft_samples_available&= 0x7FFF;
885 if(!(topright_type & type_mask))
886 h->topright_samples_available&= 0xFBFF;
888 if(IS_INTRA4x4(mb_type)){
889 if(IS_INTRA4x4(top_type)){
890 AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]);
892 h->intra4x4_pred_mode_cache[4+8*0]=
893 h->intra4x4_pred_mode_cache[5+8*0]=
894 h->intra4x4_pred_mode_cache[6+8*0]=
895 h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask);
898 if(IS_INTRA4x4(left_type[i])){
899 int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]];
900 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]];
901 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]];
903 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
904 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[i] & type_mask);
919 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
921 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
922 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
923 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
925 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
926 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
928 h->non_zero_count_cache[1+8*0]=
929 h->non_zero_count_cache[2+8*0]=
931 h->non_zero_count_cache[1+8*3]=
932 h->non_zero_count_cache[2+8*3]=
933 AV_WN32A(&h->non_zero_count_cache[4+8*0], CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040);
936 for (i=0; i<2; i++) {
938 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
939 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
940 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
941 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
943 h->non_zero_count_cache[3+8*1 + 2*8*i]=
944 h->non_zero_count_cache[3+8*2 + 2*8*i]=
945 h->non_zero_count_cache[0+8*1 + 8*i]=
946 h->non_zero_count_cache[0+8*4 + 8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
953 h->top_cbp = h->cbp_table[top_xy];
955 h->top_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
959 h->left_cbp = (h->cbp_table[left_xy[0]] & 0x1f0)
960 | ((h->cbp_table[left_xy[0]]>>(left_block[0]&(~1)))&2)
961 | (((h->cbp_table[left_xy[1]]>>(left_block[2]&(~1)))&2) << 2);
963 h->left_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
969 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
971 for(list=0; list<h->list_count; list++){
972 if(!USES_LIST(mb_type, list)){
973 /*if(!h->mv_cache_clean[list]){
974 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
975 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
976 h->mv_cache_clean[list]= 1;
980 assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
982 h->mv_cache_clean[list]= 0;
984 if(USES_LIST(top_type, list)){
985 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
986 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
987 h->ref_cache[list][scan8[0] + 0 - 1*8]=
988 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 2];
989 h->ref_cache[list][scan8[0] + 2 - 1*8]=
990 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 3];
992 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
993 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
996 if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){
998 int cache_idx = scan8[0] - 1 + i*2*8;
999 if(USES_LIST(left_type[i], list)){
1000 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1001 const int b8_xy= 4*left_xy[i] + 1;
1002 AV_COPY32(h->mv_cache[list][cache_idx ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]);
1003 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]]);
1004 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + (left_block[0+i*2]&~1)];
1005 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + (left_block[1+i*2]&~1)];
1007 AV_ZERO32(h->mv_cache [list][cache_idx ]);
1008 AV_ZERO32(h->mv_cache [list][cache_idx+8]);
1009 h->ref_cache[list][cache_idx ]=
1010 h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1014 if(USES_LIST(left_type[0], list)){
1015 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1016 const int b8_xy= 4*left_xy[0] + 1;
1017 AV_COPY32(h->mv_cache[list][scan8[0] - 1], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]]);
1018 h->ref_cache[list][scan8[0] - 1]= s->current_picture.ref_index[list][b8_xy + (left_block[0]&~1)];
1020 AV_ZERO32(h->mv_cache [list][scan8[0] - 1]);
1021 h->ref_cache[list][scan8[0] - 1]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1025 if(USES_LIST(topright_type, list)){
1026 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1027 AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]);
1028 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][4*topright_xy + 2];
1030 AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]);
1031 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1033 if(h->ref_cache[list][scan8[0] + 4 - 1*8] < 0){
1034 if(USES_LIST(topleft_type, list)){
1035 const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
1036 const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2);
1037 AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]);
1038 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1040 AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]);
1041 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1045 if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
1048 if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
1049 h->ref_cache[list][scan8[4 ]] =
1050 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1051 AV_ZERO32(h->mv_cache [list][scan8[4 ]]);
1052 AV_ZERO32(h->mv_cache [list][scan8[12]]);
1055 /* XXX beurk, Load mvd */
1056 if(USES_LIST(top_type, list)){
1057 const int b_xy= h->mb2br_xy[top_xy];
1058 AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1060 AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1062 if(USES_LIST(left_type[0], list)){
1063 const int b_xy= h->mb2br_xy[left_xy[0]] + 6;
1064 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy - left_block[0]]);
1065 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy - left_block[1]]);
1067 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]);
1068 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]);
1070 if(USES_LIST(left_type[1], list)){
1071 const int b_xy= h->mb2br_xy[left_xy[1]] + 6;
1072 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy - left_block[2]]);
1073 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy - left_block[3]]);
1075 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]);
1076 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]);
1078 AV_ZERO16(h->mvd_cache [list][scan8[4 ]]);
1079 AV_ZERO16(h->mvd_cache [list][scan8[12]]);
1080 if(h->slice_type_nos == FF_B_TYPE){
1081 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
1083 if(IS_DIRECT(top_type)){
1084 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_DIRECT2>>1));
1085 }else if(IS_8X8(top_type)){
1086 int b8_xy = 4*top_xy;
1087 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy + 2];
1088 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 3];
1090 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1));
1093 if(IS_DIRECT(left_type[0]))
1094 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
1095 else if(IS_8X8(left_type[0]))
1096 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[4*left_xy[0] + 1 + (left_block[0]&~1)];
1098 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
1100 if(IS_DIRECT(left_type[1]))
1101 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
1102 else if(IS_8X8(left_type[1]))
1103 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[4*left_xy[1] + 1 + (left_block[2]&~1)];
1105 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
