Factorize ff_generate_sliding_window_mmcos() out.
[libav.git] / libavcodec / h264.h
1 /*
2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * This file is part of FFmpeg.
6 *
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.
11 *
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.
16 *
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
20 */
21
22 /**
23 * @file
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
28 #ifndef AVCODEC_H264_H
29 #define AVCODEC_H264_H
30
31 #include "libavutil/intreadwrite.h"
32 #include "dsputil.h"
33 #include "cabac.h"
34 #include "mpegvideo.h"
35 #include "h264dsp.h"
36 #include "h264pred.h"
37 #include "rectangle.h"
38
39 #define interlaced_dct interlaced_dct_is_a_bad_name
40 #define mb_intra mb_intra_is_not_initialized_see_mb_type
41
42 #define LUMA_DC_BLOCK_INDEX 25
43 #define CHROMA_DC_BLOCK_INDEX 26
44
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
51
52 #define MAX_SPS_COUNT 32
53 #define MAX_PPS_COUNT 256
54
55 #define MAX_MMCO_COUNT 66
56
57 #define MAX_DELAYED_PIC_COUNT 16
58
59 /* Compiling in interlaced support reduces the speed
60 * of progressive decoding by about 2%. */
61 #define ALLOW_INTERLACE
62
63 #define ALLOW_NOCHROMA
64
65 #define FMO 0
66
67 /**
68 * The maximum number of slices supported by the decoder.
69 * must be a power of 2
70 */
71 #define MAX_SLICES 16
72
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)
78 #else
79 #define MB_MBAFF 0
80 #define MB_FIELD 0
81 #define FRAME_MBAFF 0
82 #define FIELD_PICTURE 0
83 #undef IS_INTERLACED
84 #define IS_INTERLACED(mb_type) 0
85 #endif
86 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
87
88 #ifdef ALLOW_NOCHROMA
89 #define CHROMA h->sps.chroma_format_idc
90 #else
91 #define CHROMA 1
92 #endif
93
94 #ifndef CABAC
95 #define CABAC h->pps.cabac
96 #endif
97
98 #define EXTENDED_SAR 255
99
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)
104
105 /**
106 * Value of Picture.reference when Picture is not a reference picture, but
107 * is held for delayed output.
108 */
109 #define DELAYED_PIC_REF 4
110
111
112 /* NAL unit types */
113 enum {
114 NAL_SLICE=1,
115 NAL_DPA,
116 NAL_DPB,
117 NAL_DPC,
118 NAL_IDR_SLICE,
119 NAL_SEI,
120 NAL_SPS,
121 NAL_PPS,
122 NAL_AUD,
123 NAL_END_SEQUENCE,
124 NAL_END_STREAM,
125 NAL_FILLER_DATA,
126 NAL_SPS_EXT,
127 NAL_AUXILIARY_SLICE=19
128 };
129
130 /**
131 * SEI message types
132 */
133 typedef enum {
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)
138 } SEI_Type;
139
140 /**
141 * pic_struct in picture timing SEI message
142 */
143 typedef enum {
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
153 } SEI_PicStructType;
154
155 /**
156 * Sequence parameter set
157 */
158 typedef struct SPS{
159
160 int profile_idc;
161 int level_idc;
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;
184 AVRational sar;
185 int video_signal_type_present_flag;
186 int full_range;
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;
193 uint32_t time_scale;
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
212 }SPS;
213
214 /**
215 * Picture parameter set
216 */
217 typedef struct PPS{
218 unsigned int sps_id;
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
236 int chroma_qp_diff;
237 }PPS;
238
239 /**
240 * Memory management control operation opcode.
241 */
242 typedef enum MMCOOpcode{
243 MMCO_END=0,
244 MMCO_SHORT2UNUSED,
245 MMCO_LONG2UNUSED,
246 MMCO_SHORT2LONG,
247 MMCO_SET_MAX_LONG,
248 MMCO_RESET,
249 MMCO_LONG,
250 } MMCOOpcode;
251
252 /**
253 * Memory management control operation.
