1 /*
2 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3 *
4 * This source code is subject to the terms of the BSD 2 Clause License and
5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6 * was not distributed with this source code in the LICENSE file, you can
7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8 * Media Patent License 1.0 was not distributed with this source code in the
9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10 */
11
12 /*!\file
13 * \brief Declares top-level encoder structures and functions.
14 */
15 #ifndef AOM_AV1_ENCODER_ENCODER_H_
16 #define AOM_AV1_ENCODER_ENCODER_H_
17
18 #include <stdbool.h>
19 #include <stdio.h>
20
21 #include "config/aom_config.h"
22
23 #include "aom/aomcx.h"
24
25 #include "av1/common/alloccommon.h"
26 #include "av1/common/av1_common_int.h"
27 #include "av1/common/blockd.h"
28 #include "av1/common/entropymode.h"
29 #include "av1/common/enums.h"
30 #include "av1/common/reconintra.h"
31 #include "av1/common/resize.h"
32 #include "av1/common/thread_common.h"
33 #include "av1/common/timing.h"
34
35 #include "av1/encoder/aq_cyclicrefresh.h"
36 #include "av1/encoder/av1_quantize.h"
37 #include "av1/encoder/block.h"
38 #include "av1/encoder/context_tree.h"
39 #include "av1/encoder/encodemb.h"
40 #include "av1/encoder/external_partition.h"
41 #include "av1/encoder/firstpass.h"
42 #include "av1/encoder/global_motion.h"
43 #include "av1/encoder/level.h"
44 #include "av1/encoder/lookahead.h"
45 #include "av1/encoder/mcomp.h"
46 #include "av1/encoder/pickcdef.h"
47 #include "av1/encoder/ratectrl.h"
48 #include "av1/encoder/rd.h"
49 #include "av1/encoder/speed_features.h"
50 #include "av1/encoder/svc_layercontext.h"
51 #include "av1/encoder/temporal_filter.h"
52 #include "av1/encoder/thirdpass.h"
53 #include "av1/encoder/tokenize.h"
54 #include "av1/encoder/tpl_model.h"
55 #include "av1/encoder/av1_noise_estimate.h"
56 #include "av1/encoder/bitstream.h"
57
58 #if CONFIG_INTERNAL_STATS
59 #include "aom_dsp/ssim.h"
60 #endif
61 #include "aom_dsp/variance.h"
62 #if CONFIG_DENOISE
63 #include "aom_dsp/noise_model.h"
64 #endif
65 #if CONFIG_TUNE_VMAF
66 #include "av1/encoder/tune_vmaf.h"
67 #endif
68 #if CONFIG_AV1_TEMPORAL_DENOISING
69 #include "av1/encoder/av1_temporal_denoiser.h"
70 #endif
71 #if CONFIG_TUNE_BUTTERAUGLI
72 #include "av1/encoder/tune_butteraugli.h"
73 #endif
74
75 #include "aom/internal/aom_codec_internal.h"
76 #include "aom_util/aom_thread.h"
77
78 #ifdef __cplusplus
79 extern "C" {
80 #endif
81
82 // TODO(yunqing, any): Added suppression tag to quiet Doxygen warnings. Need to
83 // adjust it while we work on documentation.
84 /*!\cond */
85 // Number of frames required to test for scene cut detection
86 #define SCENE_CUT_KEY_TEST_INTERVAL 16
87
88 // Lookahead index threshold to enable temporal filtering for second arf.
89 #define TF_LOOKAHEAD_IDX_THR 7
90
91 #define HDR_QP_LEVELS 10
92 #define CHROMA_CB_QP_SCALE 1.04
93 #define CHROMA_CR_QP_SCALE 1.04
94 #define CHROMA_QP_SCALE -0.46
95 #define CHROMA_QP_OFFSET 9.26
96 #define QP_SCALE_FACTOR 2.0
97 #define DISABLE_HDR_LUMA_DELTAQ 1
98
99 // Rational number with an int64 numerator
100 // This structure holds a fractional value
101 typedef struct aom_rational64 {
102 int64_t num; // fraction numerator
103 int den; // fraction denominator
104 } aom_rational64_t; // alias for struct aom_rational
105
106 enum {
107 // Good Quality Fast Encoding. The encoder balances quality with the amount of
108 // time it takes to encode the output. Speed setting controls how fast.
109 GOOD,
110 // Realtime Fast Encoding. Will force some restrictions on bitrate
111 // constraints.
112 REALTIME,
113 // All intra mode. All the frames are coded as intra frames.
114 ALLINTRA
115 } UENUM1BYTE(MODE);
116
117 enum {
118 FRAMEFLAGS_KEY = 1 << 0,
119 FRAMEFLAGS_GOLDEN = 1 << 1,
120 FRAMEFLAGS_BWDREF = 1 << 2,
121 // TODO(zoeliu): To determine whether a frame flag is needed for ALTREF2_FRAME
122 FRAMEFLAGS_ALTREF = 1 << 3,
123 FRAMEFLAGS_INTRAONLY = 1 << 4,
124 FRAMEFLAGS_SWITCH = 1 << 5,
125 FRAMEFLAGS_ERROR_RESILIENT = 1 << 6,
126 } UENUM1BYTE(FRAMETYPE_FLAGS);
127
128 #if CONFIG_FPMT_TEST
129 enum {
130 PARALLEL_ENCODE = 0,
131 PARALLEL_SIMULATION_ENCODE,
132 NUM_FPMT_TEST_ENCODES
133 } UENUM1BYTE(FPMT_TEST_ENC_CFG);
134 #endif // CONFIG_FPMT_TEST
135 // 0 level frames are sometimes used for rate control purposes, but for
136 // reference mapping purposes, the minimum level should be 1.
137 #define MIN_PYR_LEVEL 1
get_true_pyr_level(int frame_level,int frame_order,int max_layer_depth)138 static INLINE int get_true_pyr_level(int frame_level, int frame_order,
139 int max_layer_depth) {
140 if (frame_order == 0) {
141 // Keyframe case
142 return MIN_PYR_LEVEL;
143 } else if (frame_level == MAX_ARF_LAYERS) {
144 // Leaves
145 return max_layer_depth;
146 } else if (frame_level == (MAX_ARF_LAYERS + 1)) {
147 // Altrefs
148 return MIN_PYR_LEVEL;
149 }
150 return AOMMAX(MIN_PYR_LEVEL, frame_level);
151 }
152
153 enum {
154 NO_AQ = 0,
155 VARIANCE_AQ = 1,
156 COMPLEXITY_AQ = 2,
157 CYCLIC_REFRESH_AQ = 3,
158 AQ_MODE_COUNT // This should always be the last member of the enum
159 } UENUM1BYTE(AQ_MODE);
160 enum {
161 NO_DELTA_Q = 0,
162 DELTA_Q_OBJECTIVE = 1, // Modulation to improve objective quality
163 DELTA_Q_PERCEPTUAL = 2, // Modulation to improve video perceptual quality
164 DELTA_Q_PERCEPTUAL_AI = 3, // Perceptual quality opt for all intra mode
165 DELTA_Q_USER_RATING_BASED = 4, // User rating based delta q mode
166 DELTA_Q_HDR = 5, // QP adjustment based on HDR block pixel average
167 DELTA_Q_MODE_COUNT // This should always be the last member of the enum
168 } UENUM1BYTE(DELTAQ_MODE);
169
170 enum {
171 RESIZE_NONE = 0, // No frame resizing allowed.
172 RESIZE_FIXED = 1, // All frames are coded at the specified scale.
173 RESIZE_RANDOM = 2, // All frames are coded at a random scale.
174 RESIZE_DYNAMIC = 3, // Frames coded at lower scale based on rate control.
175 RESIZE_MODES
176 } UENUM1BYTE(RESIZE_MODE);
177
178 enum {
179 SS_CFG_SRC = 0,
180 SS_CFG_LOOKAHEAD = 1,
181 SS_CFG_FPF = 2,
182 SS_CFG_TOTAL = 3
183 } UENUM1BYTE(SS_CFG_OFFSET);
184
185 enum {
186 DISABLE_SCENECUT, // For LAP, lag_in_frames < 19
187 ENABLE_SCENECUT_MODE_1, // For LAP, lag_in_frames >=19 and < 33
188 ENABLE_SCENECUT_MODE_2 // For twopass and LAP - lag_in_frames >=33
189 } UENUM1BYTE(SCENECUT_MODE);
190
191 #define MAX_VBR_CORPUS_COMPLEXITY 10000
192
193 typedef enum {
194 MOD_FP, // First pass
195 MOD_TF, // Temporal filtering
196 MOD_TPL, // TPL
197 MOD_GME, // Global motion estimation
198 MOD_ENC, // Encode stage
199 MOD_LPF, // Deblocking loop filter
200 MOD_CDEF_SEARCH, // CDEF search
201 MOD_CDEF, // CDEF frame
202 MOD_LR, // Loop restoration filtering
203 MOD_PACK_BS, // Pack bitstream
204 MOD_FRAME_ENC, // Frame Parallel encode
205 MOD_AI, // All intra
206 NUM_MT_MODULES
207 } MULTI_THREADED_MODULES;
208
209 /*!\endcond */
210
211 /*!\enum COST_UPDATE_TYPE
212 * \brief This enum controls how often the entropy costs should be updated.
213 * \warning In case of any modifications/additions done to the enum
214 * COST_UPDATE_TYPE, the enum INTERNAL_COST_UPDATE_TYPE needs to be updated as
215 * well.
216 */
217 typedef enum {
218 COST_UPD_SB, /*!< Update every sb. */
219 COST_UPD_SBROW, /*!< Update every sb rows inside a tile. */
220 COST_UPD_TILE, /*!< Update every tile. */
221 COST_UPD_OFF, /*!< Turn off cost updates. */
222 NUM_COST_UPDATE_TYPES, /*!< Number of cost update types. */
223 } COST_UPDATE_TYPE;
224
225 /*!\enum LOOPFILTER_CONTROL
226 * \brief This enum controls to which frames loopfilter is applied.
227 */
228 typedef enum {
229 LOOPFILTER_NONE = 0, /*!< Disable loopfilter on all frames. */
230 LOOPFILTER_ALL = 1, /*!< Enable loopfilter for all frames. */
231 LOOPFILTER_REFERENCE = 2, /*!< Disable loopfilter on non reference frames. */
232 LOOPFILTER_SELECTIVELY =
233 3, /*!< Disable loopfilter on frames with low motion. */
234 } LOOPFILTER_CONTROL;
235
236 /*!\enum SKIP_APPLY_POSTPROC_FILTER
237 * \brief This enum controls the application of post-processing filters on a
238 * reconstructed frame.
239 */
240 typedef enum {
241 SKIP_APPLY_RESTORATION = 1 << 0,
242 SKIP_APPLY_SUPERRES = 1 << 1,
243 SKIP_APPLY_CDEF = 1 << 2,
244 SKIP_APPLY_LOOPFILTER = 1 << 3,
245 } SKIP_APPLY_POSTPROC_FILTER;
246
247 /*!
248 * \brief Encoder config related to resize.
249 */
250 typedef struct {
251 /*!
252 * Indicates the frame resize mode to be used by the encoder.
253 */
254 RESIZE_MODE resize_mode;
255 /*!
256 * Indicates the denominator for resize of inter frames, assuming 8 as the
257 * numerator. Its value ranges between 8-16.
258 */
259 uint8_t resize_scale_denominator;
260 /*!
261 * Indicates the denominator for resize of key frames, assuming 8 as the
262 * numerator. Its value ranges between 8-16.
263 */
264 uint8_t resize_kf_scale_denominator;
265 } ResizeCfg;
266
267 /*!
268 * \brief Encoder config for coding block partitioning.
269 */
270 typedef struct {
271 /*!
272 * Flag to indicate if rectanguar partitions should be enabled.
273 */
274 bool enable_rect_partitions;
275 /*!
276 * Flag to indicate if AB partitions should be enabled.
277 */
278 bool enable_ab_partitions;
279 /*!
280 * Flag to indicate if 1:4 / 4:1 partitions should be enabled.
281 */
282 bool enable_1to4_partitions;
283 /*!
284 * Indicates the minimum partition size that should be allowed. Both width and
285 * height of a partition cannot be smaller than the min_partition_size.
286 */
287 BLOCK_SIZE min_partition_size;
288 /*!
289 * Indicates the maximum partition size that should be allowed. Both width and
290 * height of a partition cannot be larger than the max_partition_size.
291 */
292 BLOCK_SIZE max_partition_size;
293 } PartitionCfg;
294
295 /*!
296 * \brief Encoder flags for intra prediction.
297 */
298 typedef struct {
299 /*!
300 * Flag to indicate if intra edge filtering process should be enabled.
301 */
302 bool enable_intra_edge_filter;
303 /*!
304 * Flag to indicate if recursive filtering based intra prediction should be
305 * enabled.
306 */
307 bool enable_filter_intra;
308 /*!
309 * Flag to indicate if smooth intra prediction modes should be enabled.
310 */
311 bool enable_smooth_intra;
312 /*!
313 * Flag to indicate if PAETH intra prediction mode should be enabled.
314 */
315 bool enable_paeth_intra;
316 /*!
317 * Flag to indicate if CFL uv intra mode should be enabled.
318 */
319 bool enable_cfl_intra;
320 /*!
321 * Flag to indicate if directional modes should be enabled.
322 */
323 bool enable_directional_intra;
324 /*!
325 * Flag to indicate if the subset of directional modes from D45 to D203 intra
326 * should be enabled. Has no effect if directional modes are disabled.
327 */
328 bool enable_diagonal_intra;
329 /*!
330 * Flag to indicate if delta angles for directional intra prediction should be
331 * enabled.
332 */
333 bool enable_angle_delta;
334 /*!
335 * Flag to indicate whether to automatically turn off several intral coding
336 * tools.
337 * This flag is only used when "--deltaq-mode=3" is true.
338 * When set to 1, the encoder will analyze the reconstruction quality
339 * as compared to the source image in the preprocessing pass.
340 * If the recontruction quality is considered high enough, we disable
341 * the following intra coding tools, for better encoding speed:
342 * "--enable_smooth_intra",
343 * "--enable_paeth_intra",
344 * "--enable_cfl_intra",
345 * "--enable_diagonal_intra".
346 */
347 bool auto_intra_tools_off;
348 } IntraModeCfg;
349
350 /*!
351 * \brief Encoder flags for transform sizes and types.
352 */
353 typedef struct {
354 /*!
355 * Flag to indicate if 64-pt transform should be enabled.
356 */
357 bool enable_tx64;
358 /*!
359 * Flag to indicate if flip and identity transform types should be enabled.
360 */
361 bool enable_flip_idtx;
362 /*!
363 * Flag to indicate if rectangular transform should be enabled.
364 */
365 bool enable_rect_tx;
366 /*!
367 * Flag to indicate whether or not to use a default reduced set for ext-tx
368 * rather than the potential full set of 16 transforms.
369 */
370 bool reduced_tx_type_set;
371 /*!
372 * Flag to indicate if transform type for intra blocks should be limited to
373 * DCT_DCT.
374 */
375 bool use_intra_dct_only;
376 /*!
377 * Flag to indicate if transform type for inter blocks should be limited to
378 * DCT_DCT.
379 */
380 bool use_inter_dct_only;
381 /*!
382 * Flag to indicate if intra blocks should use default transform type
383 * (mode-dependent) only.
384 */
385 bool use_intra_default_tx_only;
386 /*!
387 * Flag to indicate if transform size search should be enabled.
388 */
389 bool enable_tx_size_search;
390 } TxfmSizeTypeCfg;
391
392 /*!
393 * \brief Encoder flags for compound prediction modes.
394 */
395 typedef struct {
396 /*!
397 * Flag to indicate if distance-weighted compound type should be enabled.
398 */
399 bool enable_dist_wtd_comp;
400 /*!
401 * Flag to indicate if masked (wedge/diff-wtd) compound type should be
402 * enabled.
403 */
404 bool enable_masked_comp;
405 /*!
406 * Flag to indicate if smooth interintra mode should be enabled.
407 */
408 bool enable_smooth_interintra;
409 /*!
410 * Flag to indicate if difference-weighted compound type should be enabled.
411 */
412 bool enable_diff_wtd_comp;
413 /*!
414 * Flag to indicate if inter-inter wedge compound type should be enabled.
