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 / MPEG-4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
28 #ifndef AVCODEC_H264DEC_H
29 #define AVCODEC_H264DEC_H
30
31 #include "libavutil/buffer.h"
32 #include "libavutil/intreadwrite.h"
33 #include "libavutil/mem_internal.h"
34 #include "libavutil/thread.h"
35
36 #include "cabac.h"
37 #include "error_resilience.h"
38 #include "h264_parse.h"
39 #include "h264_ps.h"
40 #include "h264_sei.h"
41 #include "h2645_parse.h"
42 #include "h264chroma.h"
43 #include "h264dsp.h"
44 #include "h264pred.h"
45 #include "h264qpel.h"
46 #include "internal.h"
47 #include "mpegutils.h"
48 #include "parser.h"
49 #include "qpeldsp.h"
50 #include "rectangle.h"
51 #include "videodsp.h"
52
53 #define H264_MAX_PICTURE_COUNT 36
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 FMO 0
64
65 /**
66 * The maximum number of slices supported by the decoder.
67 * must be a power of 2
68 */
69 #define MAX_SLICES 32
70
71 #ifdef ALLOW_INTERLACE
72 #define MB_MBAFF(h) (h)->mb_mbaff
73 #define MB_FIELD(sl) (sl)->mb_field_decoding_flag
74 #define FRAME_MBAFF(h) (h)->mb_aff_frame
75 #define FIELD_PICTURE(h) ((h)->picture_structure != PICT_FRAME)
76 #define LEFT_MBS 2
77 #define LTOP 0
78 #define LBOT 1
79 #define LEFT(i) (i)
80 #else
81 #define MB_MBAFF(h) 0
82 #define MB_FIELD(sl) 0
83 #define FRAME_MBAFF(h) 0
84 #define FIELD_PICTURE(h) 0
85 #undef IS_INTERLACED
86 #define IS_INTERLACED(mb_type) 0
87 #define LEFT_MBS 1
88 #define LTOP 0
89 #define LBOT 0
90 #define LEFT(i) 0
91 #endif
92 #define FIELD_OR_MBAFF_PICTURE(h) (FRAME_MBAFF(h) || FIELD_PICTURE(h))
93
94 #ifndef CABAC
95 #define CABAC(h) (h)->ps.pps->cabac
96 #endif
97
98 #define CHROMA(h) ((h)->ps.sps->chroma_format_idc)
99 #define CHROMA422(h) ((h)->ps.sps->chroma_format_idc == 2)
100 #define CHROMA444(h) ((h)->ps.sps->chroma_format_idc == 3)
101
102 #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit
103 #define MB_TYPE_8x8DCT 0x01000000
104 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
105 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
106
107 /**
108 * Memory management control operation opcode.
109 */
110 typedef enum MMCOOpcode {
111 MMCO_END = 0,
112 MMCO_SHORT2UNUSED,
113 MMCO_LONG2UNUSED,
114 MMCO_SHORT2LONG,
115 MMCO_SET_MAX_LONG,
116 MMCO_RESET,
117 MMCO_LONG,
118 } MMCOOpcode;
119
120 /**
121 * Memory management control operation.
122 */
123 typedef struct MMCO {
124 MMCOOpcode opcode;
125 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
126 int long_arg; ///< index, pic_num, or num long refs depending on opcode
127 } MMCO;
128
129 typedef struct H264Picture {
130 AVFrame *f;
131 ThreadFrame tf;
132
133 AVBufferRef *qscale_table_buf;
134 int8_t *qscale_table;
135
136 AVBufferRef *motion_val_buf[2];
137 int16_t (*motion_val[2])[2];
138
139 AVBufferRef *mb_type_buf;
140 uint32_t *mb_type;
141
142 AVBufferRef *hwaccel_priv_buf;
143 void *hwaccel_picture_private; ///< hardware accelerator private data
144
145 AVBufferRef *ref_index_buf[2];
146 int8_t *ref_index[2];
147
148 int field_poc[2]; ///< top/bottom POC
149 int poc; ///< frame POC
150 int frame_num; ///< frame_num (raw frame_num from slice header)
151 int mmco_reset; /**< MMCO_RESET set this 1. Reordering code must
152 not mix pictures before and after MMCO_RESET. */
153 int pic_id; /**< pic_num (short -> no wrap version of pic_num,
154 pic_num & max_pic_num; long -> long_pic_num) */
155 int long_ref; ///< 1->long term reference 0->short term reference
156 int ref_poc[2][2][32]; ///< POCs of the frames/fields used as reference (FIXME need per slice)
157 int ref_count[2][2]; ///< number of entries in ref_poc (FIXME need per slice)
158 int mbaff; ///< 1 -> MBAFF frame 0-> not MBAFF
159 int field_picture; ///< whether or not picture was encoded in separate fields
160
