blob: e994f7e7fe88d744c41981ada262d667b58410d2
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/thread.h" |
34 | |
35 | #include "cabac.h" |
36 | #include "error_resilience.h" |
37 | #include "h264_parse.h" |
38 | #include "h264_ps.h" |
39 | #include "h264_sei.h" |
40 | #include "h2645_parse.h" |
41 | #include "h264chroma.h" |
42 | #include "h264dsp.h" |
43 | #include "h264pred.h" |
44 | #include "h264qpel.h" |
45 | #include "internal.h" |
46 | #include "mpegutils.h" |
47 | #include "parser.h" |
48 | #include "qpeldsp.h" |
49 | #include "rectangle.h" |
50 | #include "videodsp.h" |
51 | |
52 | #define H264_MAX_PICTURE_COUNT 36 |
53 | |
54 | #define MAX_MMCO_COUNT 66 |
55 | |
56 | #define MAX_DELAYED_PIC_COUNT 16 |
57 | |
58 | /* Compiling in interlaced support reduces the speed |
59 | * of progressive decoding by about 2%. */ |
60 | #define ALLOW_INTERLACE |
61 | |
62 | #define FMO 0 |
63 | |
64 | /** |
65 | * The maximum number of slices supported by the decoder. |
66 | * must be a power of 2 |
67 | */ |
68 | #define MAX_SLICES 32 |
69 | |
70 | #ifdef ALLOW_INTERLACE |
71 | #define MB_MBAFF(h) (h)->mb_mbaff |
72 | #define MB_FIELD(sl) (sl)->mb_field_decoding_flag |
73 | #define FRAME_MBAFF(h) (h)->mb_aff_frame |
74 | #define FIELD_PICTURE(h) ((h)->picture_structure != PICT_FRAME) |
75 | #define LEFT_MBS 2 |
76 | #define LTOP 0 |
77 | #define LBOT 1 |
78 | #define LEFT(i) (i) |
79 | #else |
80 | #define MB_MBAFF(h) 0 |
81 | #define MB_FIELD(sl) 0 |
82 | #define FRAME_MBAFF(h) 0 |
83 | #define FIELD_PICTURE(h) 0 |
84 | #undef IS_INTERLACED |
85 | #define IS_INTERLACED(mb_type) 0 |
86 | #define LEFT_MBS 1 |
87 | #define LTOP 0 |
88 | #define LBOT 0 |
89 | #define LEFT(i) 0 |
90 | #endif |
91 | #define FIELD_OR_MBAFF_PICTURE(h) (FRAME_MBAFF(h) || FIELD_PICTURE(h)) |
92 | |
93 | #ifndef CABAC |
94 | #define CABAC(h) (h)->ps.pps->cabac |
95 | #endif |
96 | |
97 | #define CHROMA(h) ((h)->ps.sps->chroma_format_idc) |
98 | #define CHROMA422(h) ((h)->ps.sps->chroma_format_idc == 2) |
99 | #define CHROMA444(h) ((h)->ps.sps->chroma_format_idc == 3) |
100 | |
101 | #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit |
102 | #define MB_TYPE_8x8DCT 0x01000000 |
103 | #define IS_REF0(a) ((a) & MB_TYPE_REF0) |
104 | #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT) |
105 | |
106 | /** |
107 | * Memory management control operation opcode. |
108 | */ |
109 | typedef enum MMCOOpcode { |
110 | MMCO_END = 0, |
111 | MMCO_SHORT2UNUSED, |
112 | MMCO_LONG2UNUSED, |
113 | MMCO_SHORT2LONG, |
114 | MMCO_SET_MAX_LONG, |
115 | MMCO_RESET, |
116 | MMCO_LONG, |
117 | } MMCOOpcode; |
118 | |
119 | /** |
120 | * Memory management control operation. |
121 | */ |
122 | typedef struct MMCO { |
123 | MMCOOpcode opcode; |
124 | int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num) |
125 | int long_arg; ///< index, pic_num, or num long refs depending on opcode |
126 | } MMCO; |
127 | |
128 | typedef struct H264Picture { |
129 | AVFrame *f; |
130 | ThreadFrame tf; |
131 | |
132 | AVBufferRef *qscale_table_buf; |
133 | int8_t *qscale_table; |
134 | |
135 | AVBufferRef *motion_val_buf[2]; |
136 | int16_t (*motion_val[2])[2]; |
137 | |
138 | AVBufferRef *mb_type_buf; |
139 | uint32_t *mb_type; |
140 | |
141 | AVBufferRef *hwaccel_priv_buf; |
142 | void *hwaccel_picture_private; ///< hardware accelerator private data |
143 | |
144 | AVBufferRef *ref_index_buf[2]; |
145 | int8_t *ref_index[2]; |
146 | |