415 */
416 bool enable_interinter_wedge;
417 /*!
418 * Flag to indicate if inter-intra wedge compound type should be enabled.
419 */
420 bool enable_interintra_wedge;
421 } CompoundTypeCfg;
422
423 /*!
424 * \brief Encoder config related to frame super-resolution.
425 */
426 typedef struct {
427 /*!
428 * Indicates the qindex based threshold to be used when AOM_SUPERRES_QTHRESH
429 * mode is used for inter frames.
430 */
431 int superres_qthresh;
432 /*!
433 * Indicates the qindex based threshold to be used when AOM_SUPERRES_QTHRESH
434 * mode is used for key frames.
435 */
436 int superres_kf_qthresh;
437 /*!
438 * Indicates the denominator of the fraction that specifies the ratio between
439 * the superblock width before and after upscaling for inter frames. The
440 * numerator of this fraction is equal to the constant SCALE_NUMERATOR.
441 */
442 uint8_t superres_scale_denominator;
443 /*!
444 * Indicates the denominator of the fraction that specifies the ratio between
445 * the superblock width before and after upscaling for key frames. The
446 * numerator of this fraction is equal to the constant SCALE_NUMERATOR.
447 */
448 uint8_t superres_kf_scale_denominator;
449 /*!
450 * Indicates the Super-resolution mode to be used by the encoder.
451 */
452 aom_superres_mode superres_mode;
453 /*!
454 * Flag to indicate if super-resolution should be enabled for the sequence.
455 */
456 bool enable_superres;
457 } SuperResCfg;
458
459 /*!
460 * \brief Encoder config related to the coding of key frames.
461 */
462 typedef struct {
463 /*!
464 * Indicates the minimum distance to a key frame.
465 */
466 int key_freq_min;
467
468 /*!
469 * Indicates the maximum distance to a key frame.
470 */
471 int key_freq_max;
472
473 /*!
474 * Indicates if temporal filtering should be applied on keyframe.
475 */
476 int enable_keyframe_filtering;
477
478 /*!
479 * Indicates the number of frames after which a frame may be coded as an
480 * S-Frame.
481 */
482 int sframe_dist;
483
484 /*!
485 * Indicates how an S-Frame should be inserted.
486 * 1: the considered frame will be made into an S-Frame only if it is an
487 * altref frame. 2: the next altref frame will be made into an S-Frame.
488 */
489 int sframe_mode;
490
491 /*!
492 * Indicates if encoder should autodetect cut scenes and set the keyframes.
493 */
494 bool auto_key;
495
496 /*!
497 * Indicates the forward key frame distance.
498 */
499 int fwd_kf_dist;
500
501 /*!
502 * Indicates if forward keyframe reference should be enabled.
503 */
504 bool fwd_kf_enabled;
505
506 /*!
507 * Indicates if S-Frames should be enabled for the sequence.
508 */
509 bool enable_sframe;
510
511 /*!
512 * Indicates if intra block copy prediction mode should be enabled or not.
513 */
514 bool enable_intrabc;
515 } KeyFrameCfg;
516
517 /*!
518 * \brief Encoder rate control configuration parameters
519 */
520 typedef struct {
521 /*!\cond */
522 // BUFFERING PARAMETERS
523 /*!\endcond */
524 /*!
525 * Indicates the amount of data that will be buffered by the decoding
526 * application prior to beginning playback, and is expressed in units of
527 * time(milliseconds).
528 */
529 int64_t starting_buffer_level_ms;
530 /*!
531 * Indicates the amount of data that the encoder should try to maintain in the
532 * decoder's buffer, and is expressed in units of time(milliseconds).
533 */
534 int64_t optimal_buffer_level_ms;
535 /*!
536 * Indicates the maximum amount of data that may be buffered by the decoding
537 * application, and is expressed in units of time(milliseconds).
538 */
539 int64_t maximum_buffer_size_ms;
540
541 /*!
542 * Indicates the bandwidth to be used in bits per second.
543 */
544 int64_t target_bandwidth;
545
546 /*!
547 * Indicates average complexity of the corpus in single pass vbr based on
548 * LAP. 0 indicates that corpus complexity vbr mode is disabled.
549 */
550 unsigned int vbr_corpus_complexity_lap;
551 /*!
552 * Indicates the maximum allowed bitrate for any intra frame as % of bitrate
553 * target.
554 */
555 unsigned int max_intra_bitrate_pct;
556 /*!
557 * Indicates the maximum allowed bitrate for any inter frame as % of bitrate
558 * target.
559 */
560 unsigned int max_inter_bitrate_pct;
561 /*!
562 * Indicates the percentage of rate boost for golden frame in CBR mode.
563 */
564 unsigned int gf_cbr_boost_pct;
565 /*!
566 * min_cr / 100 indicates the target minimum compression ratio for each
567 * frame.
568 */
569 unsigned int min_cr;
570 /*!
571 * Indicates the frame drop threshold.
572 */
573 int drop_frames_water_mark;
574 /*!
575 * under_shoot_pct indicates the tolerance of the VBR algorithm to
576 * undershoot and is used as a trigger threshold for more aggressive
577 * adaptation of Q. It's value can range from 0-100.
578 */
579 int under_shoot_pct;
580 /*!
581 * over_shoot_pct indicates the tolerance of the VBR algorithm to overshoot
582 * and is used as a trigger threshold for more aggressive adaptation of Q.
583 * It's value can range from 0-1000.
584 */
585 int over_shoot_pct;
586 /*!
587 * Indicates the maximum qindex that can be used by the quantizer i.e. the
588 * worst quality qindex.
589 */
590 int worst_allowed_q;
591 /*!
592 * Indicates the minimum qindex that can be used by the quantizer i.e. the
593 * best quality qindex.
594 */
595 int best_allowed_q;
596 /*!
597 * Indicates the Constant/Constrained Quality level.
598 */
599 int cq_level;
600 /*!
601 * Indicates if the encoding mode is vbr, cbr, constrained quality or
602 * constant quality.
603 */
604 enum aom_rc_mode mode;
605 /*!
606 * Indicates the bias (expressed on a scale of 0 to 100) for determining
607 * target size for the current frame. The value 0 indicates the optimal CBR
608 * mode value should be used, and 100 indicates the optimal VBR mode value
609 * should be used.
610 */
611 int vbrbias;
612 /*!
613 * Indicates the minimum bitrate to be used for a single frame as a percentage
614 * of the target bitrate.
615 */
616 int vbrmin_section;
617 /*!
618 * Indicates the maximum bitrate to be used for a single frame as a percentage
619 * of the target bitrate.
620 */
621 int vbrmax_section;
622 } RateControlCfg;
623
624 /*!\cond */
625 typedef struct {
626 // Indicates the number of frames lag before encoding is started.
627 int lag_in_frames;
628 // Indicates the minimum gf/arf interval to be used.
629 int min_gf_interval;
630 // Indicates the maximum gf/arf interval to be used.
631 int max_gf_interval;
632 // Indicates the minimum height for GF group pyramid structure to be used.
633 int gf_min_pyr_height;
634 // Indicates the maximum height for GF group pyramid structure to be used.
635 int gf_max_pyr_height;
636 // Indicates if automatic set and use of altref frames should be enabled.
637 bool enable_auto_arf;
638 // Indicates if automatic set and use of (b)ackward (r)ef (f)rames should be
639 // enabled.
640 bool enable_auto_brf;
641 } GFConfig;
642
643 typedef struct {
644 // Indicates the number of tile groups.
645 unsigned int num_tile_groups;
646 // Indicates the MTU size for a tile group. If mtu is non-zero,
647 // num_tile_groups is set to DEFAULT_MAX_NUM_TG.
648 unsigned int mtu;
649 // Indicates the number of tile columns in log2.
650 int tile_columns;
651 // Indicates the number of tile rows in log2.
652 int tile_rows;
653 // Indicates the number of widths in the tile_widths[] array.
654 int tile_width_count;
655 // Indicates the number of heights in the tile_heights[] array.
656 int tile_height_count;
657 // Indicates the tile widths, and may be empty.
658 int tile_widths[MAX_TILE_COLS];
659 // Indicates the tile heights, and may be empty.
660 int tile_heights[MAX_TILE_ROWS];
661 // Indicates if large scale tile coding should be used.
662 bool enable_large_scale_tile;
663 // Indicates if single tile decoding mode should be enabled.
664 bool enable_single_tile_decoding;
665 // Indicates if EXT_TILE_DEBUG should be enabled.
666 bool enable_ext_tile_debug;
667 } TileConfig;
668
669 typedef struct {
670 // Indicates the width of the input frame.
671 int width;
672 // Indicates the height of the input frame.
673 int height;
674 // If forced_max_frame_width is non-zero then it is used to force the maximum
675 // frame width written in write_sequence_header().
676 int forced_max_frame_width;
677 // If forced_max_frame_width is non-zero then it is used to force the maximum
678 // frame height written in write_sequence_header().
679 int forced_max_frame_height;
680 // Indicates the frame width after applying both super-resolution and resize
681 // to the coded frame.
682 int render_width;
683 // Indicates the frame height after applying both super-resolution and resize
684 // to the coded frame.
685 int render_height;
686 } FrameDimensionCfg;
687
688 typedef struct {
689 // Indicates if warped motion should be enabled.
690 bool enable_warped_motion;
691 // Indicates if warped motion should be evaluated or not.
692 bool allow_warped_motion;
693 // Indicates if OBMC motion should be enabled.
694 bool enable_obmc;
695 } MotionModeCfg;
696
697 typedef struct {
698 // Timing info for each frame.
699 aom_timing_info_t timing_info;
700 // Indicates the number of time units of a decoding clock.
701 uint32_t num_units_in_decoding_tick;
702 // Indicates if decoder model information is present in the coded sequence
703 // header.
704 bool decoder_model_info_present_flag;
705 // Indicates if display model information is present in the coded sequence
706 // header.
707 bool display_model_info_present_flag;
708 // Indicates if timing info for each frame is present.
709 bool timing_info_present;
710 } DecoderModelCfg;
711
712 typedef struct {
713 // Indicates the update frequency for coeff costs.
714 COST_UPDATE_TYPE coeff;
715 // Indicates the update frequency for mode costs.
716 COST_UPDATE_TYPE mode;
717 // Indicates the update frequency for mv costs.
718 COST_UPDATE_TYPE mv;
719 // Indicates the update frequency for dv costs.
720 COST_UPDATE_TYPE dv;
721 } CostUpdateFreq;
722
723 typedef struct {
724 // Indicates the maximum number of reference frames allowed per frame.
725 unsigned int max_reference_frames;
726 // Indicates if the reduced set of references should be enabled.
727 bool enable_reduced_reference_set;
728 // Indicates if one-sided compound should be enabled.
729 bool enable_onesided_comp;
730 } RefFrameCfg;
731
732 typedef struct {
733 // Indicates the color space that should be used.
734 aom_color_primaries_t color_primaries;
735 // Indicates the characteristics of transfer function to be used.
736 aom_transfer_characteristics_t transfer_characteristics;
737 // Indicates the matrix coefficients to be used for the transfer function.
738 aom_matrix_coefficients_t matrix_coefficients;
739 // Indicates the chroma 4:2:0 sample position info.
740 aom_chroma_sample_position_t chroma_sample_position;
741 // Indicates if a limited color range or full color range should be used.
742 aom_color_range_t color_range;
743 } ColorCfg;
744
745 typedef struct {
746 // Indicates if extreme motion vector unit test should be enabled or not.
747 unsigned int motion_vector_unit_test;
748 // Indicates if superblock multipass unit test should be enabled or not.
749 unsigned int sb_multipass_unit_test;
750 } UnitTestCfg;
751
752 typedef struct {
753 // Indicates the file path to the VMAF model.
754 const char *vmaf_model_path;
755 // Indicates the path to the film grain parameters.
756 const char *film_grain_table_filename;
757 // Indicates the visual tuning metric.
758 aom_tune_metric tuning;
759 // Indicates if the current content is screen or default type.
760 aom_tune_content content;
761 // Indicates the film grain parameters.
762 int film_grain_test_vector;
763 // Indicates the in-block distortion metric to use.
764 aom_dist_metric dist_metric;
765 } TuneCfg;
766
767 typedef struct {
768 // Indicates the framerate of the input video.
769 double init_framerate;
770 // Indicates the bit-depth of the input video.
771 unsigned int input_bit_depth;
772 // Indicates the maximum number of frames to be encoded.
773 unsigned int limit;
774 // Indicates the chrome subsampling x value.
775 unsigned int chroma_subsampling_x;
776 // Indicates the chrome subsampling y value.
777 unsigned int chroma_subsampling_y;
778 } InputCfg;
779
780 typedef struct {
781 // If true, encoder will use fixed QP offsets, that are either:
782 // - Given by the user, and stored in 'fixed_qp_offsets' array, OR
783 // - Picked automatically from cq_level.
784 int use_fixed_qp_offsets;
785 // Indicates the minimum flatness of the quantization matrix.
786 int qm_minlevel;
787 // Indicates the maximum flatness of the quantization matrix.
788 int qm_maxlevel;
789 // Indicates if adaptive quantize_b should be enabled.
790 int quant_b_adapt;
791 // Indicates the Adaptive Quantization mode to be used.
792 AQ_MODE aq_mode;
793 // Indicates the delta q mode to be used.
794 DELTAQ_MODE deltaq_mode;
795 // Indicates the delta q mode strength.
796 DELTAQ_MODE deltaq_strength;
797 // Indicates if delta quantization should be enabled in chroma planes.
798 bool enable_chroma_deltaq;
799 // Indicates if delta quantization should be enabled for hdr video
800 bool enable_hdr_deltaq;
801 // Indicates if encoding with quantization matrices should be enabled.
802 bool using_qm;
803 } QuantizationCfg;
804
805 /*!\endcond */
806 /*!
807 * \brief Algorithm configuration parameters.
808 */
809 typedef struct {
810 /*!
811 * Controls the level at which rate-distortion optimization of transform
812 * coefficients favours sharpness in the block. Has no impact on RD when set
813 * to zero (default). For values 1-7, eob and skip block optimization are
814 * avoided and rdmult is adjusted in favour of block sharpness.
815 */
816 int sharpness;
817
818 /*!
819 * Indicates the trellis optimization mode of quantized coefficients.
820 * 0: disabled
821 * 1: enabled
822 * 2: enabled for rd search
823 * 3: true for estimate yrd search
824 */
825 int disable_trellis_quant;
826
827 /*!
828 * The maximum number of frames used to create an arf.
829 */
830 int arnr_max_frames;
831
832 /*!
833 * The temporal filter strength for arf used when creating ARFs.
834 */
835 int arnr_strength;
836
837 /*!
838 * Indicates the CDF update mode
839 * 0: no update
840 * 1: update on every frame(default)
841 * 2: selectively update
842 */
843 uint8_t cdf_update_mode;
844
845 /*!
846 * Indicates if RDO based on frame temporal dependency should be enabled.
847 */
848 bool enable_tpl_model;
849
850 /*!
851 * Indicates if coding of overlay frames for filtered ALTREF frames is
852 * enabled.
853 */
854 bool enable_overlay;
855
856 /*!
857 * Controls loop filtering
858 * 0: Loop filter is disabled for all frames
859 * 1: Loop filter is enabled for all frames
860 * 2: Loop filter is disabled for non-reference frames
861 * 3: Loop filter is disables for the frames with low motion
862 */
863 LOOPFILTER_CONTROL loopfilter_control;
864
865 /*!
866 * Indicates if the application of post-processing filters should be skipped
867 * on reconstructed frame.
868 */
869 bool skip_postproc_filtering;
870 } AlgoCfg;
871 /*!\cond */
872
873 typedef struct {
874 // Indicates the codec bit-depth.
875 aom_bit_depth_t bit_depth;
876 // Indicates the superblock size that should be used by the encoder.
877 aom_superblock_size_t superblock_size;
878 // Indicates if loopfilter modulation should be enabled.
879 bool enable_deltalf_mode;
880 // Indicates how CDEF should be applied.
881 CDEF_CONTROL cdef_control;
882 // Indicates if loop restoration filter should be enabled.
883 bool enable_restoration;
884 // When enabled, video mode should be used even for single frame input.
885 bool force_video_mode;
886 // Indicates if the error resiliency features should be enabled.
887 bool error_resilient_mode;
888 // Indicates if frame parallel decoding feature should be enabled.