161 int reference;
162 int recovered; ///< picture at IDR or recovery point + recovery count
163 int invalid_gap;
164 int sei_recovery_frame_cnt;
165
166 AVBufferRef *pps_buf;
167 const PPS *pps;
168
169 int mb_width, mb_height;
170 int mb_stride;
171 } H264Picture;
172
173 typedef struct H264Ref {
174 uint8_t *data[3];
175 int linesize[3];
176
177 int reference;
178 int poc;
179 int pic_id;
180
181 H264Picture *parent;
182 } H264Ref;
183
184 typedef struct H264SliceContext {
185 struct H264Context *h264;
186 GetBitContext gb;
187 ERContext er;
188
189 int slice_num;
190 int slice_type;
191 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
192 int slice_type_fixed;
193
194 int qscale;
195 int chroma_qp[2]; // QPc
196 int qp_thresh; ///< QP threshold to skip loopfilter
197 int last_qscale_diff;
198
199 // deblock
200 int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
201 int slice_alpha_c0_offset;
202 int slice_beta_offset;
203
204 H264PredWeightTable pwt;
205
206 int prev_mb_skipped;
207 int next_mb_skipped;
208
209 int chroma_pred_mode;
210 int intra16x16_pred_mode;
211
212 int8_t intra4x4_pred_mode_cache[5 * 8];
213 int8_t(*intra4x4_pred_mode);
214
215 int topleft_mb_xy;
216 int top_mb_xy;
217 int topright_mb_xy;
218 int left_mb_xy[LEFT_MBS];
219
220 int topleft_type;
221 int top_type;
222 int topright_type;
223 int left_type[LEFT_MBS];
224
225 const uint8_t *left_block;
226 int topleft_partition;
227
228 unsigned int topleft_samples_available;
229 unsigned int top_samples_available;
230 unsigned int topright_samples_available;
231 unsigned int left_samples_available;
232
233 ptrdiff_t linesize, uvlinesize;
234 ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
235 ptrdiff_t mb_uvlinesize;
236
237 int mb_x, mb_y;
238 int mb_xy;
239 int resync_mb_x;
240 int resync_mb_y;
241 unsigned int first_mb_addr;
242 // index of the first MB of the next slice
243 int next_slice_idx;
244 int mb_skip_run;
245 int is_complex;
246
247 int picture_structure;
248 int mb_field_decoding_flag;
249 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
250
251 int redundant_pic_count;
252
253 /**
254 * number of neighbors (top and/or left) that used 8x8 dct
255 */
256 int neighbor_transform_size;
257
258 int direct_spatial_mv_pred;
259 int col_parity;
260 int col_fieldoff;
261
262 int cbp;
263 int top_cbp;
264 int left_cbp;
265
266 int dist_scale_factor[32];
267 int dist_scale_factor_field[2][32];
268 int map_col_to_list0[2][16 + 32];
269 int map_col_to_list0_field[2][2][16 + 32];
270
271 /**
272 * num_ref_idx_l0/1_active_minus1 + 1
273 */
274 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
275 unsigned int list_count;
276 H264Ref ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
277 * Reordered version of default_ref_list
278 * according to picture reordering in slice header */
279 struct {
280 uint8_t op;
281 uint32_t val;
282 } ref_modifications[2][32];
283 int nb_ref_modifications[2];
284
285 unsigned int pps_id;
286
287 const uint8_t *intra_pcm_ptr;
288 int16_t *dc_val_base;
289
290 uint8_t *bipred_scratchpad;
291 uint8_t *edge_emu_buffer;
292 uint8_t (*top_borders[2])[(16 * 3) * 2];
293 int bipred_scratchpad_allocated;
294 int edge_emu_buffer_allocated;
295 int top_borders_allocated[2];
296
297 /**
298 * non zero coeff count cache.
299 * is 64 if not available.
300 */
301 DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
302
303 /**
304 * Motion vector cache.
305 */
306 DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
307 DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
308 DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
309 uint8_t direct_cache[5 * 8];
310
311 DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
312
313 ///< as a DCT coefficient is int32_t in high depth, we need to reserve twice the space.