147 | int field_poc[2]; ///< top/bottom POC |
148 | int poc; ///< frame POC |
149 | int frame_num; ///< frame_num (raw frame_num from slice header) |
150 | int mmco_reset; /**< MMCO_RESET set this 1. Reordering code must |
151 | not mix pictures before and after MMCO_RESET. */ |
152 | int pic_id; /**< pic_num (short -> no wrap version of pic_num, |
153 | pic_num & max_pic_num; long -> long_pic_num) */ |
154 | int long_ref; ///< 1->long term reference 0->short term reference |
155 | int ref_poc[2][2][32]; ///< POCs of the frames/fields used as reference (FIXME need per slice) |
156 | int ref_count[2][2]; ///< number of entries in ref_poc (FIXME need per slice) |
157 | int mbaff; ///< 1 -> MBAFF frame 0-> not MBAFF |
158 | int field_picture; ///< whether or not picture was encoded in separate fields |
159 | |
160 | int reference; |
161 | int recovered; ///< picture at IDR or recovery point + recovery count |
162 | int invalid_gap; |
163 | int sei_recovery_frame_cnt; |
164 | |
165 | int crop; |
166 | int crop_left; |
167 | int crop_top; |
168 | } H264Picture; |
169 | |
170 | typedef struct H264Ref { |
171 | uint8_t *data[3]; |
172 | int linesize[3]; |
173 | |
174 | int reference; |
175 | int poc; |
176 | int pic_id; |
177 | |
178 | H264Picture *parent; |
179 | } H264Ref; |
180 | |
181 | typedef struct H264SliceContext { |
182 | struct H264Context *h264; |
183 | GetBitContext gb; |
184 | ERContext er; |
185 | |
186 | int slice_num; |
187 | int slice_type; |
188 | int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P) |
189 | int slice_type_fixed; |
190 | |
191 | int qscale; |
192 | int chroma_qp[2]; // QPc |
193 | int qp_thresh; ///< QP threshold to skip loopfilter |
194 | int last_qscale_diff; |
195 | |
196 | // deblock |
197 | int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0 |
198 | int slice_alpha_c0_offset; |
199 | int slice_beta_offset; |
200 | |
201 | H264PredWeightTable pwt; |
202 | |
203 | int prev_mb_skipped; |
204 | int next_mb_skipped; |
205 | |
206 | int chroma_pred_mode; |
207 | int intra16x16_pred_mode; |
208 | |
209 | int8_t intra4x4_pred_mode_cache[5 * 8]; |
210 | int8_t(*intra4x4_pred_mode); |
211 | |
212 | int topleft_mb_xy; |
213 | int top_mb_xy; |
214 | int topright_mb_xy; |
215 | int left_mb_xy[LEFT_MBS]; |
216 | |
217 | int topleft_type; |
218 | int top_type; |
219 | int topright_type; |
220 | int left_type[LEFT_MBS]; |
221 | |
222 | const uint8_t *left_block; |
223 | int topleft_partition; |
224 | |
225 | unsigned int topleft_samples_available; |
226 | unsigned int top_samples_available; |
227 | unsigned int topright_samples_available; |
228 | unsigned int left_samples_available; |
229 | |
230 | ptrdiff_t linesize, uvlinesize; |
231 | ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff |
232 | ptrdiff_t mb_uvlinesize; |
233 | |
234 | int mb_x, mb_y; |
235 | int mb_xy; |
236 | int resync_mb_x; |
237 | int resync_mb_y; |
238 | unsigned int first_mb_addr; |
239 | // index of the first MB of the next slice |
240 | int next_slice_idx; |
241 | int mb_skip_run; |
242 | int is_complex; |
243 | |
244 | int picture_structure; |
245 | int mb_field_decoding_flag; |
246 | int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag |
247 | |
248 | int redundant_pic_count; |
249 | |
250 | /** |
251 | * number of neighbors (top and/or left) that used 8x8 dct |
252 | */ |
253 | int neighbor_transform_size; |
254 | |
255 | int direct_spatial_mv_pred; |