889 bool frame_parallel_decoding_mode;
890 // Indicates if the input should be encoded as monochrome.
891 bool enable_monochrome;
892 // When enabled, the encoder will use a full header even for still pictures.
893 // When disabled, a reduced header is used for still pictures.
894 bool full_still_picture_hdr;
895 // Indicates if dual interpolation filters should be enabled.
896 bool enable_dual_filter;
897 // Indicates if frame order hint should be enabled or not.
898 bool enable_order_hint;
899 // Indicates if ref_frame_mvs should be enabled at the sequence level.
900 bool ref_frame_mvs_present;
901 // Indicates if ref_frame_mvs should be enabled at the frame level.
902 bool enable_ref_frame_mvs;
903 // Indicates if interintra compound mode is enabled.
904 bool enable_interintra_comp;
905 // Indicates if global motion should be enabled.
906 bool enable_global_motion;
907 // Indicates if palette should be enabled.
908 bool enable_palette;
909 } ToolCfg;
910
911 /*!\endcond */
912 /*!
913 * \brief Main encoder configuration data structure.
914 */
915 typedef struct AV1EncoderConfig {
916 /*!\cond */
917 // Configuration related to the input video.
918 InputCfg input_cfg;
919
920 // Configuration related to frame-dimensions.
921 FrameDimensionCfg frm_dim_cfg;
922
923 /*!\endcond */
924 /*!
925 * Encoder algorithm configuration.
926 */
927 AlgoCfg algo_cfg;
928
929 /*!
930 * Configuration related to key-frames.
931 */
932 KeyFrameCfg kf_cfg;
933
934 /*!
935 * Rate control configuration
936 */
937 RateControlCfg rc_cfg;
938 /*!\cond */
939
940 // Configuration related to Quantization.
941 QuantizationCfg q_cfg;
942
943 // Internal frame size scaling.
944 ResizeCfg resize_cfg;
945
946 // Frame Super-Resolution size scaling.
947 SuperResCfg superres_cfg;
948
949 /*!\endcond */
950 /*!
951 * stats_in buffer contains all of the stats packets produced in the first
952 * pass, concatenated.
953 */
954 aom_fixed_buf_t twopass_stats_in;
955 /*!\cond */
956
957 // Configuration related to encoder toolsets.
958 ToolCfg tool_cfg;
959
960 // Configuration related to Group of frames.
961 GFConfig gf_cfg;
962
963 // Tile related configuration parameters.
964 TileConfig tile_cfg;
965
966 // Configuration related to Tune.
967 TuneCfg tune_cfg;
968
969 // Configuration related to color.
970 ColorCfg color_cfg;
971
972 // Configuration related to decoder model.
973 DecoderModelCfg dec_model_cfg;
974
975 // Configuration related to reference frames.
976 RefFrameCfg ref_frm_cfg;
977
978 // Configuration related to unit tests.
979 UnitTestCfg unit_test_cfg;
980
981 // Flags related to motion mode.
982 MotionModeCfg motion_mode_cfg;
983
984 // Flags related to intra mode search.
985 IntraModeCfg intra_mode_cfg;
986
987 // Flags related to transform size/type.
988 TxfmSizeTypeCfg txfm_cfg;
989
990 // Flags related to compound type.
991 CompoundTypeCfg comp_type_cfg;
992
993 // Partition related information.
994 PartitionCfg part_cfg;
995
996 // Configuration related to frequency of cost update.
997 CostUpdateFreq cost_upd_freq;
998
999 #if CONFIG_DENOISE
1000 // Indicates the noise level.
1001 float noise_level;
1002 // Indicates the the denoisers block size.
1003 int noise_block_size;
1004 // Indicates whether to apply denoising to the frame to be encoded
1005 int enable_dnl_denoising;
1006 #endif
1007
1008 #if CONFIG_AV1_TEMPORAL_DENOISING
1009 // Noise sensitivity.
1010 int noise_sensitivity;
1011 #endif
1012 // Bit mask to specify which tier each of the 32 possible operating points
1013 // conforms to.
1014 unsigned int tier_mask;
1015
1016 // Indicates the number of pixels off the edge of a reference frame we're
1017 // allowed to go when forming an inter prediction.
1018 int border_in_pixels;
1019
1020 // Indicates the maximum number of threads that may be used by the encoder.
1021 int max_threads;
1022
1023 // Indicates the speed preset to be used.
1024 int speed;
1025
1026 // Indicates the target sequence level index for each operating point(OP).
1027 AV1_LEVEL target_seq_level_idx[MAX_NUM_OPERATING_POINTS];
1028
1029 // Indicates the bitstream profile to be used.
1030 BITSTREAM_PROFILE profile;
1031
1032 /*!\endcond */
1033 /*!
1034 * Indicates the current encoder pass :
1035 * AOM_RC_ONE_PASS = One pass encode,
1036 * AOM_RC_FIRST_PASS = First pass of multiple-pass
1037 * AOM_RC_SECOND_PASS = Second pass of multiple-pass
1038 * AOM_RC_THIRD_PASS = Third pass of multiple-pass
1039 */
1040 enum aom_enc_pass pass;
1041 /*!\cond */
1042
1043 // Total number of encoding passes.
1044 int passes;
1045
1046 // the name of the second pass output file when passes > 2
1047 const char *two_pass_output;
1048
1049 // the name of the second pass log file when passes > 2
1050 const char *second_pass_log;
1051
1052 // Indicates if the encoding is GOOD or REALTIME.
1053 MODE mode;
1054
1055 // Indicates if row-based multi-threading should be enabled or not.
1056 bool row_mt;
1057
1058 // Indicates if frame parallel multi-threading should be enabled or not.
1059 bool fp_mt;
1060
1061 // Indicates if 16bit frame buffers are to be used i.e., the content is >
1062 // 8-bit.
1063 bool use_highbitdepth;
1064
1065 // Indicates the bitstream syntax mode. 0 indicates bitstream is saved as
1066 // Section 5 bitstream, while 1 indicates the bitstream is saved in Annex - B
1067 // format.
1068 bool save_as_annexb;
1069
1070 // The path for partition stats reading and writing, used in the experiment
1071 // CONFIG_PARTITION_SEARCH_ORDER.
1072 const char *partition_info_path;
1073
1074 // Exit the encoder when it fails to encode to a given level.
1075 int strict_level_conformance;
1076
1077 // Max depth for the GOP after a key frame
1078 int kf_max_pyr_height;
1079
1080 // A flag to control if we enable the superblock qp sweep for a given lambda
1081 int sb_qp_sweep;
1082 /*!\endcond */
1083 } AV1EncoderConfig;
1084
1085 /*!\cond */
is_lossless_requested(const RateControlCfg * const rc_cfg)1086 static INLINE int is_lossless_requested(const RateControlCfg *const rc_cfg) {
1087 return rc_cfg->best_allowed_q == 0 && rc_cfg->worst_allowed_q == 0;
1088 }
1089 /*!\endcond */
1090
1091 /*!
1092 * \brief Encoder-side probabilities for pruning of various AV1 tools
1093 */
1094 typedef struct {
1095 /*!
1096 * obmc_probs[i][j] is the probability of OBMC being the best motion mode for
1097 * jth block size and ith frame update type, averaged over past frames. If
1098 * obmc_probs[i][j] < thresh, then OBMC search is pruned.
1099 */
1100 int obmc_probs[FRAME_UPDATE_TYPES][BLOCK_SIZES_ALL];
1101
1102 /*!
1103 * warped_probs[i] is the probability of warped motion being the best motion
1104 * mode for ith frame update type, averaged over past frames. If
1105 * warped_probs[i] < thresh, then warped motion search is pruned.
1106 */
1107 int warped_probs[FRAME_UPDATE_TYPES];
1108
1109 /*!
1110 * tx_type_probs[i][j][k] is the probability of kth tx_type being the best
1111 * for jth transform size and ith frame update type, averaged over past
1112 * frames. If tx_type_probs[i][j][k] < thresh, then transform search for that
1113 * type is pruned.
1114 */
1115 int tx_type_probs[FRAME_UPDATE_TYPES][TX_SIZES_ALL][TX_TYPES];
1116
1117 /*!
1118 * switchable_interp_probs[i][j][k] is the probability of kth interpolation
1119 * filter being the best for jth filter context and ith frame update type,
1120 * averaged over past frames. If switchable_interp_probs[i][j][k] < thresh,
1121 * then interpolation filter search is pruned for that case.
1122 */
1123 int switchable_interp_probs[FRAME_UPDATE_TYPES][SWITCHABLE_FILTER_CONTEXTS]
1124 [SWITCHABLE_FILTERS];
1125 } FrameProbInfo;
1126
1127 /*!\cond */
1128
1129 typedef struct FRAME_COUNTS {
1130 // Note: This structure should only contain 'unsigned int' fields, or
1131 // aggregates built solely from 'unsigned int' fields/elements
1132 #if CONFIG_ENTROPY_STATS
1133 unsigned int kf_y_mode[KF_MODE_CONTEXTS][KF_MODE_CONTEXTS][INTRA_MODES];
1134 unsigned int angle_delta[DIRECTIONAL_MODES][2 * MAX_ANGLE_DELTA + 1];
1135 unsigned int y_mode[BLOCK_SIZE_GROUPS][INTRA_MODES];
1136 unsigned int uv_mode[CFL_ALLOWED_TYPES][INTRA_MODES][UV_INTRA_MODES];
1137 unsigned int cfl_sign[CFL_JOINT_SIGNS];
1138 unsigned int cfl_alpha[CFL_ALPHA_CONTEXTS][CFL_ALPHABET_SIZE];
1139 unsigned int palette_y_mode[PALATTE_BSIZE_CTXS][PALETTE_Y_MODE_CONTEXTS][2];
1140 unsigned int palette_uv_mode[PALETTE_UV_MODE_CONTEXTS][2];
1141 unsigned int palette_y_size[PALATTE_BSIZE_CTXS][PALETTE_SIZES];
1142 unsigned int palette_uv_size[PALATTE_BSIZE_CTXS][PALETTE_SIZES];
1143 unsigned int palette_y_color_index[PALETTE_SIZES]
1144 [PALETTE_COLOR_INDEX_CONTEXTS]
1145 [PALETTE_COLORS];
1146 unsigned int palette_uv_color_index[PALETTE_SIZES]
1147 [PALETTE_COLOR_INDEX_CONTEXTS]
1148 [PALETTE_COLORS];
1149 unsigned int partition[PARTITION_CONTEXTS][EXT_PARTITION_TYPES];
1150 unsigned int txb_skip[TOKEN_CDF_Q_CTXS][TX_SIZES][TXB_SKIP_CONTEXTS][2];
1151 unsigned int eob_extra[TOKEN_CDF_Q_CTXS][TX_SIZES][PLANE_TYPES]
1152 [EOB_COEF_CONTEXTS][2];
1153 unsigned int dc_sign[PLANE_TYPES][DC_SIGN_CONTEXTS][2];
1154 unsigned int coeff_lps[TX_SIZES][PLANE_TYPES][BR_CDF_SIZE - 1][LEVEL_CONTEXTS]
1155 [2];
1156 unsigned int eob_flag[TX_SIZES][PLANE_TYPES][EOB_COEF_CONTEXTS][2];
1157 unsigned int eob_multi16[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][5];
1158 unsigned int eob_multi32[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][6];
1159 unsigned int eob_multi64[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][7];
1160 unsigned int eob_multi128[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][8];
1161 unsigned int eob_multi256[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][9];
1162 unsigned int eob_multi512[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][10];
1163 unsigned int eob_multi1024[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][11];
1164 unsigned int coeff_lps_multi[TOKEN_CDF_Q_CTXS][TX_SIZES][PLANE_TYPES]
1165 [LEVEL_CONTEXTS][BR_CDF_SIZE];
1166 unsigned int coeff_base_multi[TOKEN_CDF_Q_CTXS][TX_SIZES][PLANE_TYPES]
1167 [SIG_COEF_CONTEXTS][NUM_BASE_LEVELS + 2];
1168 unsigned int coeff_base_eob_multi[TOKEN_CDF_Q_CTXS][TX_SIZES][PLANE_TYPES]
1169 [SIG_COEF_CONTEXTS_EOB][NUM_BASE_LEVELS + 1];
1170 unsigned int newmv_mode[NEWMV_MODE_CONTEXTS][2];
1171 unsigned int zeromv_mode[GLOBALMV_MODE_CONTEXTS][2];
1172 unsigned int refmv_mode[REFMV_MODE_CONTEXTS][2];
1173 unsigned int drl_mode[DRL_MODE_CONTEXTS][2];
1174 unsigned int inter_compound_mode[INTER_MODE_CONTEXTS][INTER_COMPOUND_MODES];
1175 unsigned int wedge_idx[BLOCK_SIZES_ALL][16];
1176 unsigned int interintra[BLOCK_SIZE_GROUPS][2];
1177 unsigned int interintra_mode[BLOCK_SIZE_GROUPS][INTERINTRA_MODES];
1178 unsigned int wedge_interintra[BLOCK_SIZES_ALL][2];
1179 unsigned int compound_type[BLOCK_SIZES_ALL][MASKED_COMPOUND_TYPES];
1180 unsigned int motion_mode[BLOCK_SIZES_ALL][MOTION_MODES];
1181 unsigned int obmc[BLOCK_SIZES_ALL][2];
1182 unsigned int intra_inter[INTRA_INTER_CONTEXTS][2];
1183 unsigned int comp_inter[COMP_INTER_CONTEXTS][2];
1184 unsigned int comp_ref_type[COMP_REF_TYPE_CONTEXTS][2];
1185 unsigned int uni_comp_ref[UNI_COMP_REF_CONTEXTS][UNIDIR_COMP_REFS - 1][2];
1186 unsigned int single_ref[REF_CONTEXTS][SINGLE_REFS - 1][2];
1187 unsigned int comp_ref[REF_CONTEXTS][FWD_REFS - 1][2];
1188 unsigned int comp_bwdref[REF_CONTEXTS][BWD_REFS - 1][2];
1189 unsigned int intrabc[2];
1190
1191 unsigned int txfm_partition[TXFM_PARTITION_CONTEXTS][2];
1192 unsigned int intra_tx_size[MAX_TX_CATS][TX_SIZE_CONTEXTS][MAX_TX_DEPTH + 1];
1193 unsigned int skip_mode[SKIP_MODE_CONTEXTS][2];
1194 unsigned int skip_txfm[SKIP_CONTEXTS][2];
1195 unsigned int compound_index[COMP_INDEX_CONTEXTS][2];
1196 unsigned int comp_group_idx[COMP_GROUP_IDX_CONTEXTS][2];
1197 unsigned int delta_q[DELTA_Q_PROBS][2];
1198 unsigned int delta_lf_multi[FRAME_LF_COUNT][DELTA_LF_PROBS][2];
1199 unsigned int delta_lf[DELTA_LF_PROBS][2];
1200
1201 unsigned int inter_ext_tx[EXT_TX_SETS_INTER][EXT_TX_SIZES][TX_TYPES];
1202 unsigned int intra_ext_tx[EXT_TX_SETS_INTRA][EXT_TX_SIZES][INTRA_MODES]
1203 [TX_TYPES];
1204 unsigned int filter_intra_mode[FILTER_INTRA_MODES];
1205 unsigned int filter_intra[BLOCK_SIZES_ALL][2];
1206 unsigned int switchable_restore[RESTORE_SWITCHABLE_TYPES];
1207 unsigned int wiener_restore[2];
1208 unsigned int sgrproj_restore[2];
1209 #endif // CONFIG_ENTROPY_STATS
1210
1211 unsigned int switchable_interp[SWITCHABLE_FILTER_CONTEXTS]
1212 [SWITCHABLE_FILTERS];
1213 } FRAME_COUNTS;
1214
1215 #define INTER_MODE_RD_DATA_OVERALL_SIZE 6400
1216
1217 typedef struct {
1218 int ready;
1219 double a;
1220 double b;
1221 double dist_mean;
1222 double ld_mean;
1223 double sse_mean;
1224 double sse_sse_mean;
1225 double sse_ld_mean;
1226 int num;
1227 double dist_sum;
1228 double ld_sum;
1229 double sse_sum;
1230 double sse_sse_sum;
1231 double sse_ld_sum;
1232 } InterModeRdModel;
1233
1234 typedef struct {
1235 int idx;
1236 int64_t rd;
1237 } RdIdxPair;
1238 // TODO(angiebird): This is an estimated size. We still need to figure what is
1239 // the maximum number of modes.