314 DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2];
315 DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
316 ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either
317 ///< check that i is not too large or ensure that there is some unused stuff after mb
318 int16_t mb_padding[256 * 2];
319
320 uint8_t (*mvd_table[2])[2];
321
322 /**
323 * Cabac
324 */
325 CABACContext cabac;
326 uint8_t cabac_state[1024];
327 int cabac_init_idc;
328
329 MMCO mmco[MAX_MMCO_COUNT];
330 int nb_mmco;
331 int explicit_ref_marking;
332
333 int frame_num;
334 int poc_lsb;
335 int delta_poc_bottom;
336 int delta_poc[2];
337 int curr_pic_num;
338 int max_pic_num;
339 } H264SliceContext;
340
341 /**
342 * H264Context
343 */
344 typedef struct H264Context {
345 const AVClass *class;
346 AVCodecContext *avctx;
347 VideoDSPContext vdsp;
348 H264DSPContext h264dsp;
349 H264ChromaContext h264chroma;
350 H264QpelContext h264qpel;
351
352 H264Picture DPB[H264_MAX_PICTURE_COUNT];
353 H264Picture *cur_pic_ptr;
354 H264Picture cur_pic;
355 H264Picture last_pic_for_ec;
356
357 H264SliceContext *slice_ctx;
358 int nb_slice_ctx;
359 int nb_slice_ctx_queued;
360
361 H2645Packet pkt;
362
363 int pixel_shift; ///< 0 for 8-bit H.264, 1 for high-bit-depth H.264
364
365 /* coded dimensions -- 16 * mb w/h */
366 int width, height;
367 int chroma_x_shift, chroma_y_shift;
368
369 int droppable;
370 int coded_picture_number;
371
372 int context_initialized;
373 int flags;
374 int workaround_bugs;
375 int x264_build;
376 /* Set when slice threading is used and at least one slice uses deblocking
377 * mode 1 (i.e. across slice boundaries). Then we disable the loop filter
378 * during normal MB decoding and execute it serially at the end.
379 */
380 int postpone_filter;
381
382 /*
383 * Set to 1 when the current picture is IDR, 0 otherwise.
384 */
385 int picture_idr;
386
387 int crop_left;
388 int crop_right;
389 int crop_top;
390 int crop_bottom;
391
392 int8_t(*intra4x4_pred_mode);
393 H264PredContext hpc;
394
395 uint8_t (*non_zero_count)[48];
396
397 #define LIST_NOT_USED -1 // FIXME rename?
398 #define PART_NOT_AVAILABLE -2
399
400 /**
401 * block_offset[ 0..23] for frame macroblocks
402 * block_offset[24..47] for field macroblocks
403 */
404 int block_offset[2 * (16 * 3)];
405
406 uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
407 uint32_t *mb2br_xy;
408 int b_stride; // FIXME use s->b4_stride
409
410 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
411
412 // interlacing specific flags
413 int mb_aff_frame;
414 int picture_structure;
415 int first_field;
416
417 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
418
419 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
420 uint16_t *cbp_table;
421
422 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
423 uint8_t *chroma_pred_mode_table;
424 uint8_t (*mvd_table[2])[2];
425 uint8_t *direct_table;
426
427 uint8_t scan_padding[16];
428 uint8_t zigzag_scan[16];
429 uint8_t zigzag_scan8x8[64];
430 uint8_t zigzag_scan8x8_cavlc[64];
431 uint8_t field_scan[16];
432 uint8_t field_scan8x8[64];
433 uint8_t field_scan8x8_cavlc[64];
434 uint8_t zigzag_scan_q0[16];
435 uint8_t zigzag_scan8x8_q0[64];
436 uint8_t zigzag_scan8x8_cavlc_q0[64];
437 uint8_t field_scan_q0[16];
438 uint8_t field_scan8x8_q0[64];
439 uint8_t field_scan8x8_cavlc_q0[64];
440
441 int mb_y;
442 int mb_height, mb_width;
443 int mb_stride;
444 int mb_num;
445
446 // =============================================================
447 // Things below are not used in the MB or more inner code
448
449 int nal_ref_idc;
450 int nal_unit_type;
451
452 int has_slice; ///< slice NAL is found in the packet, set by decode_nal_units, its state does not need to be preserved outside h264_decode_frame()
453
454 /**
455 * Used to parse AVC variant of H.264
456 */
457 int is_avc; ///< this flag is != 0 if codec is avc1
458 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
459
460 int bit_depth_luma; ///< luma bit depth from sps to detect changes
461 int chroma_format_idc; ///< chroma format from sps to detect changes
462
463 H264ParamSets ps;
464
465 uint16_t *slice_table_base;
466
467 H264POCContext poc;
468
469 H264Ref default_ref[2];
470 H264Picture *short_ref[32];
471 H264Picture *long_ref[32];
472 H264Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
473 int last_pocs[MAX_DELAYED_PIC_COUNT];
474 H264Picture *next_output_pic;
475 int next_outputed_poc;
476
477 /**
478 * memory management control operations buffer.