256 | int col_parity; |
257 | int col_fieldoff; |
258 | |
259 | int cbp; |
260 | int top_cbp; |
261 | int left_cbp; |
262 | |
263 | int dist_scale_factor[32]; |
264 | int dist_scale_factor_field[2][32]; |
265 | int map_col_to_list0[2][16 + 32]; |
266 | int map_col_to_list0_field[2][2][16 + 32]; |
267 | |
268 | /** |
269 | * num_ref_idx_l0/1_active_minus1 + 1 |
270 | */ |
271 | unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode |
272 | unsigned int list_count; |
273 | H264Ref ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs. |
274 | * Reordered version of default_ref_list |
275 | * according to picture reordering in slice header */ |
276 | struct { |
277 | uint8_t op; |
278 | uint32_t val; |
279 | } ref_modifications[2][32]; |
280 | int nb_ref_modifications[2]; |
281 | |
282 | unsigned int pps_id; |
283 | |
284 | const uint8_t *intra_pcm_ptr; |
285 | int16_t *dc_val_base; |
286 | |
287 | uint8_t *bipred_scratchpad; |
288 | uint8_t *edge_emu_buffer; |
289 | uint8_t (*top_borders[2])[(16 * 3) * 2]; |
290 | int bipred_scratchpad_allocated; |
291 | int edge_emu_buffer_allocated; |
292 | int top_borders_allocated[2]; |
293 | |
294 | /** |
295 | * non zero coeff count cache. |
296 | * is 64 if not available. |
297 | */ |
298 | DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8]; |
299 | |
300 | /** |
301 | * Motion vector cache. |
302 | */ |
303 | DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2]; |
304 | DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8]; |
305 | DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2]; |
306 | uint8_t direct_cache[5 * 8]; |
307 | |
308 | DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4]; |
309 | |
310 | ///< as a DCT coefficient is int32_t in high depth, we need to reserve twice the space. |
311 | DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2]; |
312 | DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2]; |
313 | ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either |
314 | ///< check that i is not too large or ensure that there is some unused stuff after mb |
315 | int16_t mb_padding[256 * 2]; |
316 | |
317 | uint8_t (*mvd_table[2])[2]; |
318 | |
319 | /** |
320 | * Cabac |
321 | */ |
322 | CABACContext cabac; |
323 | uint8_t cabac_state[1024]; |
324 | int cabac_init_idc; |
325 | |
326 | MMCO mmco[MAX_MMCO_COUNT]; |
327 | int nb_mmco; |
328 | int explicit_ref_marking; |
329 | |
330 | int frame_num; |
331 | int poc_lsb; |
332 | int delta_poc_bottom; |
333 | int delta_poc[2]; |
334 | int curr_pic_num; |
335 | int max_pic_num; |
336 | } H264SliceContext; |
337 | |
338 | /** |
339 | * H264Context |
340 | */ |
341 | typedef struct H264Context { |
342 | const AVClass *class; |
343 | AVCodecContext *avctx; |
344 | VideoDSPContext vdsp; |
345 | H264DSPContext h264dsp; |
346 | H264ChromaContext h264chroma; |
347 | H264QpelContext h264qpel; |
348 | |
349 | H264Picture DPB[H264_MAX_PICTURE_COUNT]; |
350 | H264Picture *cur_pic_ptr; |
351 | H264Picture cur_pic; |
352 | H264Picture last_pic_for_ec; |
353 | |
354 | H264SliceContext *slice_ctx; |
355 | int nb_slice_ctx; |
356 | int nb_slice_ctx_queued; |
357 | |
358 | H2645Packet pkt; |
359 | |
360 | int pixel_shift; ///< 0 for 8-bit H.264, 1 for high-bit-depth H.264 |
361 | |
362 | /* coded dimensions -- 16 * mb w/h */ |
363 | int width, height; |
364 | int chroma_x_shift, chroma_y_shift; |
365 | |
366 | int droppable; |
367 | int coded_picture_number; |
368 | |