1240 #define MAX_INTER_MODES 1024
1241 // TODO(any): rename this struct to something else. There is already another
1242 // struct called inter_mode_info, which makes this terribly confusing.
1243 /*!\endcond */
1244 /*!
1245 * \brief Struct used to hold inter mode data for fast tx search.
1246 *
1247 * This struct is used to perform a full transform search only on winning
1248 * candidates searched with an estimate for transform coding RD.
1249 */
1250 typedef struct inter_modes_info {
1251 /*!
1252 * The number of inter modes for which data was stored in each of the
1253 * following arrays.
1254 */
1255 int num;
1256 /*!
1257 * Mode info struct for each of the candidate modes.
1258 */
1259 MB_MODE_INFO mbmi_arr[MAX_INTER_MODES];
1260 /*!
1261 * The rate for each of the candidate modes.
1262 */
1263 int mode_rate_arr[MAX_INTER_MODES];
1264 /*!
1265 * The sse of the predictor for each of the candidate modes.
1266 */
1267 int64_t sse_arr[MAX_INTER_MODES];
1268 /*!
1269 * The estimated rd of the predictor for each of the candidate modes.
1270 */
1271 int64_t est_rd_arr[MAX_INTER_MODES];
1272 /*!
1273 * The rate and mode index for each of the candidate modes.
1274 */
1275 RdIdxPair rd_idx_pair_arr[MAX_INTER_MODES];
1276 /*!
1277 * The full rd stats for each of the candidate modes.
1278 */
1279 RD_STATS rd_cost_arr[MAX_INTER_MODES];
1280 /*!
1281 * The full rd stats of luma only for each of the candidate modes.
1282 */
1283 RD_STATS rd_cost_y_arr[MAX_INTER_MODES];
1284 /*!
1285 * The full rd stats of chroma only for each of the candidate modes.
1286 */
1287 RD_STATS rd_cost_uv_arr[MAX_INTER_MODES];
1288 } InterModesInfo;
1289
1290 /*!\cond */
1291 typedef struct {
1292 // TODO(kyslov): consider changing to 64bit
1293
1294 // This struct is used for computing variance in choose_partitioning(), where
1295 // the max number of samples within a superblock is 32x32 (with 4x4 avg).
1296 // With 8bit bitdepth, uint32_t is enough for sum_square_error (2^8 * 2^8 * 32
1297 // * 32 = 2^26). For high bitdepth we need to consider changing this to 64 bit
1298 uint32_t sum_square_error;
1299 int32_t sum_error;
1300 int log2_count;
1301 int variance;
1302 } VPartVar;
1303
1304 typedef struct {
1305 VPartVar none;
1306 VPartVar horz[2];
1307 VPartVar vert[2];
1308 } VPVariance;
1309
1310 typedef struct {
1311 VPVariance part_variances;
1312 VPartVar split[4];
1313 } VP4x4;
1314
1315 typedef struct {
1316 VPVariance part_variances;
1317 VP4x4 split[4];
1318 } VP8x8;
1319
1320 typedef struct {
1321 VPVariance part_variances;
1322 VP8x8 split[4];
1323 } VP16x16;
1324
1325 typedef struct {
1326 VPVariance part_variances;
1327 VP16x16 split[4];
1328 } VP32x32;
1329
1330 typedef struct {
1331 VPVariance part_variances;
1332 VP32x32 split[4];
1333 } VP64x64;
1334
1335 typedef struct {
1336 VPVariance part_variances;
1337 VP64x64 *split;
1338 } VP128x128;
1339
1340 /*!\endcond */
1341
1342 /*!
1343 * \brief Thresholds for variance based partitioning.
1344 */
1345 typedef struct {
1346 /*!
1347 * If block variance > threshold, then that block is forced to split.
1348 * thresholds[0] - threshold for 128x128;
1349 * thresholds[1] - threshold for 64x64;
1350 * thresholds[2] - threshold for 32x32;
1351 * thresholds[3] - threshold for 16x16;
1352 * thresholds[4] - threshold for 8x8;
1353 */
1354 int64_t thresholds[5];
1355
1356 /*!
1357 * MinMax variance threshold for 8x8 sub blocks of a 16x16 block. If actual
1358 * minmax > threshold_minmax, the 16x16 is forced to split.
1359 */
1360 int64_t threshold_minmax;
1361 } VarBasedPartitionInfo;
1362
1363 /*!
1364 * \brief Encoder parameters for synchronization of row based multi-threading
1365 */
1366 typedef struct {
1367 #if CONFIG_MULTITHREAD
1368 /**
1369 * \name Synchronization objects for top-right dependency.
1370 */
1371 /**@{*/
1372 pthread_mutex_t *mutex_; /*!< Mutex lock object */
1373 pthread_cond_t *cond_; /*!< Condition variable */
1374 /**@}*/
1375 #endif // CONFIG_MULTITHREAD
1376 /*!
1377 * Buffer to store the superblock whose encoding is complete.
1378 * num_finished_cols[i] stores the number of superblocks which finished
1379 * encoding in the ith superblock row.
1380 */
1381 int *num_finished_cols;
1382 /*!
1383 * Denotes the superblock interval at which conditional signalling should
1384 * happen. Also denotes the minimum number of extra superblocks of the top row
1385 * to be complete to start encoding the current superblock. A value of 1
1386 * indicates top-right dependency.
1387 */
1388 int sync_range;
1389 /*!
1390 * Denotes the additional number of superblocks in the previous row to be
1391 * complete to start encoding the current superblock when intraBC tool is
1392 * enabled. This additional top-right delay is required to satisfy the
1393 * hardware constraints for intraBC tool when row multithreading is enabled.
1394 */
1395 int intrabc_extra_top_right_sb_delay;
1396 /*!
1397 * Number of superblock rows.
1398 */
1399 int rows;
1400 /*!
1401 * The superblock row (in units of MI blocks) to be processed next.
1402 */
1403 int next_mi_row;
1404 /*!
1405 * Number of threads processing the current tile.
1406 */
1407 int num_threads_working;
1408 } AV1EncRowMultiThreadSync;
1409
1410 /*!\cond */
1411
1412 // TODO(jingning) All spatially adaptive variables should go to TileDataEnc.
1413 typedef struct TileDataEnc {
1414 TileInfo tile_info;
1415 DECLARE_ALIGNED(16, FRAME_CONTEXT, tctx);
1416 FRAME_CONTEXT *row_ctx;
1417 uint64_t abs_sum_level;
1418 uint8_t allow_update_cdf;
1419 InterModeRdModel inter_mode_rd_models[BLOCK_SIZES_ALL];
1420 AV1EncRowMultiThreadSync row_mt_sync;
1421 MV firstpass_top_mv;
1422 } TileDataEnc;
1423
1424 typedef struct RD_COUNTS {
1425 int compound_ref_used_flag;
1426 int skip_mode_used_flag;
1427 int tx_type_used[TX_SIZES_ALL][TX_TYPES];
1428 int obmc_used[BLOCK_SIZES_ALL][2];
1429 int warped_used[2];
1430 int newmv_or_intra_blocks;
1431 uint64_t seg_tmp_pred_cost[2];
1432 } RD_COUNTS;
1433
1434 typedef struct ThreadData {
1435 MACROBLOCK mb;
1436 RD_COUNTS rd_counts;
1437 FRAME_COUNTS *counts;
1438 PC_TREE_SHARED_BUFFERS shared_coeff_buf;
1439 SIMPLE_MOTION_DATA_TREE *sms_tree;
1440 SIMPLE_MOTION_DATA_TREE *sms_root;
1441 uint32_t *hash_value_buffer[2][2];
1442 OBMCBuffer obmc_buffer;
1443 PALETTE_BUFFER *palette_buffer;
1444 CompoundTypeRdBuffers comp_rd_buffer;
1445 CONV_BUF_TYPE *tmp_conv_dst;
1446 uint64_t abs_sum_level;
1447 uint8_t *tmp_pred_bufs[2];
1448 int intrabc_used;
1449 int deltaq_used;
1450 int coefficient_size;
1451 int max_mv_magnitude;
1452 int interp_filter_selected[SWITCHABLE];
1453 FRAME_CONTEXT *tctx;
1454 VP64x64 *vt64x64;
1455 int32_t num_64x64_blocks;
1456 PICK_MODE_CONTEXT *firstpass_ctx;
1457 TemporalFilterData tf_data;
1458 TplTxfmStats tpl_txfm_stats;
1459 // Pointer to the array of structures to store gradient information of each
1460 // pixel in a superblock. The buffer constitutes of MAX_SB_SQUARE pixel level
1461 // structures for each of the plane types (PLANE_TYPE_Y and PLANE_TYPE_UV).
1462 PixelLevelGradientInfo *pixel_gradient_info;
1463 // Pointer to the array of structures to store source variance information of
1464 // each 4x4 sub-block in a superblock. Block4x4VarInfo structure is used to
1465 // store source variance and log of source variance of each 4x4 sub-block
1466 // for subsequent retrieval.
1467 Block4x4VarInfo *src_var_info_of_4x4_sub_blocks;
1468 // The pc tree root for RTC non-rd case.
1469 PC_TREE *rt_pc_root;
1470 } ThreadData;
1471
1472 struct EncWorkerData;
1473
1474 /*!\endcond */
1475
1476 /*!
1477 * \brief Encoder data related to row-based multi-threading
1478 */
1479 typedef struct {
1480 /*!
1481 * Number of tile rows for which row synchronization memory is allocated.
1482 */
1483 int allocated_tile_rows;
1484 /*!
1485 * Number of tile cols for which row synchronization memory is allocated.
1486 */
1487 int allocated_tile_cols;
1488 /*!
1489 * Number of rows for which row synchronization memory is allocated
1490 * per tile. During first-pass/look-ahead stage this equals the
1491 * maximum number of macroblock rows in a tile. During encode stage,
1492 * this equals the maximum number of superblock rows in a tile.
1493 */
1494 int allocated_rows;
1495 /*!
1496 * Number of columns for which entropy context memory is allocated
1497 * per tile. During encode stage, this equals the maximum number of
1498 * superblock columns in a tile minus 1. The entropy context memory
1499 * is not allocated during first-pass/look-ahead stage.
1500 */
1501 int allocated_cols;
1502
1503 /*!
1504 * thread_id_to_tile_id[i] indicates the tile id assigned to the ith thread.
1505 */
1506 int thread_id_to_tile_id[MAX_NUM_THREADS];
1507
1508 /*!
1509 * num_tile_cols_done[i] indicates the number of tile columns whose encoding
1510 * is complete in the ith superblock row.
1511 */
1512 int *num_tile_cols_done;
1513
1514 /*!
1515 * Number of superblock rows in a frame for which 'num_tile_cols_done' is
1516 * allocated.
1517 */
1518 int allocated_sb_rows;
1519
1520 #if CONFIG_MULTITHREAD
1521 /*!
1522 * Mutex lock used while dispatching jobs.
1523 */
1524 pthread_mutex_t *mutex_;
1525 /*!
1526 * Condition variable used to dispatch loopfilter jobs.
1527 */
1528 pthread_cond_t *cond_;
1529 #endif
1530
1531 /**
1532 * \name Row synchronization related function pointers.
1533 */
1534 /**@{*/
1535 /*!
1536 * Reader.
1537 */
1538 void (*sync_read_ptr)(AV1EncRowMultiThreadSync *const, int, int);
1539 /*!
1540 * Writer.
1541 */
1542 void (*sync_write_ptr)(AV1EncRowMultiThreadSync *const, int, int, int);
1543 /**@}*/
1544 } AV1EncRowMultiThreadInfo;
1545
1546 /*!
1547 * \brief Max number of recodes used to track the frame probabilities.
1548 */
1549 #define NUM_RECODES_PER_FRAME 10
1550
1551 /*!
1552 * \brief Max number of frames that can be encoded in a parallel encode set.
1553 */
1554 #define MAX_PARALLEL_FRAMES 4
1555
1556 /*!
1557 * \brief Buffers to be backed up during parallel encode set to be restored
1558 * later.
1559 */
1560 typedef struct RestoreStateBuffers {
1561 /*!
1562 * Backup of original CDEF srcbuf.
1563 */
1564 uint16_t *cdef_srcbuf;
1565
1566 /*!
1567 * Backup of original CDEF colbuf.
1568 */
1569 uint16_t *cdef_colbuf[MAX_MB_PLANE];
1570
1571 /*!
1572 * Backup of original LR rst_tmpbuf.
1573 */
1574 int32_t *rst_tmpbuf;
1575
1576 /*!
1577 * Backup of original LR rlbs.
1578 */
1579 RestorationLineBuffers *rlbs;
1580 } RestoreStateBuffers;
1581
1582 /*!
1583 * \brief Primary Encoder parameters related to multi-threading.
1584 */
1585 typedef struct PrimaryMultiThreadInfo {
1586 /*!
1587 * Number of workers created for multi-threading.
1588 */
1589 int num_workers;
1590
1591 /*!
1592 * Number of workers used for different MT modules.
1593 */
1594 int num_mod_workers[NUM_MT_MODULES];
1595
1596 /*!
1597 * Synchronization object used to launch job in the worker thread.
1598 */
1599 AVxWorker *workers;
1600
1601 /*!
1602 * Data specific to each worker in encoder multi-threading.
1603 * tile_thr_data[i] stores the worker data of the ith thread.
1604 */
1605 struct EncWorkerData *tile_thr_data;
1606
1607 /*!
1608 * CDEF row multi-threading data.
1609 */
1610 AV1CdefWorkerData *cdef_worker;
1611
1612 /*!
1613 * Primary(Level 1) Synchronization object used to launch job in the worker
1614 * thread.
1615 */
1616 AVxWorker *p_workers[MAX_PARALLEL_FRAMES];
1617
1618 /*!
1619 * Number of primary workers created for multi-threading.
1620 */
1621 int p_num_workers;
1622 } PrimaryMultiThreadInfo;
1623
1624 /*!
1625 * \brief Encoder parameters related to multi-threading.
1626 */
1627 typedef struct MultiThreadInfo {
1628 /*!
1629 * Number of workers created for multi-threading.
1630 */
1631 int num_workers;
1632
1633 /*!
1634 * Number of workers used for different MT modules.
1635 */
1636 int num_mod_workers[NUM_MT_MODULES];
1637
1638 /*!
1639 * Synchronization object used to launch job in the worker thread.
1640 */
1641 AVxWorker *workers;
1642
1643 /*!
1644 * Data specific to each worker in encoder multi-threading.
1645 * tile_thr_data[i] stores the worker data of the ith thread.
1646 */
1647 struct EncWorkerData *tile_thr_data;
1648
1649 /*!
1650 * When set, indicates that row based multi-threading of the encoder is
1651 * enabled.
1652 */
1653 bool row_mt_enabled;
1654
1655 /*!
1656 * When set, indicates that multi-threading for bitstream packing is enabled.
1657 */
1658 bool pack_bs_mt_enabled;
1659
1660 /*!
1661 * Encoder row multi-threading data.
1662 */
1663 AV1EncRowMultiThreadInfo enc_row_mt;
1664
1665 /*!
1666 * Tpl row multi-threading data.
1667 */
1668 AV1TplRowMultiThreadInfo tpl_row_mt;
1669
1670 /*!
1671 * Loop Filter multi-threading object.
1672 */
1673 AV1LfSync lf_row_sync;
1674
1675 /*!
1676 * Loop Restoration multi-threading object.
1677 */
1678 AV1LrSync lr_row_sync;
1679
1680 /*!
1681 * Pack bitstream multi-threading object.
1682 */
1683 AV1EncPackBSSync pack_bs_sync;
1684
1685 /*!
1686 * Global Motion multi-threading object.
1687 */
1688 AV1GlobalMotionSync gm_sync;
1689
1690 /*!
1691 * Temporal Filter multi-threading object.
1692 */
1693 AV1TemporalFilterSync tf_sync;
1694
1695 /*!
1696 * CDEF search multi-threading object.
1697 */
1698 AV1CdefSync cdef_sync;
1699
1700 /*!
1701 * Pointer to CDEF row multi-threading data for the frame.