479 */
480 MMCO mmco[MAX_MMCO_COUNT];
481 int nb_mmco;
482 int mmco_reset;
483 int explicit_ref_marking;
484
485 int long_ref_count; ///< number of actual long term references
486 int short_ref_count; ///< number of actual short term references
487
488 /**
489 * @name Members for slice based multithreading
490 * @{
491 */
492 /**
493 * current slice number, used to initialize slice_num of each thread/context
494 */
495 int current_slice;
496
497 /** @} */
498
499 /**
500 * Complement sei_pic_struct
501 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
502 * However, soft telecined frames may have these values.
503 * This is used in an attempt to flag soft telecine progressive.
504 */
505 int prev_interlaced_frame;
506
507 /**
508 * Are the SEI recovery points looking valid.
509 */
510 int valid_recovery_point;
511
512 /**
513 * recovery_frame is the frame_num at which the next frame should
514 * be fully constructed.
515 *
516 * Set to -1 when not expecting a recovery point.
517 */
518 int recovery_frame;
519
520 /**
521 * We have seen an IDR, so all the following frames in coded order are correctly
522 * decodable.
523 */
524 #define FRAME_RECOVERED_IDR (1 << 0)
525 /**
526 * Sufficient number of frames have been decoded since a SEI recovery point,
527 * so all the following frames in presentation order are correct.
528 */
529 #define FRAME_RECOVERED_SEI (1 << 1)
530
531 int frame_recovered; ///< Initial frame has been completely recovered
532
533 int has_recovery_point;
534
535 int missing_fields;
536
537 /* for frame threading, this is set to 1
538 * after finish_setup() has been called, so we cannot modify
539 * some context properties (which are supposed to stay constant between
540 * slices) anymore */
541 int setup_finished;
542
543 int cur_chroma_format_idc;
544 int cur_bit_depth_luma;
545 int16_t slice_row[MAX_SLICES]; ///< to detect when MAX_SLICES is too low
546
547 /* original AVCodecContext dimensions, used to handle container
548 * cropping */
549 int width_from_caller;
550 int height_from_caller;
551
552 int enable_er;
553
554 H264SEIContext sei;
555
556 AVBufferPool *qscale_table_pool;
557 AVBufferPool *mb_type_pool;
558 AVBufferPool *motion_val_pool;
559 AVBufferPool *ref_index_pool;
560 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
561 } H264Context;
562
563 extern const uint16_t ff_h264_mb_sizes[4];
564
565 /**
566 * Reconstruct bitstream slice_type.
567 */
568 int ff_h264_get_slice_type(const H264SliceContext *sl);
569
570 /**
571 * Allocate tables.
572 * needs width/height
573 */
574 int ff_h264_alloc_tables(H264Context *h);
575
576 int ff_h264_decode_ref_pic_list_reordering(H264SliceContext *sl, void *logctx);
577 int ff_h264_build_ref_list(H264Context *h, H264SliceContext *sl);
578 void ff_h264_remove_all_refs(H264Context *h);
579
580 /**
581 * Execute the reference picture marking (memory management control operations).