369 | int context_initialized; |
370 | int flags; |
371 | int workaround_bugs; |
372 | /* Set when slice threading is used and at least one slice uses deblocking |
373 | * mode 1 (i.e. across slice boundaries). Then we disable the loop filter |
374 | * during normal MB decoding and execute it serially at the end. |
375 | */ |
376 | int postpone_filter; |
377 | |
378 | /* |
379 | * Set to 1 when the current picture is IDR, 0 otherwise. |
380 | */ |
381 | int picture_idr; |
382 | |
383 | int8_t(*intra4x4_pred_mode); |
384 | H264PredContext hpc; |
385 | |
386 | uint8_t (*non_zero_count)[48]; |
387 | |
388 | #define LIST_NOT_USED -1 // FIXME rename? |
389 | #define PART_NOT_AVAILABLE -2 |
390 | |
391 | /** |
392 | * block_offset[ 0..23] for frame macroblocks |
393 | * block_offset[24..47] for field macroblocks |
394 | */ |
395 | int block_offset[2 * (16 * 3)]; |
396 | |
397 | uint32_t *mb2b_xy; // FIXME are these 4 a good idea? |
398 | uint32_t *mb2br_xy; |
399 | int b_stride; // FIXME use s->b4_stride |
400 | |
401 | uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1 |
402 | |
403 | // interlacing specific flags |
404 | int mb_aff_frame; |
405 | int picture_structure; |
406 | int first_field; |
407 | |
408 | uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type |
409 | |
410 | /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */ |
411 | uint16_t *cbp_table; |
412 | |
413 | /* chroma_pred_mode for i4x4 or i16x16, else 0 */ |
414 | uint8_t *chroma_pred_mode_table; |
415 | uint8_t (*mvd_table[2])[2]; |
416 | uint8_t *direct_table; |
417 | |
418 | uint8_t zigzag_scan[16]; |
419 | uint8_t zigzag_scan8x8[64]; |
420 | uint8_t zigzag_scan8x8_cavlc[64]; |
421 | uint8_t field_scan[16]; |
422 | uint8_t field_scan8x8[64]; |
423 | uint8_t field_scan8x8_cavlc[64]; |
424 | uint8_t zigzag_scan_q0[16]; |
425 | uint8_t zigzag_scan8x8_q0[64]; |
426 | uint8_t zigzag_scan8x8_cavlc_q0[64]; |
427 | uint8_t field_scan_q0[16]; |
428 | uint8_t field_scan8x8_q0[64]; |
429 | uint8_t field_scan8x8_cavlc_q0[64]; |
430 | |
431 | int mb_y; |
432 | int mb_height, mb_width; |
433 | int mb_stride; |
434 | int mb_num; |
435 | |
436 | // ============================================================= |
437 | // Things below are not used in the MB or more inner code |
438 | |
439 | int nal_ref_idc; |
440 | int nal_unit_type; |
441 | |
442 | 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() |
443 | |
444 | /** |
445 | * Used to parse AVC variant of H.264 |
446 | */ |
447 | int is_avc; ///< this flag is != 0 if codec is avc1 |
448 | int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4) |
449 | |
450 | int bit_depth_luma; ///< luma bit depth from sps to detect changes |
451 | int chroma_format_idc; ///< chroma format from sps to detect changes |
452 | |
453 | H264ParamSets ps; |
454 | |
455 | uint16_t *slice_table_base; |
456 | |
457 | H264POCContext poc; |
458 | |
459 | H264Ref default_ref[2]; |
460 | H264Picture *short_ref[32]; |
461 | H264Picture *long_ref[32]; |
462 | H264Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size? |
463 | int last_pocs[MAX_DELAYED_PIC_COUNT]; |
464 | H264Picture *next_output_pic; |
465 | int next_outputed_poc; |
466 | |
467 | /** |
468 | * memory management control operations buffer. |
469 | */ |
470 | MMCO mmco[MAX_MMCO_COUNT]; |
471 | int nb_mmco; |
472 | int mmco_reset; |
473 | int explicit_ref_marking; |
474 | |
475 | int long_ref_count; ///< number of actual long term references |