254 */
255 typedef struct MMCO{
256 MMCOOpcode opcode;
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
259 } MMCO;
260
261 /**
262 * H264Context
263 */
264 typedef struct H264Context{
265 MpegEncContext s;
266 H264DSPContext h264dsp;
267 int chroma_qp[2]; //QPc
268
269 int qp_thresh; ///< QP threshold to skip loopfilter
270
271 int prev_mb_skipped;
272 int next_mb_skipped;
273
274 //prediction stuff
275 int chroma_pred_mode;
276 int intra16x16_pred_mode;
277
278 int topleft_mb_xy;
279 int top_mb_xy;
280 int topright_mb_xy;
281 int left_mb_xy[2];
282
283 int topleft_type;
284 int top_type;
285 int topright_type;
286 int left_type[2];
287
288 const uint8_t * left_block;
289 int topleft_partition;
290
291 int8_t intra4x4_pred_mode_cache[5*8];
292 int8_t (*intra4x4_pred_mode);
293 H264PredContext hpc;
294 unsigned int topleft_samples_available;
295 unsigned int top_samples_available;
296 unsigned int topright_samples_available;
297 unsigned int left_samples_available;
298 uint8_t (*top_borders[2])[16+2*8];
299
300 /**
301 * non zero coeff count cache.
302 * is 64 if not available.
303 */
304 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[6*8];
305
306 /*
307 .UU.YYYY
308 .UU.YYYY
309 .vv.YYYY
310 .VV.YYYY
311 */
312 uint8_t (*non_zero_count)[32];
313
314 /**
315 * Motion vector cache.
316 */
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
321
322 /**
323 * is 1 if the specific list MV&references are set to 0,0,-2.
324 */
325 int mv_cache_clean[2];
326
327 /**
328 * number of neighbors (top and/or left) that used 8x8 dct
329 */
330 int neighbor_transform_size;
331
332 /**
333 * block_offset[ 0..23] for frame macroblocks
334 * block_offset[24..47] for field macroblocks
335 */
336 int block_offset[2*(16+8)];
337
338 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
339 uint32_t *mb2br_xy;
340 int b_stride; //FIXME use s->b4_stride
341
342 int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff
343 int mb_uvlinesize;
344
345 int emu_edge_width;
346 int emu_edge_height;
347
348 SPS sps; ///< current sps
349
350 /**
351 * current pps
352 */
353 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
354
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];
359
360 int slice_num;
361 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
362 int slice_type;
363 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
364 int slice_type_fixed;
365
366 //interlacing specific flags
367 int mb_aff_frame;
368 int mb_field_decoding_flag;
369 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
370
371 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
372
373 //Weighted pred stuff
374 int use_weight;
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];
382
383 int direct_spatial_mv_pred;
384 int col_parity;
385 int col_fieldoff;
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];
390
391 /**
392 * num_ref_idx_l0/1_active_minus1 + 1
393 */
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
401
402 //data partitioning
403 GetBitContext intra_gb;
404 GetBitContext inter_gb;
405 GetBitContext *intra_gb_ptr;
406 GetBitContext *inter_gb_ptr;
407
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
410
411 /**
412 * Cabac
413 */
414 CABACContext cabac;
415 uint8_t cabac_state[460];
416
417 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
418 uint16_t *cbp_table;
419 int cbp;
420 int top_cbp;
421 int left_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];
429
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;
442
443 int x264_build;
444
445 int mb_xy;
446
447 int is_complex;
448
449 //deblock
450 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
451 int slice_alpha_c0_offset;
452 int slice_beta_offset;
453
454 //=============================================================
455 //Things below are not used in the MB or more inner code
456
457 int nal_ref_idc;
458 int nal_unit_type;
459 uint8_t *rbsp_buffer[2];
460 unsigned int rbsp_buffer_size[2];
461
462 /**
463 * Used to parse AVC variant of h264
464 */
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)
467 int got_first; ///< this flag is != 0 if we've parsed a frame
468
469 SPS *sps_buffers[MAX_SPS_COUNT];
470 PPS *pps_buffers[MAX_PPS_COUNT];
471
472 int dequant_coeff_pps; ///< reinit tables when pps changes
473
474 uint16_t *slice_table_base;
475
476
477 //POC stuff
478 int poc_lsb;
479 int poc_msb;
480 int delta_poc_bottom;
481 int delta_poc[2];
482 int frame_num;
483 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
484 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
485 int frame_num_offset; ///< for POC type 2
486 int prev_frame_num_offset; ///< for POC type 2
487 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
488
489 /**
490 * frame_num for frames or 2*frame_num+1 for field pics.