1111 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1112 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1113 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1114 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1115 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1116 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1117 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1118 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1119 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1120 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1122 #define MAP_F2F(idx, mb_type)\
1123 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1124 h->ref_cache[list][idx] <<= 1;\
1125 h->mv_cache[list][idx][1] /= 2;\
1126 h->mvd_cache[list][idx][1] >>=1;\
1131 #define MAP_F2F(idx, mb_type)\
1132 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1133 h->ref_cache[list][idx] >>= 1;\
1134 h->mv_cache[list][idx][1] <<= 1;\
1135 h->mvd_cache[list][idx][1] <<= 1;\
1145 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1150 * @returns non zero if the loop filter can be skiped
1152 static int fill_filter_caches(H264Context *h, int mb_type){
1153 MpegEncContext * const s = &h->s;
1154 const int mb_xy= h->mb_xy;
1155 int top_xy, left_xy[2];
1156 int top_type, left_type[2];
1158 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
1160 //FIXME deblocking could skip the intra and nnz parts.
1162 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1163 * stuff, I can't imagine that these complex rules are worth it. */
1165 left_xy[1] = left_xy[0] = mb_xy-1;
1167 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
1168 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1170 if (left_mb_field_flag != curr_mb_field_flag) {
1171 left_xy[0] -= s->mb_stride;
1174 if(curr_mb_field_flag){
1175 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
1177 if (left_mb_field_flag != curr_mb_field_flag) {
1178 left_xy[1] += s->mb_stride;
1183 h->top_mb_xy = top_xy;
1184 h->left_mb_xy[0] = left_xy[0];
1185 h->left_mb_xy[1] = left_xy[1];
1187 //for sufficiently low qp, filtering wouldn't do anything
1188 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
1189 int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
1190 int qp = s->current_picture.qscale_table[mb_xy];
1192 && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
1193 && (top_xy < 0 || ((qp + s->current_picture.qscale_table[top_xy ] + 1)>>1) <= qp_thresh)){
1196 if( (left_xy[0]< 0 || ((qp + s->current_picture.qscale_table[left_xy[1] ] + 1)>>1) <= qp_thresh)
1197 && (top_xy < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy -s->mb_stride] + 1)>>1) <= qp_thresh))
1202 if(h->deblocking_filter == 2){
1203 h->top_type = top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
1204 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;
1205 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;
1207 h->top_type = top_type = h->slice_table[top_xy ] < 0xFFFF ? s->current_picture.mb_type[top_xy] : 0;
1208 h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
1209 h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
1211 if(IS_INTRA(mb_type))
1214 AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
1215 AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
1216 AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
1217 AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
1218 AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
1220 h->cbp= h->cbp_table[mb_xy];
1224 for(list=0; list<h->list_count; list++){
1227 int16_t (*mv_dst)[2];
1228 int16_t (*mv_src)[2];
1230 if(!USES_LIST(mb_type, list)){
1231 fill_rectangle( h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
1232 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1233 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1234 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1235 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1239 ref = &s->current_picture.ref_index[list][4*mb_xy];
1241 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1242 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1243 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1245 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1246 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1249 b_stride = h->b_stride;
1250 mv_dst = &h->mv_cache[list][scan8[0]];
1251 mv_src = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
1253 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
1268 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
1270 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
1274 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
1275 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
1276 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
1277 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
1280 // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
1281 if(!CABAC && h->pps.transform_8x8_mode){
1282 if(IS_8x8DCT(top_type)){
1283 h->non_zero_count_cache[4+8*0]=
1284 h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
1285 h->non_zero_count_cache[6+8*0]=
1286 h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
1288 if(IS_8x8DCT(left_type[0])){
1289 h->non_zero_count_cache[3+8*1]=
1290 h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
1292 if(IS_8x8DCT(left_type[1])){
1293 h->non_zero_count_cache[3+8*3]=
1294 h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
1297 if(IS_8x8DCT(mb_type)){
1298 h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]=
1299 h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= h->cbp & 1;
1301 h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
1302 h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
1304 h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
1305 h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
1307 h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
1308 h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
1312 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
1314 for(list=0; list<h->list_count; list++){
1315 if(USES_LIST(top_type, list)){
1316 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1317 const int b8_xy= 4*top_xy + 2;
1318 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1319 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1320 h->ref_cache[list][scan8[0] + 0 - 1*8]=
1321 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
1322 h->ref_cache[list][scan8[0] + 2 - 1*8]=
1323 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
1325 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1326 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1329 if(!IS_INTERLACED(mb_type^left_type[0])){
1330 if(USES_LIST(left_type[0], list)){
1331 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1332 const int b8_xy= 4*left_xy[0] + 1;
1333 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1334 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
1335 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
1336 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
1337 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
1338 h->ref_cache[list][scan8[0] - 1 + 0 ]=
1339 h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
1340 h->ref_cache[list][scan8[0] - 1 +16 ]=
1341 h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
1343 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
1344 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
1345 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
1346 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
1347 h->ref_cache[list][scan8[0] - 1 + 0 ]=
1348 h->ref_cache[list][scan8[0] - 1 + 8 ]=
1349 h->ref_cache[list][scan8[0] - 1 + 16 ]=
1350 h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
1360 * gets the predicted intra4x4 prediction mode.