1702 */
1703 AV1CdefWorkerData *cdef_worker;
1704
1705 /*!
1706 * Buffers to be stored/restored before/after parallel encode.
1707 */
1708 RestoreStateBuffers restore_state_buf;
1709
1710 /*!
1711 * In multi-threaded realtime encoding with row-mt enabled, pipeline
1712 * loop-filtering after encoding.
1713 */
1714 int pipeline_lpf_mt_with_enc;
1715 } MultiThreadInfo;
1716
1717 /*!\cond */
1718
1719 typedef struct ActiveMap {
1720 int enabled;
1721 int update;
1722 unsigned char *map;
1723 } ActiveMap;
1724
1725 /*!\endcond */
1726
1727 /*!
1728 * \brief Encoder info used for decision on forcing integer motion vectors.
1729 */
1730 typedef struct {
1731 /*!
1732 * cs_rate_array[i] is the fraction of blocks in a frame which either match
1733 * with the collocated block or are smooth, where i is the rate_index.
1734 */
1735 double cs_rate_array[32];
1736 /*!
1737 * rate_index is used to index cs_rate_array.
1738 */
1739 int rate_index;
1740 /*!
1741 * rate_size is the total number of entries populated in cs_rate_array.
1742 */
1743 int rate_size;
1744 } ForceIntegerMVInfo;
1745
1746 /*!\cond */
1747
1748 #if CONFIG_INTERNAL_STATS
1749 // types of stats
1750 enum {
1751 STAT_Y,
1752 STAT_U,
1753 STAT_V,
1754 STAT_ALL,
1755 NUM_STAT_TYPES // This should always be the last member of the enum
1756 } UENUM1BYTE(StatType);
1757
1758 typedef struct IMAGE_STAT {
1759 double stat[NUM_STAT_TYPES];
1760 double worst;
1761 } ImageStat;
1762 #endif // CONFIG_INTERNAL_STATS
1763
1764 typedef struct {
1765 int ref_count;
1766 YV12_BUFFER_CONFIG buf;
1767 } EncRefCntBuffer;
1768
1769 /*!\endcond */
1770
1771 /*!
1772 * \brief Buffer to store mode information at mi_alloc_bsize (4x4 or 8x8) level
1773 *
1774 * This is used for bitstream preparation.
1775 */
1776 typedef struct {
1777 /*!
1778 * frame_base[mi_row * stride + mi_col] stores the mode information of
1779 * block (mi_row,mi_col).
1780 */
1781 MB_MODE_INFO_EXT_FRAME *frame_base;
1782 /*!
1783 * Size of frame_base buffer.
1784 */
1785 int alloc_size;
1786 /*!
1787 * Stride of frame_base buffer.
1788 */
1789 int stride;
1790 } MBMIExtFrameBufferInfo;
1791
1792 /*!\cond */
1793
1794 #if CONFIG_COLLECT_PARTITION_STATS
1795 typedef struct FramePartitionTimingStats {
1796 int partition_decisions[6][EXT_PARTITION_TYPES];
1797 int partition_attempts[6][EXT_PARTITION_TYPES];
1798 int64_t partition_times[6][EXT_PARTITION_TYPES];
1799
1800 int partition_redo;
1801 } FramePartitionTimingStats;
1802 #endif // CONFIG_COLLECT_PARTITION_STATS
1803
1804 #if CONFIG_COLLECT_COMPONENT_TIMING
1805 #include "aom_ports/aom_timer.h"
1806 // Adjust the following to add new components.
1807 enum {
1808 av1_encode_strategy_time,
1809 av1_get_one_pass_rt_params_time,
1810 av1_get_second_pass_params_time,
1811 denoise_and_encode_time,
1812 apply_filtering_time,
1813 av1_tpl_setup_stats_time,
1814 encode_frame_to_data_rate_time,
1815 encode_with_or_without_recode_time,
1816 loop_filter_time,
1817 cdef_time,
1818 loop_restoration_time,
1819 av1_pack_bitstream_final_time,
1820 av1_encode_frame_time,
1821 av1_compute_global_motion_time,
1822 av1_setup_motion_field_time,
1823 encode_sb_row_time,
1824
1825 rd_pick_partition_time,
1826 rd_use_partition_time,
1827 choose_var_based_partitioning_time,
1828 av1_prune_partitions_time,
1829 none_partition_search_time,
1830 split_partition_search_time,
1831 rectangular_partition_search_time,
1832 ab_partitions_search_time,
1833 rd_pick_4partition_time,
1834 encode_sb_time,
1835
1836 rd_pick_sb_modes_time,
1837 av1_rd_pick_intra_mode_sb_time,
1838 av1_rd_pick_inter_mode_sb_time,
1839 set_params_rd_pick_inter_mode_time,
1840 skip_inter_mode_time,
1841 handle_inter_mode_time,
1842 evaluate_motion_mode_for_winner_candidates_time,
1843 do_tx_search_time,
1844 handle_intra_mode_time,
1845 refine_winner_mode_tx_time,
1846 av1_search_palette_mode_time,
1847 handle_newmv_time,
1848 compound_type_rd_time,
1849 interpolation_filter_search_time,
1850 motion_mode_rd_time,
1851
1852 nonrd_use_partition_time,
1853 pick_sb_modes_nonrd_time,
1854 hybrid_intra_mode_search_time,
1855 nonrd_pick_inter_mode_sb_time,
1856 encode_b_nonrd_time,
1857
1858 kTimingComponents,
1859 } UENUM1BYTE(TIMING_COMPONENT);
1860
get_component_name(int index)1861 static INLINE char const *get_component_name(int index) {
1862 switch (index) {
1863 case av1_encode_strategy_time: return "av1_encode_strategy_time";
1864 case av1_get_one_pass_rt_params_time:
1865 return "av1_get_one_pass_rt_params_time";
1866 case av1_get_second_pass_params_time:
1867 return "av1_get_second_pass_params_time";
1868 case denoise_and_encode_time: return "denoise_and_encode_time";
1869 case apply_filtering_time: return "apply_filtering_time";
1870 case av1_tpl_setup_stats_time: return "av1_tpl_setup_stats_time";
1871 case encode_frame_to_data_rate_time:
1872 return "encode_frame_to_data_rate_time";
1873 case encode_with_or_without_recode_time:
1874 return "encode_with_or_without_recode_time";
1875 case loop_filter_time: return "loop_filter_time";
1876 case cdef_time: return "cdef_time";
1877 case loop_restoration_time: return "loop_restoration_time";
1878 case av1_pack_bitstream_final_time: return "av1_pack_bitstream_final_time";
1879 case av1_encode_frame_time: return "av1_encode_frame_time";
1880 case av1_compute_global_motion_time:
1881 return "av1_compute_global_motion_time";
1882 case av1_setup_motion_field_time: return "av1_setup_motion_field_time";
1883 case encode_sb_row_time: return "encode_sb_row_time";
1884
1885 case rd_pick_partition_time: return "rd_pick_partition_time";
1886 case rd_use_partition_time: return "rd_use_partition_time";
1887 case choose_var_based_partitioning_time:
1888 return "choose_var_based_partitioning_time";
1889 case av1_prune_partitions_time: return "av1_prune_partitions_time";
1890 case none_partition_search_time: return "none_partition_search_time";
1891 case split_partition_search_time: return "split_partition_search_time";
1892 case rectangular_partition_search_time:
1893 return "rectangular_partition_search_time";
1894 case ab_partitions_search_time: return "ab_partitions_search_time";
1895 case rd_pick_4partition_time: return "rd_pick_4partition_time";
1896 case encode_sb_time: return "encode_sb_time";
1897
1898 case rd_pick_sb_modes_time: return "rd_pick_sb_modes_time";
1899 case av1_rd_pick_intra_mode_sb_time:
1900 return "av1_rd_pick_intra_mode_sb_time";
1901 case av1_rd_pick_inter_mode_sb_time:
1902 return "av1_rd_pick_inter_mode_sb_time";
1903 case set_params_rd_pick_inter_mode_time:
1904 return "set_params_rd_pick_inter_mode_time";
1905 case skip_inter_mode_time: return "skip_inter_mode_time";
1906 case handle_inter_mode_time: return "handle_inter_mode_time";
1907 case evaluate_motion_mode_for_winner_candidates_time:
1908 return "evaluate_motion_mode_for_winner_candidates_time";
1909 case do_tx_search_time: return "do_tx_search_time";
1910 case handle_intra_mode_time: return "handle_intra_mode_time";
1911 case refine_winner_mode_tx_time: return "refine_winner_mode_tx_time";
1912 case av1_search_palette_mode_time: return "av1_search_palette_mode_time";
1913 case handle_newmv_time: return "handle_newmv_time";
1914 case compound_type_rd_time: return "compound_type_rd_time";
1915 case interpolation_filter_search_time:
1916 return "interpolation_filter_search_time";
1917 case motion_mode_rd_time: return "motion_mode_rd_time";
1918
1919 case nonrd_use_partition_time: return "nonrd_use_partition_time";
1920 case pick_sb_modes_nonrd_time: return "pick_sb_modes_nonrd_time";
1921 case hybrid_intra_mode_search_time: return "hybrid_intra_mode_search_time";
1922 case nonrd_pick_inter_mode_sb_time: return "nonrd_pick_inter_mode_sb_time";
1923 case encode_b_nonrd_time: return "encode_b_nonrd_time";
1924
1925 default: assert(0);
1926 }
1927 return "error";
1928 }
1929 #endif
1930
1931 // The maximum number of internal ARFs except ALTREF_FRAME
1932 #define MAX_INTERNAL_ARFS (REF_FRAMES - BWDREF_FRAME - 1)
1933
1934 /*!\endcond */
1935
1936 /*!
1937 * \brief Parameters related to global motion search
1938 */
1939 typedef struct {
1940 /*!
1941 * Flag to indicate if global motion search needs to be rerun.
1942 */
1943 bool search_done;
1944
1945 /*!
1946 * Array of pointers to the frame buffers holding the reference frames.
1947 * ref_buf[i] stores the pointer to the reference frame of the ith
1948 * reference frame type.
1949 */
1950 YV12_BUFFER_CONFIG *ref_buf[REF_FRAMES];
1951
1952 /*!
1953 * Pointer to the source frame buffer.
1954 */
1955 unsigned char *src_buffer;
1956
1957 /*!
1958 * Holds the number of valid reference frames in past and future directions
1959 * w.r.t. the current frame. num_ref_frames[i] stores the total number of
1960 * valid reference frames in 'i' direction.
1961 */
1962 int num_ref_frames[MAX_DIRECTIONS];
1963
1964 /*!
1965 * Array of structure which stores the valid reference frames in past and
1966 * future directions and their corresponding distance from the source frame.
1967 * reference_frames[i][j] holds the jth valid reference frame type in the
1968 * direction 'i' and its temporal distance from the source frame .
1969 */
1970 FrameDistPair reference_frames[MAX_DIRECTIONS][REF_FRAMES - 1];
1971
1972 /**
1973 * \name Dimensions for which segment map is allocated.
1974 */
1975 /**@{*/
1976 int segment_map_w; /*!< segment map width */
1977 int segment_map_h; /*!< segment map height */
1978 /**@}*/
1979
1980 /*!
1981 * Holds the total number of corner points detected in the source frame.
1982 */
1983 int num_src_corners;
1984
1985 /*!
1986 * Holds the x and y co-ordinates of the corner points detected in the source
1987 * frame. src_corners[i] holds the x co-ordinate and src_corners[i+1] holds
1988 * the y co-ordinate of the ith corner point detected.
1989 */
1990 int src_corners[2 * MAX_CORNERS];
1991 } GlobalMotionInfo;
1992
1993 /*!
1994 * \brief Initial frame dimensions
1995 *
1996 * Tracks the frame dimensions using which:
1997 * - Frame buffers (like altref and util frame buffers) were allocated
1998 * - Motion estimation related initializations were done
1999 * This structure is helpful to reallocate / reinitialize the above when there
2000 * is a change in frame dimensions.
2001 */
2002 typedef struct {
2003 int width; /*!< initial width */
2004 int height; /*!< initial height */
2005 } InitialDimensions;
2006
2007 /*!
2008 * \brief Flags related to interpolation filter search
2009 */
2010 typedef struct {
2011 /*!
2012 * Stores the default value of skip flag depending on chroma format
2013 * Set as 1 for monochrome and 3 for other color formats
2014 */
2015 int default_interp_skip_flags;
2016 /*!
2017 * Filter mask to allow certain interp_filter type.
2018 */
2019 uint16_t interp_filter_search_mask;
2020 } InterpSearchFlags;
2021
2022 /*!
2023 * \brief Parameters for motion vector search process
2024 */
2025 typedef struct {
2026 /*!
2027 * Largest MV component used in a frame.
2028 * The value from the previous frame is used to set the full pixel search
2029 * range for the current frame.
2030 */
2031 int max_mv_magnitude;
2032 /*!
2033 * Parameter indicating initial search window to be used in full-pixel search.
2034 * Range [0, MAX_MVSEARCH_STEPS-2]. Lower value indicates larger window.
2035 */
2036 int mv_step_param;
2037 /*!
2038 * Pointer to sub-pixel search function.
2039 * In encoder: av1_find_best_sub_pixel_tree
2040 * av1_find_best_sub_pixel_tree_pruned
2041 * av1_find_best_sub_pixel_tree_pruned_more
2042 * In MV unit test: av1_return_max_sub_pixel_mv
2043 * av1_return_min_sub_pixel_mv
2044 */
2045 fractional_mv_step_fp *find_fractional_mv_step;
2046 /*!
2047 * Search site configuration for full-pel MV search.
2048 * search_site_cfg[SS_CFG_SRC]: Used in tpl, rd/non-rd inter mode loop, simple
2049 * motion search. search_site_cfg[SS_CFG_LOOKAHEAD]: Used in intraBC, temporal
2050 * filter search_site_cfg[SS_CFG_FPF]: Used during first pass and lookahead
2051 */
2052 search_site_config search_site_cfg[SS_CFG_TOTAL][NUM_DISTINCT_SEARCH_METHODS];
2053 } MotionVectorSearchParams;
2054
2055 /*!
2056 * \brief Refresh frame flags for different type of frames.
2057 *
2058 * If the refresh flag is true for a particular reference frame, after the
2059 * current frame is encoded, the reference frame gets refreshed (updated) to
2060 * be the current frame. Note: Usually at most one flag will be set to true at
2061 * a time. But, for key-frames, all flags are set to true at once.
2062 */
2063 typedef struct {
2064 bool golden_frame; /*!< Refresh flag for golden frame */
2065 bool bwd_ref_frame; /*!< Refresh flag for bwd-ref frame */
2066 bool alt_ref_frame; /*!< Refresh flag for alt-ref frame */
2067 } RefreshFrameInfo;
2068
2069 /*!
2070 * \brief Desired dimensions for an externally triggered resize.
2071 *
2072 * When resize is triggered externally, the desired dimensions are stored in
2073 * this struct until used in the next frame to be coded. These values are
2074 * effective only for one frame and are reset after they are used.
2075 */
2076 typedef struct {
2077 int width; /*!< Desired resized width */
2078 int height; /*!< Desired resized height */
2079 } ResizePendingParams;
2080
2081 /*!
2082 * \brief Refrence frame distance related variables.
2083 */
2084 typedef struct {
2085 /*!
2086 * True relative distance of reference frames w.r.t. the current frame.
2087 */
2088 int ref_relative_dist[INTER_REFS_PER_FRAME];
2089 /*!
2090 * The nearest reference w.r.t. current frame in the past.
2091 */
2092 int8_t nearest_past_ref;
2093 /*!
2094 * The nearest reference w.r.t. current frame in the future.
2095 */
2096 int8_t nearest_future_ref;
2097 } RefFrameDistanceInfo;
2098
2099 /*!
2100 * \brief Parameters used for winner mode processing.
2101 *
2102 * This is a basic two pass approach: in the first pass, we reduce the number of
2103 * transform searches based on some thresholds during the rdopt process to find
2104 * the "winner mode". In the second pass, we perform a more through tx search
2105 * on the winner mode.
2106 * There are some arrays in the struct, and their indices are used in the
2107 * following manner:
2108 * Index 0: Default mode evaluation, Winner mode processing is not applicable
2109 * (Eg : IntraBc).
2110 * Index 1: Mode evaluation.