491 */
492 int curr_pic_num;
493
494 /**
495 * max_frame_num or 2*max_frame_num for field pics.
496 */
497 int max_pic_num;
498
499 int redundant_pic_count;
500
501 Picture *short_ref[32];
502 Picture *long_ref[32];
503 Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
504 Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
505 int outputed_poc;
506
507 /**
508 * memory management control operations buffer.
509 */
510 MMCO mmco[MAX_MMCO_COUNT];
511 int mmco_index;
512
513 int long_ref_count; ///< number of actual long term references
514 int short_ref_count; ///< number of actual short term references
515
516 int cabac_init_idc;
517
518 /**
519 * @defgroup multithreading Members for slice based multithreading
520 * @{
521 */
522 struct H264Context *thread_context[MAX_THREADS];
523
524 /**
525 * current slice number, used to initalize slice_num of each thread/context
526 */
527 int current_slice;
528
529 /**
530 * Max number of threads / contexts.
531 * This is equal to AVCodecContext.thread_count unless
532 * multithreaded decoding is impossible, in which case it is
533 * reduced to 1.
534 */
535 int max_contexts;
536
537 /**
538 * 1 if the single thread fallback warning has already been
539 * displayed, 0 otherwise.
540 */
541 int single_decode_warning;
542
543 int last_slice_type;
544 /** @} */
545
546 /**
547 * pic_struct in picture timing SEI message
548 */
549 SEI_PicStructType sei_pic_struct;
550
551 /**
552 * Complement sei_pic_struct
553 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
554 * However, soft telecined frames may have these values.
555 * This is used in an attempt to flag soft telecine progressive.
556 */
557 int prev_interlaced_frame;
558
559 /**
560 * Bit set of clock types for fields/frames in picture timing SEI message.
561 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
562 * interlaced).
563 */
564 int sei_ct_type;
565
566 /**
567 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
568 */
569 int sei_dpb_output_delay;
570
571 /**
572 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
573 */
574 int sei_cpb_removal_delay;
575
576 /**
577 * recovery_frame_cnt from SEI message
578 *
579 * Set to -1 if no recovery point SEI message found or to number of frames
580 * before playback synchronizes. Frames having recovery point are key
581 * frames.
582 */
583 int sei_recovery_frame_cnt;
584
585 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
586 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
587
588 // Timestamp stuff
589 int sei_buffering_period_present; ///< Buffering period SEI flag
590 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
591
592 //SVQ3 specific fields
593 int halfpel_flag;
594 int thirdpel_flag;
595 int unknown_svq3_flag;
596 int next_slice_index;
597 uint32_t svq3_watermark_key;
598 }H264Context;
599
600
601 extern const uint8_t ff_h264_chroma_qp[52];
602
603 void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
604
605 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
606
607 /**
608 * Decode SEI
609 */
610 int ff_h264_decode_sei(H264Context *h);
611
612 /**
613 * Decode SPS
614 */
615 int ff_h264_decode_seq_parameter_set(H264Context *h);
616
617 /**
618 * Decode PPS
619 */
620 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
621
622 /**
623 * Decode a network abstraction layer unit.
624 * @param consumed is the number of bytes used as input
625 * @param length is the length of the array
626 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
627 * @return decoded bytes, might be src+1 if no escapes
628 */
629 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
630
631 /**
632 * Identify the exact end of the bitstream
633 * @return the length of the trailing, or 0 if damaged
634 */
635 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
636
637 /**
638 * Free any data that may have been allocated in the H264 context like SPS, PPS etc.
639 */
640 av_cold void ff_h264_free_context(H264Context *h);
641
642 /**
643 * Reconstruct bitstream slice_type.
644 */
645 int ff_h264_get_slice_type(const H264Context *h);
646
647 /**
648 * Allocate tables.