1362 static inline int pred_intra_mode(H264Context *h, int n){
1363 const int index8= scan8[n];
1364 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1365 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1366 const int min= FFMIN(left, top);
1368 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1370 if(min<0) return DC_PRED;
1374 static inline void write_back_non_zero_count(H264Context *h){
1375 const int mb_xy= h->mb_xy;
1377 AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1378 AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1379 AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[0+8*5]);
1380 AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8*3]);
1381 AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1384 static inline void write_back_motion(H264Context *h, int mb_type){
1385 MpegEncContext * const s = &h->s;
1386 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
1387 const int b8_xy= 4*h->mb_xy;
1390 if(!USES_LIST(mb_type, 0))
1391 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1393 for(list=0; list<h->list_count; list++){
1395 int16_t (*mv_dst)[2];
1396 int16_t (*mv_src)[2];
1398 if(!USES_LIST(mb_type, list))
1401 b_stride = h->b_stride;
1402 mv_dst = &s->current_picture.motion_val[list][b_xy];
1403 mv_src = &h->mv_cache[list][scan8[0]];
1405 AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1408 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
1409 uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1410 if(IS_SKIP(mb_type))
1411 AV_ZERO128(mvd_dst);
1413 AV_COPY64(mvd_dst, mvd_src + 8*3);
1414 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
1415 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
1416 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
1421 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1422 ref_index[0+0*2]= h->ref_cache[list][scan8[0]];
1423 ref_index[1+0*2]= h->ref_cache[list][scan8[4]];
1424 ref_index[0+1*2]= h->ref_cache[list][scan8[8]];
1425 ref_index[1+1*2]= h->ref_cache[list][scan8[12]];
1429 if(h->slice_type_nos == FF_B_TYPE && CABAC){
1430 if(IS_8X8(mb_type)){
1431 uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
1432 direct_table[1] = h->sub_mb_type[1]>>1;
1433 direct_table[2] = h->sub_mb_type[2]>>1;
1434 direct_table[3] = h->sub_mb_type[3]>>1;
1439 static inline int get_dct8x8_allowed(H264Context *h){
1440 if(h->sps.direct_8x8_inference_flag)
1441 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL));
1443 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1447 * decodes a P_SKIP or B_SKIP macroblock
1449 static void decode_mb_skip(H264Context *h){
1450 MpegEncContext * const s = &h->s;
1451 const int mb_xy= h->mb_xy;
1454 memset(h->non_zero_count[mb_xy], 0, 32);
1455 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1458 mb_type|= MB_TYPE_INTERLACED;
1460 if( h->slice_type_nos == FF_B_TYPE )
1462 // just for fill_caches. pred_direct_motion will set the real mb_type
1463 mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1464 if(h->direct_spatial_mv_pred){
1465 fill_decode_neighbors(h, mb_type);
1466 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1468 ff_h264_pred_direct_motion(h, &mb_type);
1469 mb_type|= MB_TYPE_SKIP;
1474 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1476 fill_decode_neighbors(h, mb_type);
1477 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1478 pred_pskip_motion(h, &mx, &my);
1479 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1480 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1483 write_back_motion(h, mb_type);
1484 s->current_picture.mb_type[mb_xy]= mb_type;
1485 s->current_picture.qscale_table[mb_xy]= s->qscale;
1486 h->slice_table[ mb_xy ]= h->slice_num;
1487 h->prev_mb_skipped= 1;
1490 #include "h264_mvpred.h" //For pred_pskip_motion()
1492 #endif /* AVCODEC_H264_H */