476 | int short_ref_count; ///< number of actual short term references |
477 | |
478 | /** |
479 | * @name Members for slice based multithreading |
480 | * @{ |
481 | */ |
482 | /** |
483 | * current slice number, used to initialize slice_num of each thread/context |
484 | */ |
485 | int current_slice; |
486 | |
487 | /** @} */ |
488 | |
489 | /** |
490 | * Complement sei_pic_struct |
491 | * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames. |
492 | * However, soft telecined frames may have these values. |
493 | * This is used in an attempt to flag soft telecine progressive. |
494 | */ |
495 | int prev_interlaced_frame; |
496 | |
497 | /** |
498 | * Are the SEI recovery points looking valid. |
499 | */ |
500 | int valid_recovery_point; |
501 | |
502 | /** |
503 | * recovery_frame is the frame_num at which the next frame should |
504 | * be fully constructed. |
505 | * |
506 | * Set to -1 when not expecting a recovery point. |
507 | */ |
508 | int recovery_frame; |
509 | |
510 | /** |
511 | * We have seen an IDR, so all the following frames in coded order are correctly |
512 | * decodable. |
513 | */ |
514 | #define FRAME_RECOVERED_IDR (1 << 0) |
515 | /** |
516 | * Sufficient number of frames have been decoded since a SEI recovery point, |
517 | * so all the following frames in presentation order are correct. |
518 | */ |
519 | #define FRAME_RECOVERED_SEI (1 << 1) |
520 | |
521 | int frame_recovered; ///< Initial frame has been completely recovered |
522 | |
523 | int has_recovery_point; |
524 | |
525 | int missing_fields; |
526 | |
527 | /* for frame threading, this is set to 1 |
528 | * after finish_setup() has been called, so we cannot modify |
529 | * some context properties (which are supposed to stay constant between |
530 | * slices) anymore */ |
531 | int setup_finished; |
532 | |
533 | int cur_chroma_format_idc; |
534 | int cur_bit_depth_luma; |
535 | int16_t slice_row[MAX_SLICES]; ///< to detect when MAX_SLICES is too low |
536 | |
537 | int enable_er; |
538 | |
539 | H264SEIContext sei; |
540 | |
541 | AVBufferPool *qscale_table_pool; |
542 | AVBufferPool *mb_type_pool; |
543 | AVBufferPool *motion_val_pool; |
544 | AVBufferPool *ref_index_pool; |
545 | int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1 |
546 | } H264Context; |
547 | |
548 | extern const uint16_t ff_h264_mb_sizes[4]; |
549 | |
550 | /** |
551 | * Reconstruct bitstream slice_type. |
552 | */ |
553 | int ff_h264_get_slice_type(const H264SliceContext *sl); |
554 | |
555 | /** |
556 | * Allocate tables. |
557 | * needs width/height |
558 | */ |
559 | int ff_h264_alloc_tables(H264Context *h); |
560 | |
561 | int ff_h264_decode_ref_pic_list_reordering(H264SliceContext *sl, void *logctx); |
562 | int ff_h264_build_ref_list(H264Context *h, H264SliceContext *sl); |
563 | void ff_h264_remove_all_refs(H264Context *h); |
564 | |
565 | /** |
566 | * Execute the reference picture marking (memory management control operations). |
567 | */ |
568 | int ff_h264_execute_ref_pic_marking(H264Context *h); |
569 | |
570 | int ff_h264_decode_ref_pic_marking(H264SliceContext *sl, GetBitContext *gb, |
571 | const H2645NAL *nal, void *logctx); |
572 | |
573 | void ff_h264_hl_decode_mb(const H264Context *h, H264SliceContext *sl); |
574 | int ff_h264_decode_init(AVCodecContext *avctx); |
575 | void ff_h264_decode_init_vlc(void); |
576 | |
577 | /** |
578 | * Decode a macroblock |