649 * needs width/height
650 */
651 int ff_h264_alloc_tables(H264Context *h);
652
653 /**
654 * Fill the default_ref_list.
655 */
656 int ff_h264_fill_default_ref_list(H264Context *h);
657
658 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
659 void ff_h264_fill_mbaff_ref_list(H264Context *h);
660 void ff_h264_remove_all_refs(H264Context *h);
661
662 /**
663 * Execute the reference picture marking (memory management control operations).
664 */
665 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
666
667 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
668
669 void ff_generate_sliding_window_mmcos(H264Context *h);
670
671
672 /**
673 * Check if the top & left blocks are available if needed & change the dc mode so it only uses the available blocks.
674 */
675 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
676
677 /**
678 * Check if the top & left blocks are available if needed & change the dc mode so it only uses the available blocks.
679 */
680 int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
681
682 void ff_h264_write_back_intra_pred_mode(H264Context *h);
683 void ff_h264_hl_decode_mb(H264Context *h);
684 int ff_h264_frame_start(H264Context *h);
685 int ff_h264_decode_extradata(H264Context *h);
686 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
687 av_cold int ff_h264_decode_end(AVCodecContext *avctx);
688 av_cold void ff_h264_decode_init_vlc(void);
689
690 /**
691 * Decode a macroblock
692 * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
693 */
694 int ff_h264_decode_mb_cavlc(H264Context *h);
695
696 /**
697 * Decode a CABAC coded macroblock
698 * @return 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
699 */
700 int ff_h264_decode_mb_cabac(H264Context *h);
701
702 void ff_h264_init_cabac_states(H264Context *h);
703
704 void ff_h264_direct_dist_scale_factor(H264Context * const h);
705 void ff_h264_direct_ref_list_init(H264Context * const h);
706 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
707
708 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);
709 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);
710
711 /**
712 * Reset SEI values at the beginning of the frame.
713 *
714 * @param h H.264 context.
715 */
716 void ff_h264_reset_sei(H264Context *h);
717
718
719 /*
720 o-o o-o
721 / / /
722 o-o o-o
723 ,---'
724 o-o o-o
725 / / /
726 o-o o-o
727 */
728 //This table must be here because scan8[constant] must be known at compiletime
729 static const uint8_t scan8[16 + 2*4]={
730 4+1*8, 5+1*8, 4+2*8, 5+2*8,
731 6+1*8, 7+1*8, 6+2*8, 7+2*8,
732 4+3*8, 5+3*8, 4+4*8, 5+4*8,
733 6+3*8, 7+3*8, 6+4*8, 7+4*8,
734 1+1*8, 2+1*8,
735 1+2*8, 2+2*8,
736 1+4*8, 2+4*8,
737 1+5*8, 2+5*8,
738 };
739
740 static av_always_inline uint32_t pack16to32(int a, int b){
741 #if HAVE_BIGENDIAN
742 return (b&0xFFFF) + (a<<16);
743 #else
744 return (a&0xFFFF) + (b<<16);
745 #endif
746 }
747
748 static av_always_inline uint16_t pack8to16(int a, int b){
749 #if HAVE_BIGENDIAN
750 return (b&0xFF) + (a<<8);
751 #else
752 return (a&0xFF) + (b<<8);
753 #endif
754 }
755
756 /**
757 * gets the chroma qp.