582 */
583 int ff_h264_execute_ref_pic_marking(H264Context *h);
584
585 int ff_h264_decode_ref_pic_marking(H264SliceContext *sl, GetBitContext *gb,
586 const H2645NAL *nal, void *logctx);
587
588 void ff_h264_hl_decode_mb(const H264Context *h, H264SliceContext *sl);
589 void ff_h264_decode_init_vlc(void);
590
591 /**
592 * Decode a macroblock
593 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
594 */
595 int ff_h264_decode_mb_cavlc(const H264Context *h, H264SliceContext *sl);
596
597 /**
598 * Decode a CABAC coded macroblock
599 * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
600 */
601 int ff_h264_decode_mb_cabac(const H264Context *h, H264SliceContext *sl);
602
603 void ff_h264_init_cabac_states(const H264Context *h, H264SliceContext *sl);
604
605 void ff_h264_direct_dist_scale_factor(const H264Context *const h, H264SliceContext *sl);
606 void ff_h264_direct_ref_list_init(const H264Context *const h, H264SliceContext *sl);
607 void ff_h264_pred_direct_motion(const H264Context *const h, H264SliceContext *sl,
608 int *mb_type);
609
610 void ff_h264_filter_mb_fast(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
611 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
612 unsigned int linesize, unsigned int uvlinesize);
613 void ff_h264_filter_mb(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
614 uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
615 unsigned int linesize, unsigned int uvlinesize);
616
617 /*
618 * o-o o-o
619 * / / /
620 * o-o o-o
621 * ,---'
622 * o-o o-o
623 * / / /
624 * o-o o-o
625 */
626
627 /* Scan8 organization:
628 * 0 1 2 3 4 5 6 7
629 * 0 DY y y y y y
630 * 1 y Y Y Y Y
631 * 2 y Y Y Y Y
632 * 3 y Y Y Y Y
633 * 4 y Y Y Y Y
634 * 5 DU u u u u u
635 * 6 u U U U U
636 * 7 u U U U U
637 * 8 u U U U U
638 * 9 u U U U U
639 * 10 DV v v v v v
640 * 11 v V V V V
641 * 12 v V V V V
642 * 13 v V V V V
643 * 14 v V V V V
644 * DY/DU/DV are for luma/chroma DC.
645 */
646
647 #define LUMA_DC_BLOCK_INDEX 48
648 #define CHROMA_DC_BLOCK_INDEX 49
649
650 // This table must be here because scan8[constant] must be known at compiletime
651 static const uint8_t scan8[16 * 3 + 3] = {
652 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
653 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
654 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
655 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
656 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
657 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
658 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
659 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
660 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
661 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
662 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
663 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
664 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
665 };
666
pack16to32(unsigned a,unsigned b)667 static av_always_inline uint32_t pack16to32(unsigned a, unsigned b)
668 {
669 #if HAVE_BIGENDIAN
670 return (b & 0xFFFF) + (a << 16);
671 #else
672 return (a & 0xFFFF) + (b << 16);
673 #endif
674 }
675
pack8to16(unsigned a,unsigned b)676 static av_always_inline uint16_t pack8to16(unsigned a, unsigned b)
677 {
678 #if HAVE_BIGENDIAN
679 return (b & 0xFF) + (a << 8);
680 #else
681 return (a & 0xFF) + (b << 8);
682 #endif
683 }
684
685 /**
686 * Get the chroma qp.
687 */
get_chroma_qp(const PPS * pps,int t,int qscale)688 static av_always_inline int get_chroma_qp(const PPS *pps, int t, int qscale)
689 {
690 return pps->chroma_qp_table[t][qscale];
691 }
692
693 /**
694 * Get the predicted intra4x4 prediction mode.
695 */
pred_intra_mode(const H264Context * h,H264SliceContext * sl,int n)696 static av_always_inline int pred_intra_mode(const H264Context *h,
697 H264SliceContext *sl, int n)
698 {
699 const int index8 = scan8[n];
700 const int left = sl->intra4x4_pred_mode_cache[index8 - 1];
701 const int top = sl->intra4x4_pred_mode_cache[index8 - 8];
702 const int min = FFMIN(left, top);
703
704 ff_tlog(h->avctx, "mode:%d %d min:%d\n", left, top, min);
705
706 if (min < 0)
707 return DC_PRED;
708 else
709 return min;
710 }
711
write_back_intra_pred_mode(const H264Context * h,H264SliceContext * sl)712 static av_always_inline void write_back_intra_pred_mode(const H264Context *h,
713 H264SliceContext *sl)
714 {
715 int8_t *i4x4 = sl->intra4x4_pred_mode + h->mb2br_xy[sl->mb_xy];
716 int8_t *i4x4_cache = sl->intra4x4_pred_mode_cache;
717
718 AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
719 i4x4[4] = i4x4_cache[7 + 8 * 3];
720 i4x4[5] = i4x4_cache[7 + 8 * 2];
721 i4x4[6] = i4x4_cache[7 + 8 * 1];
722 }
723
write_back_non_zero_count(const H264Context * h,H264SliceContext * sl)724 static av_always_inline void write_back_non_zero_count(const H264Context *h,
725 H264SliceContext *sl)
726 {
727 const int mb_xy = sl->mb_xy;
728 uint8_t *nnz = h->non_zero_count[mb_xy];
729 uint8_t *nnz_cache = sl->non_zero_count_cache;
730
731 AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