579 | * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error |
580 | */ |
581 | int ff_h264_decode_mb_cavlc(const H264Context *h, H264SliceContext *sl); |
582 | |
583 | /** |
584 | * Decode a CABAC coded macroblock |
585 | * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error |
586 | */ |
587 | int ff_h264_decode_mb_cabac(const H264Context *h, H264SliceContext *sl); |
588 | |
589 | void ff_h264_init_cabac_states(const H264Context *h, H264SliceContext *sl); |
590 | |
591 | void ff_h264_direct_dist_scale_factor(const H264Context *const h, H264SliceContext *sl); |
592 | void ff_h264_direct_ref_list_init(const H264Context *const h, H264SliceContext *sl); |
593 | void ff_h264_pred_direct_motion(const H264Context *const h, H264SliceContext *sl, |
594 | int *mb_type); |
595 | |
596 | void ff_h264_filter_mb_fast(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y, |
597 | uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, |
598 | unsigned int linesize, unsigned int uvlinesize); |
599 | void ff_h264_filter_mb(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y, |
600 | uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, |
601 | unsigned int linesize, unsigned int uvlinesize); |
602 | |
603 | /* |
604 | * o-o o-o |
605 | * / / / |
606 | * o-o o-o |
607 | * ,---' |
608 | * o-o o-o |
609 | * / / / |
610 | * o-o o-o |
611 | */ |
612 | |
613 | /* Scan8 organization: |
614 | * 0 1 2 3 4 5 6 7 |
615 | * 0 DY y y y y y |
616 | * 1 y Y Y Y Y |
617 | * 2 y Y Y Y Y |
618 | * 3 y Y Y Y Y |
619 | * 4 y Y Y Y Y |
620 | * 5 DU u u u u u |
621 | * 6 u U U U U |
622 | * 7 u U U U U |
623 | * 8 u U U U U |
624 | * 9 u U U U U |
625 | * 10 DV v v v v v |
626 | * 11 v V V V V |
627 | * 12 v V V V V |
628 | * 13 v V V V V |
629 | * 14 v V V V V |
630 | * DY/DU/DV are for luma/chroma DC. |
631 | */ |
632 | |
633 | #define LUMA_DC_BLOCK_INDEX 48 |
634 | #define CHROMA_DC_BLOCK_INDEX 49 |
635 | |
636 | // This table must be here because scan8[constant] must be known at compiletime |
637 | static const uint8_t scan8[16 * 3 + 3] = { |
638 | 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8, |
639 | 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8, |
640 | 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8, |
641 | 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8, |
642 | 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8, |
643 | 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8, |
644 | 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8, |
645 | 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8, |
646 | 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8, |
647 | 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8, |
648 | 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8, |
649 | 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8, |
650 | 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8 |
651 | }; |
652 | |
653 | static av_always_inline uint32_t pack16to32(unsigned a, unsigned b) |
654 | { |
655 | #if HAVE_BIGENDIAN |
656 | return (b & 0xFFFF) + (a << 16); |
657 | #else |
658 | return (a & 0xFFFF) + (b << 16); |
659 | #endif |
660 | } |
661 | |
662 | static av_always_inline uint16_t pack8to16(unsigned a, unsigned b) |
663 | { |
664 | #if HAVE_BIGENDIAN |
665 | return (b & 0xFF) + (a << 8); |
666 | #else |
667 | return (a & 0xFF) + (b << 8); |
668 | #endif |
669 | } |
670 | |
671 | /** |
672 | * Get the chroma qp. |
673 | */ |