758 */
759 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
760 return h->pps.chroma_qp_table[t][qscale];
761 }
762
763 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
764
765 static void fill_decode_neighbors(H264Context *h, int mb_type){
766 MpegEncContext * const s = &h->s;
767 const int mb_xy= h->mb_xy;
768 int topleft_xy, top_xy, topright_xy, left_xy[2];
769 static const uint8_t left_block_options[4][16]={
770 {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},
771 {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},
772 {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},
773 {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}
774 };
775
776 h->topleft_partition= -1;
777
778 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
779
780 /* Wow, what a mess, why didn't they simplify the interlacing & intra
781 * stuff, I can't imagine that these complex rules are worth it. */
782
783 topleft_xy = top_xy - 1;
784 topright_xy= top_xy + 1;
785 left_xy[1] = left_xy[0] = mb_xy-1;
786 h->left_block = left_block_options[0];
787 if(FRAME_MBAFF){
788 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
789 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
790 if(s->mb_y&1){
791 if (left_mb_field_flag != curr_mb_field_flag) {
792 left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
793 if (curr_mb_field_flag) {
794 left_xy[1] += s->mb_stride;
795 h->left_block = left_block_options[3];
796 } else {
797 topleft_xy += s->mb_stride;
798 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
799 h->topleft_partition = 0;
800 h->left_block = left_block_options[1];
801 }
802 }
803 }else{
804 if(curr_mb_field_flag){
805 topleft_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
806 topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
807 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
808 }
809 if (left_mb_field_flag != curr_mb_field_flag) {
810 if (curr_mb_field_flag) {
811 left_xy[1] += s->mb_stride;
812 h->left_block = left_block_options[3];
813 } else {
814 h->left_block = left_block_options[2];
815 }
816 }
817 }
818 }
819
820 h->topleft_mb_xy = topleft_xy;
821 h->top_mb_xy = top_xy;
822 h->topright_mb_xy= topright_xy;
823 h->left_mb_xy[0] = left_xy[0];
824 h->left_mb_xy[1] = left_xy[1];
825 //FIXME do we need all in the context?
826
827 h->topleft_type = s->current_picture.mb_type[topleft_xy] ;
828 h->top_type = s->current_picture.mb_type[top_xy] ;
829 h->topright_type= s->current_picture.mb_type[topright_xy];
830 h->left_type[0] = s->current_picture.mb_type[left_xy[0]] ;
831 h->left_type[1] = s->current_picture.mb_type[left_xy[1]] ;
832
833 if(FMO){
834 if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0;
835 if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0;
836 if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
837 }else{
838 if(h->slice_table[topleft_xy ] != h->slice_num){
839 h->topleft_type = 0;
840 if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0;
841 if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
842 }
843 }
844 if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
845 }
846
847 static void fill_decode_caches(H264Context *h, int mb_type){
848 MpegEncContext * const s = &h->s;
849 int topleft_xy, top_xy, topright_xy, left_xy[2];
850 int topleft_type, top_type, topright_type, left_type[2];
851 const uint8_t * left_block= h->left_block;
852 int i;
853
854 topleft_xy = h->topleft_mb_xy ;
855 top_xy = h->top_mb_xy ;
856 topright_xy = h->topright_mb_xy;
857 left_xy[0] = h->left_mb_xy[0] ;
858 left_xy[1] = h->left_mb_xy[1] ;
859 topleft_type = h->topleft_type ;
860 top_type = h->top_type ;
861 topright_type= h->topright_type ;
862 left_type[0] = h->left_type[0] ;
863 left_type[1] = h->left_type[1] ;
864
865 if(!IS_SKIP(mb_type)){
866 if(IS_INTRA(mb_type)){
867 int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
868 h->topleft_samples_available=
869 h->top_samples_available=
870 h->left_samples_available= 0xFFFF;
871 h->topright_samples_available= 0xEEEA;
872
873 if(!(top_type & type_mask)){
874 h->topleft_samples_available= 0xB3FF;
875 h->top_samples_available= 0x33FF;
876 h->topright_samples_available= 0x26EA;
877 }
878 if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
879 if(IS_INTERLACED(mb_type)){
880 if(!(left_type[0] & type_mask)){
881 h->topleft_samples_available&= 0xDFFF;
882 h->left_samples_available&= 0x5FFF;
883 }
884 if(!(left_type[1] & type_mask)){
885 h->topleft_samples_available&= 0xFF5F;
886 h->left_samples_available&= 0xFF5F;
887 }
888 }else{
889 int left_typei = s->current_picture.mb_type[left_xy[0] + s->mb_stride];
890
891 assert(left_xy[0] == left_xy[1]);
892 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
893 h->topleft_samples_available&= 0xDF5F;
894 h->left_samples_available&= 0x5F5F;
895 }
896 }
897 }else{
898 if(!(left_type[0] & type_mask)){
899 h->topleft_samples_available&= 0xDF5F;
900 h->left_samples_available&= 0x5F5F;
901 }
902 }
903
904 if(!(topleft_type & type_mask))
905 h->topleft_samples_available&= 0x7FFF;
906
907 if(!(topright_type & type_mask))
908 h->topright_samples_available&= 0xFBFF;
909
910 if(IS_INTRA4x4(mb_type)){
911 if(IS_INTRA4x4(top_type)){
912 AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]);
913 }else{
914 h->intra4x4_pred_mode_cache[4+8*0]=
915 h->intra4x4_pred_mode_cache[5+8*0]=
916 h->intra4x4_pred_mode_cache[6+8*0]=
917 h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask);
918 }
919 for(i=0; i<2; i++){
920 if(IS_INTRA4x4(left_type[i])){
921 int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]];
922 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]];
923 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]];
924 }else{
925 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
926 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[i] & type_mask);
927 }
928 }
929 }
930 }
931
932
933 /*
934 0 . T T. T T T T
935 1 L . .L . . . .