2111 * Index 2: Winner mode evaluation
2112 * Index 1 and 2 are only used when the respective speed feature is on.
2113 */
2114 typedef struct {
2115 /*!
2116 * Threshold to determine if trellis optimization is to be enabled
2117 * based on :
2118 * 0 : dist threshold
2119 * 1 : satd threshold
2120 * Corresponds to enable_winner_mode_for_coeff_opt speed feature.
2121 */
2122 unsigned int coeff_opt_thresholds[MODE_EVAL_TYPES][2];
2123
2124 /*!
2125 * Determines the tx size search method during rdopt.
2126 * Corresponds to enable_winner_mode_for_tx_size_srch speed feature.
2127 */
2128 TX_SIZE_SEARCH_METHOD tx_size_search_methods[MODE_EVAL_TYPES];
2129
2130 /*!
2131 * Controls how often we should approximate prediction error with tx
2132 * coefficients. If it's 0, then never. If 1, then it's during the tx_type
2133 * search only. If 2, then always.
2134 * Corresponds to tx_domain_dist_level speed feature.
2135 */
2136 unsigned int use_transform_domain_distortion[MODE_EVAL_TYPES];
2137
2138 /*!
2139 * Threshold to approximate pixel domain distortion with transform domain
2140 * distortion. This is only used if use_transform_domain_distortion is on.
2141 * Corresponds to enable_winner_mode_for_use_tx_domain_dist speed feature.
2142 */
2143 unsigned int tx_domain_dist_threshold[MODE_EVAL_TYPES];
2144
2145 /*!
2146 * Controls how often we should try to skip the transform process based on
2147 * result from dct.
2148 * Corresponds to use_skip_flag_prediction speed feature.
2149 */
2150 unsigned int skip_txfm_level[MODE_EVAL_TYPES];
2151
2152 /*!
2153 * Predict DC only txfm blocks for default, mode and winner mode evaluation.
2154 * Index 0: Default mode evaluation, Winner mode processing is not applicable.
2155 * Index 1: Mode evaluation, Index 2: Winner mode evaluation
2156 */
2157 unsigned int predict_dc_level[MODE_EVAL_TYPES];
2158 } WinnerModeParams;
2159
2160 /*!
2161 * \brief Frame refresh flags set by the external interface.
2162 *
2163 * Flags set by external interface to determine which reference buffers are
2164 * refreshed by this frame. When set, the encoder will update the particular
2165 * reference frame buffer with the contents of the current frame.
2166 */
2167 typedef struct {
2168 bool last_frame; /*!< Refresh flag for last frame */
2169 bool golden_frame; /*!< Refresh flag for golden frame */
2170 bool bwd_ref_frame; /*!< Refresh flag for bwd-ref frame */
2171 bool alt2_ref_frame; /*!< Refresh flag for alt2-ref frame */
2172 bool alt_ref_frame; /*!< Refresh flag for alt-ref frame */
2173 /*!
2174 * Flag indicating if the update of refresh frame flags is pending.
2175 */
2176 bool update_pending;
2177 } ExtRefreshFrameFlagsInfo;
2178
2179 /*!
2180 * \brief Flags signalled by the external interface at frame level.
2181 */
2182 typedef struct {
2183 /*!
2184 * Bit mask to disable certain reference frame types.
2185 */
2186 int ref_frame_flags;
2187
2188 /*!
2189 * Frame refresh flags set by the external interface.
2190 */
2191 ExtRefreshFrameFlagsInfo refresh_frame;
2192
2193 /*!
2194 * Flag to enable the update of frame contexts at the end of a frame decode.
2195 */
2196 bool refresh_frame_context;
2197
2198 /*!
2199 * Flag to indicate that update of refresh_frame_context from external
2200 * interface is pending.
2201 */
2202 bool refresh_frame_context_pending;
2203
2204 /*!
2205 * Flag to enable temporal MV prediction.
2206 */
2207 bool use_ref_frame_mvs;
2208
2209 /*!
2210 * Indicates whether the current frame is to be coded as error resilient.
2211 */
2212 bool use_error_resilient;
2213
2214 /*!
2215 * Indicates whether the current frame is to be coded as s-frame.
2216 */
2217 bool use_s_frame;
2218
2219 /*!
2220 * Indicates whether the current frame's primary_ref_frame is set to
2221 * PRIMARY_REF_NONE.
2222 */
2223 bool use_primary_ref_none;
2224 } ExternalFlags;
2225
2226 /*!\cond */
2227
2228 typedef struct {
2229 // Some misc info
2230 int high_prec;
2231 int q;
2232 int order;
2233
2234 // MV counters
2235 int inter_count;
2236 int intra_count;
2237 int default_mvs;
2238 int mv_joint_count[4];
2239 int last_bit_zero;
2240 int last_bit_nonzero;
2241
2242 // Keep track of the rates
2243 int total_mv_rate;
2244 int hp_total_mv_rate;
2245 int lp_total_mv_rate;
2246
2247 // Texture info
2248 int horz_text;
2249 int vert_text;
2250 int diag_text;
2251
2252 // Whether the current struct contains valid data
2253 int valid;
2254 } MV_STATS;
2255
2256 typedef struct WeberStats {
2257 int64_t mb_wiener_variance;
2258 int64_t src_variance;
2259 int64_t rec_variance;
2260 int16_t src_pix_max;
2261 int16_t rec_pix_max;
2262 int64_t distortion;
2263 int64_t satd;
2264 double max_scale;
2265 } WeberStats;
2266
2267 typedef struct {
2268 struct loopfilter lf;
2269 CdefInfo cdef_info;
2270 YV12_BUFFER_CONFIG copy_buffer;
2271 RATE_CONTROL rc;
2272 MV_STATS mv_stats;
2273 } CODING_CONTEXT;
2274
2275 typedef struct {
2276 int frame_width;
2277 int frame_height;
2278 int mi_rows;
2279 int mi_cols;
2280 int mb_rows;
2281 int mb_cols;
2282 int num_mbs;
2283 aom_bit_depth_t bit_depth;
2284 int subsampling_x;
2285 int subsampling_y;
2286 } FRAME_INFO;
2287
2288 /*!
2289 * \brief This structure stores different types of frame indices.
2290 */
2291 typedef struct {
2292 int show_frame_count;
2293 } FRAME_INDEX_SET;
2294
2295 /*!\endcond */
2296
2297 /*!
2298 * \brief Segmentation related information for the current frame.
2299 */
2300 typedef struct {
2301 /*!
2302 * 3-bit number containing the segment affiliation for each 4x4 block in the
2303 * frame. map[y * stride + x] contains the segment id of the 4x4 block at
2304 * (x,y) position.
2305 */
2306 uint8_t *map;
2307 /*!
2308 * Flag to indicate if current frame has lossless segments or not.
2309 * 1: frame has at least one lossless segment.
2310 * 0: frame has no lossless segments.
2311 */
2312 bool has_lossless_segment;
2313 } EncSegmentationInfo;
2314
2315 /*!
2316 * \brief Frame time stamps.
2317 */
2318 typedef struct {
2319 /*!
2320 * Start time stamp of the previous frame
2321 */
2322 int64_t prev_ts_start;
2323 /*!
2324 * End time stamp of the previous frame
2325 */
2326 int64_t prev_ts_end;
2327 /*!
2328 * Start time stamp of the first frame
2329 */
2330 int64_t first_ts_start;
2331 } TimeStamps;
2332
2333 /*!
2334 * Pointers to the memory allocated for frame level transform coeff related
2335 * info.
2336 */
2337 typedef struct {
2338 /*!
2339 * Pointer to the transformed coefficients buffer.
2340 */
2341 tran_low_t *tcoeff;
2342 /*!
2343 * Pointer to the eobs buffer.
2344 */
2345 uint16_t *eobs;
2346 /*!
2347 * Pointer to the entropy_ctx buffer.
2348 */
2349 uint8_t *entropy_ctx;
2350 } CoeffBufferPool;
2351
2352 #if !CONFIG_REALTIME_ONLY
2353 /*!\cond */
2354 // DUCKY_ENCODE_FRAME_MODE is c version of EncodeFrameMode
2355 enum {
2356 DUCKY_ENCODE_FRAME_MODE_NONE, // Let native AV1 determine q index and rdmult
2357 DUCKY_ENCODE_FRAME_MODE_QINDEX, // DuckyEncode determines q index and AV1
2358 // determines rdmult
2359 DUCKY_ENCODE_FRAME_MODE_QINDEX_RDMULT, // DuckyEncode determines q index and
2360 // rdmult
2361 } UENUM1BYTE(DUCKY_ENCODE_FRAME_MODE);
2362
2363 enum {
2364 DUCKY_ENCODE_GOP_MODE_NONE, // native AV1 decides GOP
2365 DUCKY_ENCODE_GOP_MODE_RCL, // rate control lib decides GOP
2366 } UENUM1BYTE(DUCKY_ENCODE_GOP_MODE);
2367
2368 typedef struct DuckyEncodeFrameInfo {
2369 DUCKY_ENCODE_FRAME_MODE qp_mode;
2370 DUCKY_ENCODE_GOP_MODE gop_mode;
2371 int q_index;
2372 int rdmult;
2373 // These two arrays are equivalent to std::vector<SuperblockEncodeParameters>
2374 int *superblock_encode_qindex;
2375 int *superblock_encode_rdmult;
2376 int delta_q_enabled;
2377 } DuckyEncodeFrameInfo;
2378
2379 typedef struct DuckyEncodeFrameResult {
2380 int global_order_idx;
2381 int q_index;
2382 int rdmult;
2383 int rate;
2384 int64_t dist;
2385 double psnr;
2386 } DuckyEncodeFrameResult;
2387
2388 typedef struct DuckyEncodeInfo {
2389 DuckyEncodeFrameInfo frame_info;
2390 DuckyEncodeFrameResult frame_result;
2391 } DuckyEncodeInfo;
2392 /*!\endcond */
2393 #endif
2394
2395 /*!\cond */
2396 typedef struct RTC_REF {
2397 /*!
2398 * LAST_FRAME (0), LAST2_FRAME(1), LAST3_FRAME(2), GOLDEN_FRAME(3),
2399 * BWDREF_FRAME(4), ALTREF2_FRAME(5), ALTREF_FRAME(6).
2400 */
2401 int reference[INTER_REFS_PER_FRAME];
2402 int ref_idx[INTER_REFS_PER_FRAME];
2403 int refresh[REF_FRAMES];
2404 int set_ref_frame_config;
2405 int non_reference_frame;
2406 int ref_frame_comp[3];
2407 int gld_idx_1layer;
2408 } RTC_REF;
2409 /*!\endcond */
2410
2411 /*!
2412 * \brief Structure to hold data corresponding to an encoded frame.
2413 */
2414 typedef struct AV1_COMP_DATA {
2415 /*!
2416 * Buffer to store packed bitstream data of a frame.
2417 */
2418 unsigned char *cx_data;
2419
2420 /*!
2421 * Allocated size of the cx_data buffer.
2422 */
2423 size_t cx_data_sz;
2424
2425 /*!
2426 * Size of data written in the cx_data buffer.
2427 */
2428 size_t frame_size;
2429
2430 /*!
2431 * Flags for the frame.
2432 */
2433 unsigned int lib_flags;
2434
2435 /*!
2436 * Time stamp for start of frame.
2437 */
2438 int64_t ts_frame_start;
2439
2440 /*!
2441 * Time stamp for end of frame.
2442 */
2443 int64_t ts_frame_end;
2444
2445 /*!
2446 * Flag to indicate flush call.
2447 */
2448 int flush;
2449
2450 /*!
2451 * Time base for sequence.
2452 */
2453 const aom_rational64_t *timestamp_ratio;
2454
2455 /*!
2456 * Decide to pop the source for this frame from input buffer queue.
2457 */
2458 int pop_lookahead;
2459
2460 /*!
2461 * Display order hint of frame whose packed data is in cx_data buffer.
2462 */
2463 int frame_display_order_hint;
2464 } AV1_COMP_DATA;
2465
2466 /*!
2467 * \brief Top level primary encoder structure
2468 */
2469 typedef struct AV1_PRIMARY {
2470 /*!
2471 * Array of frame level encoder stage top level structures
2472 */
2473 struct AV1_COMP *parallel_cpi[MAX_PARALLEL_FRAMES];
2474
2475 /*!
2476 * Array of structures to hold data of frames encoded in a given parallel
2477 * encode set.
2478 */
2479 struct AV1_COMP_DATA parallel_frames_data[MAX_PARALLEL_FRAMES - 1];
2480 #if CONFIG_FPMT_TEST
2481 /*!
2482 * Flag which enables/disables simulation path for fpmt unit test.
2483 * 0 - FPMT integration
2484 * 1 - FPMT simulation
2485 */
2486 FPMT_TEST_ENC_CFG fpmt_unit_test_cfg;
2487
2488 /*!
2489 * Temporary variable simulating the delayed frame_probability update.
2490 */
2491 FrameProbInfo temp_frame_probs;
2492
2493 /*!
2494 * Temporary variable holding the updated frame probability across
2495 * frames. Copy its value to temp_frame_probs for frame_parallel_level 0
2496 * frames or last frame in parallel encode set.
2497 */
2498 FrameProbInfo temp_frame_probs_simulation;
2499
2500 /*!
2501 * Temporary variable simulating the delayed update of valid global motion
2502 * model across frames.
2503 */
2504 int temp_valid_gm_model_found[FRAME_UPDATE_TYPES];
2505 #endif // CONFIG_FPMT_TEST
2506 /*!
2507 * Copy of cm->ref_frame_map maintained to facilitate sequential update of
2508 * ref_frame_map by lower layer depth frames encoded ahead of time in a
2509 * parallel encode set.
2510 */
2511 RefCntBuffer *ref_frame_map_copy[REF_FRAMES];
2512
2513 /*!
2514 * Start time stamp of the last encoded show frame
2515 */
2516 int64_t ts_start_last_show_frame;
2517
2518 /*!
2519 * End time stamp of the last encoded show frame
2520 */
2521 int64_t ts_end_last_show_frame;
2522
2523 /*!
2524 * Number of frame level contexts(cpis)
2525 */
2526 int num_fp_contexts;
2527
2528 /*!
2529 * Loopfilter levels of the previous encoded frame.
2530 */
2531 int filter_level[2];
2532
2533 /*!
2534 * Chrominance component loopfilter level of the previous encoded frame.
2535 */
2536 int filter_level_u;
2537
2538 /*!
2539 * Chrominance component loopfilter level of the previous encoded frame.
2540 */
2541 int filter_level_v;
2542
2543 /*!
2544 * Encode stage top level structure
2545 * During frame parallel encode, this is the same as parallel_cpi[0]
2546 */
2547 struct AV1_COMP *cpi;
2548
2549 /*!
2550 * Lookahead processing stage top level structure
2551 */
2552 struct AV1_COMP *cpi_lap;
2553
2554 /*!
2555 * Look-ahead context.
2556 */
2557 struct lookahead_ctx *lookahead;
2558
2559 /*!
2560 * Sequence parameters have been transmitted already and locked
2561 * or not. Once locked av1_change_config cannot change the seq
2562 * parameters.
2563 */
2564 int seq_params_locked;
2565
2566 /*!
2567 * Pointer to internal utility functions that manipulate aom_codec_* data
2568 * structures.
2569 */
2570 struct aom_codec_pkt_list *output_pkt_list;
2571
2572 /*!
2573 * When set, indicates that internal ARFs are enabled.
2574 */
2575 int internal_altref_allowed;
2576
2577 /*!
2578 * Tell if OVERLAY frame shows existing alt_ref frame.
2579 */
2580 int show_existing_alt_ref;
2581
2582 /*!
2583 * Information related to a gf group.
2584 */
2585 GF_GROUP gf_group;
2586
2587 /*!
2588 * Track prior gf group state.
2589 */
2590 GF_STATE gf_state;
2591
2592 /*!
2593 * Flag indicating whether look ahead processing (LAP) is enabled.
2594 */
2595 int lap_enabled;
2596
2597 /*!
2598 * Parameters for AV1 bitstream levels.
2599 */
2600 AV1LevelParams level_params;
2601
2602 /*!
2603 * Calculates PSNR on each frame when set to 1.
2604 */
2605 int b_calculate_psnr;
2606
2607 /*!
2608 * Number of frames left to be encoded, is 0 if limit is not set.