674 | static av_always_inline int get_chroma_qp(const PPS *pps, int t, int qscale) |
675 | { |
676 | return pps->chroma_qp_table[t][qscale]; |
677 | } |
678 | |
679 | /** |
680 | * Get the predicted intra4x4 prediction mode. |
681 | */ |
682 | static av_always_inline int pred_intra_mode(const H264Context *h, |
683 | H264SliceContext *sl, int n) |
684 | { |
685 | const int index8 = scan8[n]; |
686 | const int left = sl->intra4x4_pred_mode_cache[index8 - 1]; |
687 | const int top = sl->intra4x4_pred_mode_cache[index8 - 8]; |
688 | const int min = FFMIN(left, top); |
689 | |
690 | ff_tlog(h->avctx, "mode:%d %d min:%d\n", left, top, min); |
691 | |
692 | if (min < 0) |
693 | return DC_PRED; |
694 | else |
695 | return min; |
696 | } |
697 | |
698 | static av_always_inline void write_back_intra_pred_mode(const H264Context *h, |
699 | H264SliceContext *sl) |
700 | { |
701 | int8_t *i4x4 = sl->intra4x4_pred_mode + h->mb2br_xy[sl->mb_xy]; |
702 | int8_t *i4x4_cache = sl->intra4x4_pred_mode_cache; |
703 | |
704 | AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4); |
705 | i4x4[4] = i4x4_cache[7 + 8 * 3]; |
706 | i4x4[5] = i4x4_cache[7 + 8 * 2]; |
707 | i4x4[6] = i4x4_cache[7 + 8 * 1]; |
708 | } |
709 | |
710 | static av_always_inline void write_back_non_zero_count(const H264Context *h, |
711 | H264SliceContext *sl) |
712 | { |
713 | const int mb_xy = sl->mb_xy; |
714 | uint8_t *nnz = h->non_zero_count[mb_xy]; |
715 | uint8_t *nnz_cache = sl->non_zero_count_cache; |
716 | |
717 | AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]); |
718 | AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]); |
719 | AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]); |
720 | AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]); |
721 | AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]); |
722 | AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]); |
723 | AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]); |
724 | AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]); |
725 | |
726 | if (!h->chroma_y_shift) { |
727 | AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]); |
728 | AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]); |
729 | AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]); |
730 | AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]); |
731 | } |
732 | } |
733 | |
734 | static av_always_inline void write_back_motion_list(const H264Context *h, |
735 | H264SliceContext *sl, |
736 | int b_stride, |
737 | int b_xy, int b8_xy, |
738 | int mb_type, int list) |
739 | { |
740 | int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy]; |
741 | int16_t(*mv_src)[2] = &sl->mv_cache[list][scan8[0]]; |
742 | AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0); |
743 | AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1); |
744 | AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2); |
745 | AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3); |
746 | if (CABAC(h)) { |
747 | uint8_t (*mvd_dst)[2] = &sl->mvd_table[list][FMO ? 8 * sl->mb_xy |
748 | : h->mb2br_xy[sl->mb_xy]]; |
749 | uint8_t(*mvd_src)[2] = &sl->mvd_cache[list][scan8[0]]; |
750 | if (IS_SKIP(mb_type)) { |
751 | AV_ZERO128(mvd_dst); |
752 | } else { |
753 | AV_COPY64(mvd_dst, mvd_src + 8 * 3); |
754 | AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0); |
755 | AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1); |
756 | AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2); |