936 2 L . .L . . . .
937 3 . T TL . . . .
938 4 L . .L . . . .
939 5 L . .. . . . .
940 */
941 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
942 if(top_type){
943 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
944 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
945 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
946
947 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
948 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
949 }else {
950 h->non_zero_count_cache[1+8*0]=
951 h->non_zero_count_cache[2+8*0]=
952
953 h->non_zero_count_cache[1+8*3]=
954 h->non_zero_count_cache[2+8*3]=
955 AV_WN32A(&h->non_zero_count_cache[4+8*0], CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040);
956 }
957
958 for (i=0; i<2; i++) {
959 if(left_type[i]){
960 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
961 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
962 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
963 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
964 }else{
965 h->non_zero_count_cache[3+8*1 + 2*8*i]=
966 h->non_zero_count_cache[3+8*2 + 2*8*i]=
967 h->non_zero_count_cache[0+8*1 + 8*i]=
968 h->non_zero_count_cache[0+8*4 + 8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
969 }
970 }
971
972 if( CABAC ) {
973 // top_cbp
974 if(top_type) {
975 h->top_cbp = h->cbp_table[top_xy];
976 } else {
977 h->top_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
978 }
979 // left_cbp
980 if (left_type[0]) {
981 h->left_cbp = (h->cbp_table[left_xy[0]] & 0x1f0)
982 | ((h->cbp_table[left_xy[0]]>>(left_block[0]&(~1)))&2)
983 | (((h->cbp_table[left_xy[1]]>>(left_block[2]&(~1)))&2) << 2);
984 } else {
985 h->left_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
986 }
987 }
988 }
989
990 #if 1
991 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
992 int list;
993 for(list=0; list<h->list_count; list++){
994 if(!USES_LIST(mb_type, list)){
995 /*if(!h->mv_cache_clean[list]){
996 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
997 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
998 h->mv_cache_clean[list]= 1;
999 }*/
1000 continue;
1001 }
1002 assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
1003
1004 h->mv_cache_clean[list]= 0;
1005
1006 if(USES_LIST(top_type, list)){
1007 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1008 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1009 h->ref_cache[list][scan8[0] + 0 - 1*8]=
1010 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 2];
1011 h->ref_cache[list][scan8[0] + 2 - 1*8]=
1012 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 3];
1013 }else{
1014 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1015 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
1016 }
1017
1018 if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){
1019 for(i=0; i<2; i++){
1020 int cache_idx = scan8[0] - 1 + i*2*8;
1021 if(USES_LIST(left_type[i], list)){
1022 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1023 const int b8_xy= 4*left_xy[i] + 1;
1024 AV_COPY32(h->mv_cache[list][cache_idx ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]);
1025 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]]);
1026 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + (left_block[0+i*2]&~1)];
1027 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + (left_block[1+i*2]&~1)];
1028 }else{
1029 AV_ZERO32(h->mv_cache [list][cache_idx ]);
1030 AV_ZERO32(h->mv_cache [list][cache_idx+8]);
1031 h->ref_cache[list][cache_idx ]=
1032 h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1033 }
1034 }
1035 }else{
1036 if(USES_LIST(left_type[0], list)){
1037 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1038 const int b8_xy= 4*left_xy[0] + 1;
1039 AV_COPY32(h->mv_cache[list][scan8[0] - 1], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]]);
1040 h->ref_cache[list][scan8[0] - 1]= s->current_picture.ref_index[list][b8_xy + (left_block[0]&~1)];
1041 }else{
1042 AV_ZERO32(h->mv_cache [list][scan8[0] - 1]);
1043 h->ref_cache[list][scan8[0] - 1]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1044 }
1045 }
1046
1047 if(USES_LIST(topright_type, list)){
1048 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1049 AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]);
1050 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][4*topright_xy + 2];
1051 }else{
1052 AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]);
1053 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1054 }