732 AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
733 AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
734 AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
735 AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
736 AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
737 AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
738 AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
739
740 if (!h->chroma_y_shift) {
741 AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
742 AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
743 AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
744 AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
745 }
746 }
747
write_back_motion_list(const H264Context * h,H264SliceContext * sl,int b_stride,int b_xy,int b8_xy,int mb_type,int list)748 static av_always_inline void write_back_motion_list(const H264Context *h,
749 H264SliceContext *sl,
750 int b_stride,
751 int b_xy, int b8_xy,
752 int mb_type, int list)
753 {
754 int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
755 int16_t(*mv_src)[2] = &sl->mv_cache[list][scan8[0]];
756 AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
757 AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
758 AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
759 AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
760 if (CABAC(h)) {
761 uint8_t (*mvd_dst)[2] = &sl->mvd_table[list][FMO ? 8 * sl->mb_xy
762 : h->mb2br_xy[sl->mb_xy]];
763 uint8_t(*mvd_src)[2] = &sl->mvd_cache[list][scan8[0]];
764 if (IS_SKIP(mb_type)) {
765 AV_ZERO128(mvd_dst);
766 } else {
767 AV_COPY64(mvd_dst, mvd_src + 8 * 3);
768 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
769 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
770 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
771 }
772 }
773
774 {
775 int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
776 int8_t *ref_cache = sl->ref_cache[list];
777 ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
778 ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
779 ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
780 ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
781 }
782 }
783
write_back_motion(const H264Context * h,H264SliceContext * sl,int mb_type)784 static av_always_inline void write_back_motion(const H264Context *h,
785 H264SliceContext *sl,
786 int mb_type)
787 {
788 const int b_stride = h->b_stride;
789 const int b_xy = 4 * sl->mb_x + 4 * sl->mb_y * h->b_stride; // try mb2b(8)_xy
790 const int b8_xy = 4 * sl->mb_xy;
791
792 if (USES_LIST(mb_type, 0)) {
793 write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 0);
794 } else {
795 fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
796 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
797 }
798 if (USES_LIST(mb_type, 1))
799 write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 1);
800
801 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) {
802 if (IS_8X8(mb_type)) {
803 uint8_t *direct_table = &h->direct_table[4 * sl->mb_xy];
804 direct_table[1] = sl->sub_mb_type[1] >> 1;
805 direct_table[2] = sl->sub_mb_type[2] >> 1;
806 direct_table[3] = sl->sub_mb_type[3] >> 1;
807 }
808 }
809 }
810
get_dct8x8_allowed(const H264Context * h,H264SliceContext * sl)811 static av_always_inline int get_dct8x8_allowed(const H264Context *h, H264SliceContext *sl)
812 {
813 if (h->ps.sps->direct_8x8_inference_flag)
814 return !(AV_RN64A(sl->sub_mb_type) &
815 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
816 0x0001000100010001ULL));
817 else
818 return !(AV_RN64A(sl->sub_mb_type) &
819 ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
820 0x0001000100010001ULL));
821 }
822
find_start_code(const uint8_t * buf,int buf_size,int buf_index,int next_avc)823 static inline int find_start_code(const uint8_t *buf, int buf_size,
824 int buf_index, int next_avc)
825 {
826 uint32_t state = -1;
827
828 buf_index = avpriv_find_start_code(buf + buf_index, buf + next_avc + 1, &state) - buf - 1;
829
830 return FFMIN(buf_index, buf_size);
831 }
832
833 int ff_h264_field_end(H264Context *h, H264SliceContext *sl, int in_setup);
834
835 int ff_h264_ref_picture(H264Context *h, H264Picture *dst, H264Picture *src);
836 void ff_h264_unref_picture(H264Context *h, H264Picture *pic);
837
838 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl);
839
840 void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl, int y, int height);
841
842 /**
843 * Submit a slice for decoding.
844 *
845 * Parse the slice header, starting a new field/frame if necessary. If any
846 * slices are queued for the previous field, they are decoded.
847 */
848 int ff_h264_queue_decode_slice(H264Context *h, const H2645NAL *nal);
849 int ff_h264_execute_decode_slices(H264Context *h);
850 int ff_h264_update_thread_context(AVCodecContext *dst,
851 const AVCodecContext *src);
852
853 void ff_h264_flush_change(H264Context *h);
854
855 void ff_h264_free_tables(H264Context *h);
856
857 void ff_h264_set_erpic(ERPicture *dst, H264Picture *src);
858
859 #endif /* AVCODEC_H264DEC_H */
860