757 | } |
758 | } |
759 | |
760 | { |
761 | int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy]; |
762 | int8_t *ref_cache = sl->ref_cache[list]; |
763 | ref_index[0 + 0 * 2] = ref_cache[scan8[0]]; |
764 | ref_index[1 + 0 * 2] = ref_cache[scan8[4]]; |
765 | ref_index[0 + 1 * 2] = ref_cache[scan8[8]]; |
766 | ref_index[1 + 1 * 2] = ref_cache[scan8[12]]; |
767 | } |
768 | } |
769 | |
770 | static av_always_inline void write_back_motion(const H264Context *h, |
771 | H264SliceContext *sl, |
772 | int mb_type) |
773 | { |
774 | const int b_stride = h->b_stride; |
775 | const int b_xy = 4 * sl->mb_x + 4 * sl->mb_y * h->b_stride; // try mb2b(8)_xy |
776 | const int b8_xy = 4 * sl->mb_xy; |
777 | |
778 | if (USES_LIST(mb_type, 0)) { |
779 | write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 0); |
780 | } else { |
781 | fill_rectangle(&h->cur_pic.ref_index[0][b8_xy], |
782 | 2, 2, 2, (uint8_t)LIST_NOT_USED, 1); |
783 | } |
784 | if (USES_LIST(mb_type, 1)) |
785 | write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 1); |
786 | |
787 | if (sl->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) { |
788 | if (IS_8X8(mb_type)) { |
789 | uint8_t *direct_table = &h->direct_table[4 * sl->mb_xy]; |
790 | direct_table[1] = sl->sub_mb_type[1] >> 1; |
791 | direct_table[2] = sl->sub_mb_type[2] >> 1; |
792 | direct_table[3] = sl->sub_mb_type[3] >> 1; |
793 | } |
794 | } |
795 | } |
796 | |
797 | static av_always_inline int get_dct8x8_allowed(const H264Context *h, H264SliceContext *sl) |
798 | { |
799 | if (h->ps.sps->direct_8x8_inference_flag) |
800 | return !(AV_RN64A(sl->sub_mb_type) & |
801 | ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) * |
802 | 0x0001000100010001ULL)); |
803 | else |
804 | return !(AV_RN64A(sl->sub_mb_type) & |
805 | ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) * |
806 | 0x0001000100010001ULL)); |
807 | } |
808 | |
809 | static inline int find_start_code(const uint8_t *buf, int buf_size, |
810 | int buf_index, int next_avc) |
811 | { |
812 | uint32_t state = -1; |
813 | |
814 | buf_index = avpriv_find_start_code(buf + buf_index, buf + next_avc + 1, &state) - buf - 1; |
815 | |
816 | return FFMIN(buf_index, buf_size); |
817 | } |
818 | |
819 | int ff_h264_field_end(H264Context *h, H264SliceContext *sl, int in_setup); |
820 | |
821 | int ff_h264_ref_picture(H264Context *h, H264Picture *dst, H264Picture *src); |
822 | void ff_h264_unref_picture(H264Context *h, H264Picture *pic); |
823 | |
824 | int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl); |
825 | |
826 | void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl, int y, int height); |
827 | |
828 | int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, |
829 | const H2645NAL *nal); |
830 | /** |
831 | * Submit a slice for decoding. |
832 | * |
833 | * Parse the slice header, starting a new field/frame if necessary. If any |
834 | * slices are queued for the previous field, they are decoded. |
835 | */ |
836 | int ff_h264_queue_decode_slice(H264Context *h, const H2645NAL *nal); |
837 | int ff_h264_execute_decode_slices(H264Context *h); |
838 | int ff_h264_update_thread_context(AVCodecContext *dst, |
839 | const AVCodecContext *src); |
840 | |
841 | void ff_h264_flush_change(H264Context *h); |
842 | |
843 | void ff_h264_free_tables(H264Context *h); |
844 | |
845 | void ff_h264_set_erpic(ERPicture *dst, H264Picture *src); |
846 | |
847 | #endif /* AVCODEC_H264DEC_H */ |
848 |