1055 if(h->ref_cache[list][scan8[0] + 4 - 1*8] < 0){
1056 if(USES_LIST(topleft_type, list)){
1057 const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
1058 const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2);
1059 AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]);
1060 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1061 }else{
1062 AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]);
1063 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1064 }
1065 }
1066
1067 if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
1068 continue;
1069
1070 if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
1071 h->ref_cache[list][scan8[4 ]] =
1072 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1073 AV_ZERO32(h->mv_cache [list][scan8[4 ]]);
1074 AV_ZERO32(h->mv_cache [list][scan8[12]]);
1075
1076 if( CABAC ) {
1077 /* XXX beurk, Load mvd */
1078 if(USES_LIST(top_type, list)){
1079 const int b_xy= h->mb2br_xy[top_xy];
1080 AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1081 }else{
1082 AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1083 }
1084 if(USES_LIST(left_type[0], list)){
1085 const int b_xy= h->mb2br_xy[left_xy[0]] + 6;
1086 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy - left_block[0]]);
1087 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy - left_block[1]]);
1088 }else{
1089 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]);
1090 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]);
1091 }
1092 if(USES_LIST(left_type[1], list)){
1093 const int b_xy= h->mb2br_xy[left_xy[1]] + 6;
1094 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy - left_block[2]]);
1095 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy - left_block[3]]);
1096 }else{
1097 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]);
1098 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]);
1099 }
1100 AV_ZERO16(h->mvd_cache [list][scan8[4 ]]);
1101 AV_ZERO16(h->mvd_cache [list][scan8[12]]);
1102 if(h->slice_type_nos == FF_B_TYPE){
1103 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
1104
1105 if(IS_DIRECT(top_type)){
1106 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101u*(MB_TYPE_DIRECT2>>1));
1107 }else if(IS_8X8(top_type)){
1108 int b8_xy = 4*top_xy;
1109 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy + 2];
1110 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 3];
1111 }else{
1112 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1));
1113 }
1114
1115 if(IS_DIRECT(left_type[0]))
1116 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
1117 else if(IS_8X8(left_type[0]))
1118 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[4*left_xy[0] + 1 + (left_block[0]&~1)];
1119 else
1120 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
1121
1122 if(IS_DIRECT(left_type[1]))
1123 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
1124 else if(IS_8X8(left_type[1]))
1125 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[4*left_xy[1] + 1 + (left_block[2]&~1)];
1126 else
1127 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
1128 }
1129 }
1130 }
1131 if(FRAME_MBAFF){
1132 #define MAP_MVS\
1133 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1134 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1135 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1136 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1137 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1138 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1139 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1140 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1141 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1142 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1143 if(MB_FIELD){
1144 #define MAP_F2F(idx, mb_type)\
1145 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1146 h->ref_cache[list][idx] <<= 1;\
1147 h->mv_cache[list][idx][1] /= 2;\
1148 h->mvd_cache[list][idx][1] >>=1;\
1149 }
1150 MAP_MVS
1151 #undef MAP_F2F
1152 }else{
1153 #define MAP_F2F(idx, mb_type)\
1154 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1155 h->ref_cache[list][idx] >>= 1;\
1156 h->mv_cache[list][idx][1] <<= 1;\
1157 h->mvd_cache[list][idx][1] <<= 1;\
1158 }
1159 MAP_MVS
1160 #undef MAP_F2F
1161 }
1162 }
1163 }
1164 }
1165 #endif
1166
1167 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1168 }
1169
1170 /**
1171 * gets the predicted intra4x4 prediction mode.