2609 */
2610 int frames_left;
2611
2612 /*!
2613 * Information related to two pass encoding.
2614 */
2615 TWO_PASS twopass;
2616
2617 /*!
2618 * Rate control related parameters.
2619 */
2620 PRIMARY_RATE_CONTROL p_rc;
2621
2622 /*!
2623 * Info and resources used by temporal filtering.
2624 */
2625 TEMPORAL_FILTER_INFO tf_info;
2626 /*!
2627 * Elements part of the sequence header, that are applicable for all the
2628 * frames in the video.
2629 */
2630 SequenceHeader seq_params;
2631
2632 /*!
2633 * Indicates whether to use SVC.
2634 */
2635 int use_svc;
2636
2637 /*!
2638 * If true, buffer removal times are present.
2639 */
2640 bool buffer_removal_time_present;
2641
2642 /*!
2643 * Number of temporal layers: may be > 1 for SVC (scalable vector coding).
2644 */
2645 unsigned int number_temporal_layers;
2646
2647 /*!
2648 * Number of spatial layers: may be > 1 for SVC (scalable vector coding).
2649 */
2650 unsigned int number_spatial_layers;
2651
2652 /*!
2653 * Code and details about current error status.
2654 */
2655 struct aom_internal_error_info error;
2656
2657 /*!
2658 * Function pointers to variants of sse/sad/variance computation functions.
2659 * fn_ptr[i] indicates the list of function pointers corresponding to block
2660 * size i.
2661 */
2662 aom_variance_fn_ptr_t fn_ptr[BLOCK_SIZES_ALL];
2663
2664 /*!
2665 * tpl_sb_rdmult_scaling_factors[i] stores the RD multiplier scaling factor of
2666 * the ith 16 x 16 block in raster scan order.
2667 */
2668 double *tpl_sb_rdmult_scaling_factors;
2669
2670 /*!
2671 * Parameters related to tpl.
2672 */
2673 TplParams tpl_data;
2674
2675 /*!
2676 * Motion vector stats of the previous encoded frame.
2677 */
2678 MV_STATS mv_stats;
2679
2680 #if CONFIG_INTERNAL_STATS
2681 /*!\cond */
2682 uint64_t total_time_receive_data;
2683 uint64_t total_time_compress_data;
2684
2685 unsigned int total_mode_chosen_counts[MAX_MODES];
2686
2687 int count[2];
2688 uint64_t total_sq_error[2];
2689 uint64_t total_samples[2];
2690 ImageStat psnr[2];
2691
2692 double total_blockiness;
2693 double worst_blockiness;
2694
2695 int total_bytes;
2696 double summed_quality;
2697 double summed_weights;
2698 double summed_quality_hbd;
2699 double summed_weights_hbd;
2700 unsigned int total_recode_hits;
2701 double worst_ssim;
2702 double worst_ssim_hbd;
2703
2704 ImageStat fastssim;
2705 ImageStat psnrhvs;
2706
2707 int b_calculate_blockiness;
2708 int b_calculate_consistency;
2709
2710 double total_inconsistency;
2711 double worst_consistency;
2712 Ssimv *ssim_vars;
2713 Metrics metrics;
2714 /*!\endcond */
2715 #endif
2716
2717 #if CONFIG_ENTROPY_STATS
2718 /*!
2719 * Aggregates frame counts for the sequence.
2720 */
2721 FRAME_COUNTS aggregate_fc;
2722 #endif // CONFIG_ENTROPY_STATS
2723
2724 /*!
2725 * For each type of reference frame, this contains the index of a reference
2726 * frame buffer for a reference frame of the same type. We use this to
2727 * choose our primary reference frame (which is the most recent reference
2728 * frame of the same type as the current frame).
2729 */
2730 int fb_of_context_type[REF_FRAMES];
2731
2732 /*!
2733 * Primary Multi-threading parameters.
2734 */
2735 PrimaryMultiThreadInfo p_mt_info;
2736
2737 /*!
2738 * Probabilities for pruning of various AV1 tools.
2739 */
2740 FrameProbInfo frame_probs;
2741
2742 /*!
2743 * Indicates if a valid global motion model has been found in the different
2744 * frame update types of a GF group.
2745 * valid_gm_model_found[i] indicates if valid global motion model has been
2746 * found in the frame update type with enum value equal to i
2747 */
2748 int valid_gm_model_found[FRAME_UPDATE_TYPES];
2749
2750 /*!
2751 * Struct for the reference structure for RTC.
2752 */
2753 RTC_REF rtc_ref;
2754 } AV1_PRIMARY;
2755
2756 /*!
2757 * \brief Top level encoder structure.
2758 */
2759 typedef struct AV1_COMP {
2760 /*!
2761 * Pointer to top level primary encoder structure
2762 */
2763 AV1_PRIMARY *ppi;
2764
2765 /*!
2766 * Quantization and dequantization parameters for internal quantizer setup
2767 * in the encoder.
2768 */
2769 EncQuantDequantParams enc_quant_dequant_params;
2770
2771 /*!
2772 * Structure holding thread specific variables.
2773 */
2774 ThreadData td;
2775
2776 /*!
2777 * Statistics collected at frame level.
2778 */
2779 FRAME_COUNTS counts;
2780
2781 /*!
2782 * Holds buffer storing mode information at 4x4/8x8 level.
2783 */
2784 MBMIExtFrameBufferInfo mbmi_ext_info;
2785
2786 /*!
2787 * Buffer holding the transform block related information.
2788 * coeff_buffer_base[i] stores the transform block related information of the
2789 * ith superblock in raster scan order.
2790 */
2791 CB_COEFF_BUFFER *coeff_buffer_base;
2792
2793 /*!
2794 * Structure holding pointers to frame level memory allocated for transform
2795 * block related information.
2796 */
2797 CoeffBufferPool coeff_buffer_pool;
2798
2799 /*!
2800 * Structure holding variables common to encoder and decoder.
2801 */
2802 AV1_COMMON common;
2803
2804 /*!
2805 * Encoder configuration related parameters.
2806 */
2807 AV1EncoderConfig oxcf;
2808
2809 /*!
2810 * Stores the trellis optimization type at segment level.
2811 * optimize_seg_arr[i] stores the trellis opt type for ith segment.
2812 */
2813 TRELLIS_OPT_TYPE optimize_seg_arr[MAX_SEGMENTS];
2814
2815 /*!
2816 * Pointer to the frame buffer holding the source frame to be used during the
2817 * current stage of encoding. It can be the raw input, temporally filtered
2818 * input or scaled input.
2819 */
2820 YV12_BUFFER_CONFIG *source;
2821
2822 /*!
2823 * Pointer to the frame buffer holding the last raw source frame.
2824 * last_source is NULL for the following cases:
2825 * 1) First frame
2826 * 2) Alt-ref frames
2827 * 3) All frames for all-intra frame encoding.
2828 */
2829 YV12_BUFFER_CONFIG *last_source;
2830
2831 /*!
2832 * Pointer to the frame buffer holding the unscaled source frame.
2833 * It can be either the raw input or temporally filtered input.
2834 */
2835 YV12_BUFFER_CONFIG *unscaled_source;
2836
2837 /*!
2838 * Frame buffer holding the resized source frame (cropping / superres).
2839 */
2840 YV12_BUFFER_CONFIG scaled_source;
2841
2842 /*!
2843 * Pointer to the frame buffer holding the unscaled last source frame.
2844 */
2845 YV12_BUFFER_CONFIG *unscaled_last_source;
2846
2847 /*!
2848 * Frame buffer holding the resized last source frame.
2849 */
2850 YV12_BUFFER_CONFIG scaled_last_source;
2851
2852 /*!
2853 * Pointer to the original source frame. This is used to determine if the
2854 * content is screen.
2855 */
2856 YV12_BUFFER_CONFIG *unfiltered_source;
2857
2858 /*!
2859 * Frame buffer holding the orig source frame for PSNR calculation in rtc tf
2860 * case.
2861 */
2862 YV12_BUFFER_CONFIG orig_source;
2863
2864 /*!
2865 * Skip tpl setup when tpl data from gop length decision can be reused.
2866 */
2867 int skip_tpl_setup_stats;
2868
2869 /*!
2870 * Scaling factors used in the RD multiplier modulation.
2871 * TODO(sdeng): consider merge the following arrays.
2872 * tpl_rdmult_scaling_factors is a temporary buffer used to store the
2873 * intermediate scaling factors which are used in the calculation of
2874 * tpl_sb_rdmult_scaling_factors. tpl_rdmult_scaling_factors[i] stores the
2875 * intermediate scaling factor of the ith 16 x 16 block in raster scan order.
2876 */
2877 double *tpl_rdmult_scaling_factors;
2878
2879 /*!
2880 * Temporal filter context.
2881 */
2882 TemporalFilterCtx tf_ctx;
2883
2884 /*!
2885 * Variables related to forcing integer mv decisions for the current frame.
2886 */
2887 ForceIntegerMVInfo force_intpel_info;
2888
2889 /*!
2890 * Pointer to the buffer holding the scaled reference frames.
2891 * scaled_ref_buf[i] holds the scaled reference frame of type i.
2892 */
2893 RefCntBuffer *scaled_ref_buf[INTER_REFS_PER_FRAME];
2894
2895 /*!
2896 * Pointer to the buffer holding the last show frame.
2897 */
2898 RefCntBuffer *last_show_frame_buf;
2899
2900 /*!
2901 * Refresh frame flags for golden, bwd-ref and alt-ref frames.
2902 */
2903 RefreshFrameInfo refresh_frame;
2904
2905 /*!
2906 * Flag to reduce the number of reference frame buffers used in rt.
2907 */
2908 int rt_reduce_num_ref_buffers;
2909
2910 /*!
2911 * Flags signalled by the external interface at frame level.
2912 */
2913 ExternalFlags ext_flags;
2914
2915 /*!
2916 * Temporary frame buffer used to store the non-loop filtered reconstructed
2917 * frame during the search of loop filter level.
2918 */
2919 YV12_BUFFER_CONFIG last_frame_uf;
2920
2921 /*!
2922 * Temporary frame buffer used to store the loop restored frame during loop
2923 * restoration search.
2924 */
2925 YV12_BUFFER_CONFIG trial_frame_rst;
2926
2927 /*!
2928 * Ambient reconstruction err target for force key frames.
2929 */
2930 int64_t ambient_err;
2931
2932 /*!
2933 * Parameters related to rate distortion optimization.
2934 */
2935 RD_OPT rd;
2936
2937 /*!
2938 * Temporary coding context used to save and restore when encoding with and
2939 * without super-resolution.
2940 */
2941 CODING_CONTEXT coding_context;
2942
2943 /*!
2944 * Parameters related to global motion search.
2945 */
2946 GlobalMotionInfo gm_info;
2947
2948 /*!
2949 * Parameters related to winner mode processing.
2950 */
2951 WinnerModeParams winner_mode_params;
2952
2953 /*!
2954 * Frame time stamps.
2955 */
2956 TimeStamps time_stamps;
2957
2958 /*!
2959 * Rate control related parameters.
2960 */
2961 RATE_CONTROL rc;
2962
2963 /*!
2964 * Frame rate of the video.
2965 */
2966 double framerate;
2967
2968 /*!
2969 * Bitmask indicating which reference buffers may be referenced by this frame.
2970 */
2971 int ref_frame_flags;
2972
2973 /*!
2974 * speed is passed as a per-frame parameter into the encoder.
2975 */
2976 int speed;
2977
2978 /*!
2979 * sf contains fine-grained config set internally based on speed.
2980 */
2981 SPEED_FEATURES sf;
2982
2983 /*!
2984 * Parameters for motion vector search process.
2985 */
2986 MotionVectorSearchParams mv_search_params;
2987
2988 /*!
2989 * When set, indicates that all reference frames are forward references,
2990 * i.e., all the reference frames are output before the current frame.
2991 */
2992 int all_one_sided_refs;
2993
2994 /*!
2995 * Segmentation related information for current frame.
2996 */
2997 EncSegmentationInfo enc_seg;
2998
2999 /*!
3000 * Parameters related to cyclic refresh aq-mode.
3001 */
3002 CYCLIC_REFRESH *cyclic_refresh;
3003 /*!
3004 * Parameters related to active map. Active maps indicate
3005 * if there is any activity on a 4x4 block basis.
3006 */
3007 ActiveMap active_map;
3008
3009 /*!
3010 * The frame processing order within a GOP.
3011 */
3012 unsigned char gf_frame_index;
3013
3014 #if CONFIG_INTERNAL_STATS
3015 /*!\cond */
3016 uint64_t time_compress_data;
3017
3018 unsigned int mode_chosen_counts[MAX_MODES];
3019 int bytes;
3020 unsigned int frame_recode_hits;
3021 /*!\endcond */
3022 #endif
3023
3024 #if CONFIG_SPEED_STATS
3025 /*!
3026 * For debugging: number of transform searches we have performed.
3027 */
3028 unsigned int tx_search_count;
3029 #endif // CONFIG_SPEED_STATS
3030
3031 /*!
3032 * When set, indicates that the frame is droppable, i.e., this frame
3033 * does not update any reference buffers.
3034 */
3035 int droppable;
3036
3037 /*!
3038 * Stores the frame parameters during encoder initialization.
3039 */
3040 FRAME_INFO frame_info;
3041
3042 /*!
3043 * Stores different types of frame indices.
3044 */
3045 FRAME_INDEX_SET frame_index_set;
3046
3047 /*!
3048 * Structure to store the dimensions of current frame.
3049 */
3050 InitialDimensions initial_dimensions;
3051
3052 /*!
3053 * Number of MBs in the full-size frame; to be used to
3054 * normalize the firstpass stats. This will differ from the
3055 * number of MBs in the current frame when the frame is
3056 * scaled.
3057 */
3058 int initial_mbs;
3059
3060 /*!
3061 * Resize related parameters.
3062 */
3063 ResizePendingParams resize_pending_params;
3064
3065 /*!
3066 * Pointer to struct holding adaptive data/contexts/models for the tile during
3067 * encoding.
3068 */
3069 TileDataEnc *tile_data;
3070 /*!
3071 * Number of tiles for which memory has been allocated for tile_data.
3072 */
3073 int allocated_tiles;
3074
3075 /*!
3076 * Structure to store the palette token related information.
3077 */
3078 TokenInfo token_info;
3079
3080 /*!
3081 * VARIANCE_AQ segment map refresh.
3082 */
3083 int vaq_refresh;
3084
3085 /*!
3086 * Thresholds for variance based partitioning.
3087 */
3088 VarBasedPartitionInfo vbp_info;
3089
3090 /*!
3091 * Number of recodes in the frame.
3092 */
3093 int num_frame_recode;
3094
3095 /*!
3096 * Current frame probability of parallel frames, across recodes.
3097 */
3098 FrameProbInfo frame_new_probs[NUM_RECODES_PER_FRAME];
3099
3100 /*!
3101 * Retain condition for transform type frame_probability calculation
3102 */
3103 int do_update_frame_probs_txtype[NUM_RECODES_PER_FRAME];
3104
3105 /*!
3106 * Retain condition for obmc frame_probability calculation
3107 */
3108 int do_update_frame_probs_obmc[NUM_RECODES_PER_FRAME];
3109
3110 /*!
3111 * Retain condition for warped motion frame_probability calculation
3112 */
3113 int do_update_frame_probs_warp[NUM_RECODES_PER_FRAME];
3114
3115 /*!
3116 * Retain condition for interpolation filter frame_probability calculation
3117 */
3118 int do_update_frame_probs_interpfilter[NUM_RECODES_PER_FRAME];
3119
3120 #if CONFIG_FPMT_TEST
3121 /*!
3122 * Temporary variable for simulation.
3123 * Previous frame's framerate.
3124 */
3125 double temp_framerate;
3126 #endif
3127 /*!
3128 * Updated framerate for the current parallel frame.
3129 * cpi->framerate is updated with new_framerate during
3130 * post encode updates for parallel frames.
3131 */
3132 double new_framerate;
3133
3134 /*!
3135 * Retain condition for fast_extra_bits calculation.
3136 */
3137 int do_update_vbr_bits_off_target_fast;
3138
3139 /*!
3140 * Multi-threading parameters.
3141 */
3142 MultiThreadInfo mt_info;
3143
3144 /*!