1172 */
1173 static inline int pred_intra_mode(H264Context *h, int n){
1174 const int index8= scan8[n];
1175 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1176 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1177 const int min= FFMIN(left, top);
1178
1179 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1180
1181 if(min<0) return DC_PRED;
1182 else return min;
1183 }
1184
1185 static inline void write_back_non_zero_count(H264Context *h){
1186 const int mb_xy= h->mb_xy;
1187
1188 AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1189 AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1190 AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[0+8*5]);
1191 AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8*3]);
1192 AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1193 }
1194
1195 static inline void write_back_motion(H264Context *h, int mb_type){
1196 MpegEncContext * const s = &h->s;
1197 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
1198 const int b8_xy= 4*h->mb_xy;
1199 int list;
1200
1201 if(!USES_LIST(mb_type, 0))
1202 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
1203
1204 for(list=0; list<h->list_count; list++){
1205 int y, b_stride;
1206 int16_t (*mv_dst)[2];
1207 int16_t (*mv_src)[2];
1208
1209 if(!USES_LIST(mb_type, list))
1210 continue;
1211
1212 b_stride = h->b_stride;
1213 mv_dst = &s->current_picture.motion_val[list][b_xy];
1214 mv_src = &h->mv_cache[list][scan8[0]];
1215 for(y=0; y<4; y++){
1216 AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1217 }
1218 if( CABAC ) {
1219 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
1220 uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1221 if(IS_SKIP(mb_type))
1222 AV_ZERO128(mvd_dst);
1223 else{
1224 AV_COPY64(mvd_dst, mvd_src + 8*3);
1225 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
1226 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
1227 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
1228 }
1229 }
1230
1231 {
1232 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1233 ref_index[0+0*2]= h->ref_cache[list][scan8[0]];
1234 ref_index[1+0*2]= h->ref_cache[list][scan8[4]];
1235 ref_index[0+1*2]= h->ref_cache[list][scan8[8]];
1236 ref_index[1+1*2]= h->ref_cache[list][scan8[12]];
1237 }
1238 }
1239
1240 if(h->slice_type_nos == FF_B_TYPE && CABAC){
1241 if(IS_8X8(mb_type)){
1242 uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
1243 direct_table[1] = h->sub_mb_type[1]>>1;
1244 direct_table[2] = h->sub_mb_type[2]>>1;
1245 direct_table[3] = h->sub_mb_type[3]>>1;
1246 }
1247 }
1248 }
1249
1250 static inline int get_dct8x8_allowed(H264Context *h){
1251 if(h->sps.direct_8x8_inference_flag)
1252 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL));
1253 else
1254 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1255 }
1256
1257 /**
1258 * decodes a P_SKIP or B_SKIP macroblock
1259 */
1260 static void decode_mb_skip(H264Context *h){
1261 MpegEncContext * const s = &h->s;
1262 const int mb_xy= h->mb_xy;
1263 int mb_type=0;
1264
1265 memset(h->non_zero_count[mb_xy], 0, 32);
1266 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1267
1268 if(MB_FIELD)
1269 mb_type|= MB_TYPE_INTERLACED;
1270
1271 if( h->slice_type_nos == FF_B_TYPE )
1272 {
1273 // just for fill_caches. pred_direct_motion will set the real mb_type
1274 mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1275 if(h->direct_spatial_mv_pred){
1276 fill_decode_neighbors(h, mb_type);
1277 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1278 }
1279 ff_h264_pred_direct_motion(h, &mb_type);
1280 mb_type|= MB_TYPE_SKIP;
1281 }
1282 else
1283 {
1284 int mx, my;
1285 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1286
1287 fill_decode_neighbors(h, mb_type);
1288 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1289 pred_pskip_motion(h, &mx, &my);
1290 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1291 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1292 }
1293
1294 write_back_motion(h, mb_type);
1295 s->current_picture.mb_type[mb_xy]= mb_type;
1296 s->current_picture.qscale_table[mb_xy]= s->qscale;
1297 h->slice_table[ mb_xy ]= h->slice_num;
1298 h->prev_mb_skipped= 1;
1299 }
1300
1301 #include "h264_mvpred.h" //For pred_pskip_motion()
1302
1303 #endif /* AVCODEC_H264_H */