3145 * Specifies the frame to be output. It is valid only if show_existing_frame
3146 * is 1. When show_existing_frame is 0, existing_fb_idx_to_show is set to
3147 * INVALID_IDX.
3148 */
3149 int existing_fb_idx_to_show;
3150
3151 /*!
3152 * A flag to indicate if intrabc is ever used in current frame.
3153 */
3154 int intrabc_used;
3155
3156 /*!
3157 * Mark which ref frames can be skipped for encoding current frame during RDO.
3158 */
3159 int prune_ref_frame_mask;
3160
3161 /*!
3162 * Loop Restoration context.
3163 */
3164 AV1LrStruct lr_ctxt;
3165
3166 /*!
3167 * Pointer to list of tables with film grain parameters.
3168 */
3169 aom_film_grain_table_t *film_grain_table;
3170
3171 #if CONFIG_DENOISE
3172 /*!
3173 * Pointer to structure holding the denoised image buffers and the helper
3174 * noise models.
3175 */
3176 struct aom_denoise_and_model_t *denoise_and_model;
3177 #endif
3178
3179 /*!
3180 * Flags related to interpolation filter search.
3181 */
3182 InterpSearchFlags interp_search_flags;
3183
3184 /*!
3185 * Turn on screen content tools flag.
3186 * Note that some videos are not screen content videos, but
3187 * screen content tools could also improve coding efficiency.
3188 * For example, videos with large flat regions, gaming videos that look
3189 * like natural videos.
3190 */
3191 int use_screen_content_tools;
3192
3193 /*!
3194 * A flag to indicate "real" screen content videos.
3195 * For example, screen shares, screen editing.
3196 * This type is true indicates |use_screen_content_tools| must be true.
3197 * In addition, rate control strategy is adjusted when this flag is true.
3198 */
3199 int is_screen_content_type;
3200
3201 #if CONFIG_COLLECT_PARTITION_STATS
3202 /*!
3203 * Accumulates the partition timing stat over the whole frame.
3204 */
3205 FramePartitionTimingStats partition_stats;
3206 #endif // CONFIG_COLLECT_PARTITION_STATS
3207
3208 #if CONFIG_COLLECT_COMPONENT_TIMING
3209 /*!
3210 * component_time[] are initialized to zero while encoder starts.
3211 */
3212 uint64_t component_time[kTimingComponents];
3213 /*!
3214 * Stores timing for individual components between calls of start_timing()
3215 * and end_timing().
3216 */
3217 struct aom_usec_timer component_timer[kTimingComponents];
3218 /*!
3219 * frame_component_time[] are initialized to zero at beginning of each frame.
3220 */
3221 uint64_t frame_component_time[kTimingComponents];
3222 #endif
3223
3224 /*!
3225 * Count the number of OBU_FRAME and OBU_FRAME_HEADER for level calculation.
3226 */
3227 int frame_header_count;
3228
3229 /*!
3230 * Whether any no-zero delta_q was actually used.
3231 */
3232 int deltaq_used;
3233
3234 /*!
3235 * Refrence frame distance related variables.
3236 */
3237 RefFrameDistanceInfo ref_frame_dist_info;
3238
3239 /*!
3240 * ssim_rdmult_scaling_factors[i] stores the RD multiplier scaling factor of
3241 * the ith 16 x 16 block in raster scan order. This scaling factor is used for
3242 * RD multiplier modulation when SSIM tuning is enabled.
3243 */
3244 double *ssim_rdmult_scaling_factors;
3245
3246 #if CONFIG_TUNE_VMAF
3247 /*!
3248 * Parameters for VMAF tuning.
3249 */
3250 TuneVMAFInfo vmaf_info;
3251 #endif
3252
3253 #if CONFIG_TUNE_BUTTERAUGLI
3254 /*!
3255 * Parameters for Butteraugli tuning.
3256 */
3257 TuneButteraugliInfo butteraugli_info;
3258 #endif
3259
3260 /*!
3261 * Parameters for scalable video coding.
3262 */
3263 SVC svc;
3264
3265 /*!
3266 * Indicates whether current processing stage is encode stage or LAP stage.
3267 */
3268 COMPRESSOR_STAGE compressor_stage;
3269
3270 /*!
3271 * Frame type of the last frame. May be used in some heuristics for speeding
3272 * up the encoding.
3273 */
3274 FRAME_TYPE last_frame_type;
3275
3276 /*!
3277 * Number of tile-groups.
3278 */
3279 int num_tg;
3280
3281 /*!
3282 * Super-resolution mode currently being used by the encoder.
3283 * This may / may not be same as user-supplied mode in oxcf->superres_mode
3284 * (when we are recoding to try multiple options for example).
3285 */
3286 aom_superres_mode superres_mode;
3287
3288 /*!
3289 * First pass related data.
3290 */
3291 FirstPassData firstpass_data;
3292
3293 /*!
3294 * Temporal Noise Estimate
3295 */
3296 NOISE_ESTIMATE noise_estimate;
3297
3298 #if CONFIG_AV1_TEMPORAL_DENOISING
3299 /*!
3300 * Temporal Denoiser
3301 */
3302 AV1_DENOISER denoiser;
3303 #endif
3304
3305 /*!
3306 * Count on how many consecutive times a block uses small/zeromv for encoding
3307 * in a scale of 8x8 block.
3308 */
3309 uint8_t *consec_zero_mv;
3310
3311 /*!
3312 * Block size of first pass encoding
3313 */
3314 BLOCK_SIZE fp_block_size;
3315
3316 /*!
3317 * The counter of encoded super block, used to differentiate block names.
3318 * This number starts from 0 and increases whenever a super block is encoded.
3319 */
3320 int sb_counter;
3321
3322 /*!
3323 * Available bitstream buffer size in bytes
3324 */
3325 size_t available_bs_size;
3326
3327 /*!
3328 * The controller of the external partition model.
3329 * It is used to do partition type selection based on external models.
3330 */
3331 ExtPartController ext_part_controller;
3332
3333 /*!
3334 * Motion vector stats of the current encoded frame, used to update the
3335 * ppi->mv_stats during postencode.
3336 */
3337 MV_STATS mv_stats;
3338 /*!
3339 * Stores the reference refresh index for the current frame.
3340 */
3341 int ref_refresh_index;
3342
3343 /*!
3344 * A flag to indicate if the reference refresh index is available for the
3345 * current frame.
3346 */
3347 bool refresh_idx_available;
3348
3349 /*!
3350 * Reference frame index corresponding to the frame to be excluded from being
3351 * used as a reference by frame_parallel_level 2 frame in a parallel
3352 * encode set of lower layer frames.
3353 */
3354 int ref_idx_to_skip;
3355 #if CONFIG_FPMT_TEST
3356 /*!
3357 * Stores the wanted frame buffer index for choosing primary ref frame by a
3358 * frame_parallel_level 2 frame in a parallel encode set of lower layer
3359 * frames.
3360 */
3361
3362 int wanted_fb;
3363 #endif // CONFIG_FPMT_TEST
3364
3365 /*!
3366 * A flag to indicate frames that will update their data to the primary
3367 * context at the end of the encode. It is set for non-parallel frames and the
3368 * last frame in encode order in a given parallel encode set.
3369 */
3370 bool do_frame_data_update;
3371
3372 #if CONFIG_RD_COMMAND
3373 /*!
3374 * A structure for assigning external q_index / rdmult for experiments
3375 */
3376 RD_COMMAND rd_command;
3377 #endif // CONFIG_RD_COMMAND
3378
3379 /*!
3380 * Buffer to store MB variance after Wiener filter.
3381 */
3382 WeberStats *mb_weber_stats;
3383
3384 /*!
3385 * Buffer to store MB variance after Wiener filter.
3386 */
3387 BLOCK_SIZE weber_bsize;
3388
3389 /*!
3390 * Frame level Wiener filter normalization.
3391 */
3392 int64_t norm_wiener_variance;
3393
3394 /*!
3395 * Buffer to store delta-q values for delta-q mode 4.
3396 */
3397 int *mb_delta_q;
3398
3399 /*!
3400 * Flag to indicate that current frame is dropped.
3401 */
3402 bool is_dropped_frame;
3403
3404 #if CONFIG_BITRATE_ACCURACY
3405 /*!
3406 * Structure stores information needed for bitrate accuracy experiment.
3407 */
3408 VBR_RATECTRL_INFO vbr_rc_info;
3409 #endif
3410
3411 #if CONFIG_RATECTRL_LOG
3412 /*!
3413 * Structure stores information of rate control decisions.
3414 */
3415 RATECTRL_LOG rc_log;
3416 #endif // CONFIG_RATECTRL_LOG
3417
3418 /*!
3419 * Frame level twopass status and control data
3420 */
3421 TWO_PASS_FRAME twopass_frame;
3422
3423 /*!
3424 * Context needed for third pass encoding.
3425 */
3426 THIRD_PASS_DEC_CTX *third_pass_ctx;
3427
3428 /*!
3429 * File pointer to second pass log
3430 */
3431 FILE *second_pass_log_stream;
3432
3433 /*!
3434 * Buffer to store 64x64 SAD
3435 */
3436 uint64_t *src_sad_blk_64x64;
3437
3438 /*!
3439 * SSE between the current frame and the reconstructed last frame
3440 */
3441 uint64_t rec_sse;
3442
3443 /*!
3444 * A flag to indicate whether the encoder is controlled by DuckyEncode or not.
3445 * 1:yes 0:no
3446 */
3447 int use_ducky_encode;
3448
3449 #if !CONFIG_REALTIME_ONLY
3450 /*! A structure that facilitates the communication between DuckyEncode and AV1
3451 * encoder.
3452 */
3453 DuckyEncodeInfo ducky_encode_info;
3454 #endif // CONFIG_REALTIME_ONLY
3455 //
3456 /*!
3457 * Frames since last frame with cdf update.
3458 */
3459 int frames_since_last_update;
3460
3461 /*!
3462 * Block level thresholds to force zeromv-skip at partition level.
3463 */
3464 unsigned int zeromv_skip_thresh_exit_part[BLOCK_SIZES_ALL];
3465 } AV1_COMP;
3466
3467 /*!
3468 * \brief Input frames and last input frame
3469 */
3470 typedef struct EncodeFrameInput {
3471 /*!\cond */
3472 YV12_BUFFER_CONFIG *source;
3473 YV12_BUFFER_CONFIG *last_source;
3474 int64_t ts_duration;
3475 /*!\endcond */
3476 } EncodeFrameInput;
3477
3478 /*!
3479 * \brief contains per-frame encoding parameters decided upon by
3480 * av1_encode_strategy() and passed down to av1_encode().
3481 */
3482 typedef struct EncodeFrameParams {
3483 /*!
3484 * Is error resilient mode enabled
3485 */
3486 int error_resilient_mode;
3487 /*!
3488 * Frame type (eg KF vs inter frame etc)
3489 */
3490 FRAME_TYPE frame_type;
3491
3492 /*!\cond */
3493 int primary_ref_frame;
3494 int order_offset;
3495
3496 /*!\endcond */
3497 /*!
3498 * Should the current frame be displayed after being decoded
3499 */
3500 int show_frame;
3501
3502 /*!\cond */
3503 int refresh_frame_flags;
3504
3505 int show_existing_frame;
3506 int existing_fb_idx_to_show;
3507
3508 /*!\endcond */
3509 /*!
3510 * Bitmask of which reference buffers may be referenced by this frame.
3511 */
3512 int ref_frame_flags;
3513
3514 /*!
3515 * Reference buffer assignment for this frame.
3516 */
3517 int remapped_ref_idx[REF_FRAMES];
3518
3519 /*!
3520 * Flags which determine which reference buffers are refreshed by this
3521 * frame.
3522 */
3523 RefreshFrameInfo refresh_frame;
3524
3525 /*!
3526 * Speed level to use for this frame: Bigger number means faster.
3527 */
3528 int speed;
3529 } EncodeFrameParams;
3530
3531 /*!\cond */
3532
3533 // EncodeFrameResults contains information about the result of encoding a
3534 // single frame
3535 typedef struct {
3536 size_t size; // Size of resulting bitstream
3537 } EncodeFrameResults;
3538
3539 void av1_initialize_enc(unsigned int usage, enum aom_rc_mode end_usage);
3540
3541 struct AV1_COMP *av1_create_compressor(AV1_PRIMARY *ppi,
3542 const AV1EncoderConfig *oxcf,
3543 BufferPool *const pool,
3544 COMPRESSOR_STAGE stage,
3545 int lap_lag_in_frames);
3546
3547 struct AV1_PRIMARY *av1_create_primary_compressor(
3548 struct aom_codec_pkt_list *pkt_list_head, int num_lap_buffers,
3549 const AV1EncoderConfig *oxcf);
3550
3551 void av1_remove_compressor(AV1_COMP *cpi);
3552
3553 void av1_remove_primary_compressor(AV1_PRIMARY *ppi);
3554
3555 #if CONFIG_ENTROPY_STATS
3556 void print_entropy_stats(AV1_PRIMARY *const ppi);
3557 #endif
3558 #if CONFIG_INTERNAL_STATS
3559 void print_internal_stats(AV1_PRIMARY *ppi);
3560 #endif
3561
3562 void av1_change_config_seq(AV1_PRIMARY *ppi, const AV1EncoderConfig *oxcf,
3563 bool *sb_size_changed);
3564
3565 void av1_change_config(AV1_COMP *cpi, const AV1EncoderConfig *oxcf,
3566 bool sb_size_changed);
3567
3568 void av1_check_initial_width(AV1_COMP *cpi, int use_highbitdepth,
3569 int subsampling_x, int subsampling_y);
3570
3571 void av1_init_seq_coding_tools(AV1_PRIMARY *const ppi,
3572 const AV1EncoderConfig *oxcf, int use_svc);
3573
3574 void av1_post_encode_updates(AV1_COMP *const cpi,
3575 const AV1_COMP_DATA *const cpi_data);
3576
3577 void av1_scale_references_fpmt(AV1_COMP *cpi, int *ref_buffers_used_map);
3578
3579 void av1_increment_scaled_ref_counts_fpmt(BufferPool *buffer_pool,
3580 int ref_buffers_used_map);
3581
3582 void av1_release_scaled_references_fpmt(AV1_COMP *cpi);
3583
3584 void av1_decrement_ref_counts_fpmt(BufferPool *buffer_pool,
3585 int ref_buffers_used_map);
3586
3587 void av1_init_sc_decisions(AV1_PRIMARY *const ppi);
3588
3589 AV1_COMP *av1_get_parallel_frame_enc_data(AV1_PRIMARY *const ppi,
3590 AV1_COMP_DATA *const first_cpi_data);
3591
3592 int av1_init_parallel_frame_context(const AV1_COMP_DATA *const first_cpi_data,
3593 AV1_PRIMARY *const ppi,
3594 int *ref_buffers_used_map);
3595 /*!\endcond */
3596
3597 /*!\brief Obtain the raw frame data
3598 *
3599 * \ingroup high_level_algo
3600 * This function receives the raw frame data from input.
3601 *
3602 * \param[in] cpi Top-level encoder structure
3603 * \param[in] frame_flags Flags to decide how to encoding the frame
3604 * \param[in] sd Contain raw frame data
3605 * \param[in] time_stamp Time stamp of the frame
3606 * \param[in] end_time_stamp End time stamp
3607 *
3608 * \return Returns a value to indicate if the frame data is received
3609 * successfully.
3610 * \note The caller can assume that a copy of this frame is made and not just a
3611 * copy of the pointer.
3612 */
3613 int av1_receive_raw_frame(AV1_COMP *cpi, aom_enc_frame_flags_t frame_flags,
3614 YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
3615 int64_t end_time_stamp);
3616
3617 /*!\brief Encode a frame
3618 *
3619 * \ingroup high_level_algo
3620 * \callgraph
3621 * \callergraph
3622 * This function encodes the raw frame data, and outputs the frame bit stream
3623 * to the designated buffer. The caller should use the output parameters
3624 * cpi_data->ts_frame_start and cpi_data->ts_frame_end only when this function
3625 * returns AOM_CODEC_OK.
3626 *
3627 * \param[in] cpi Top-level encoder structure
3628 * \param[in,out] cpi_data Data corresponding to a frame encode
3629 *
3630 * \return Returns a value to indicate if the encoding is done successfully.
3631 * \retval #AOM_CODEC_OK
3632 * \retval -1
3633 * No frame encoded; more input is required.
3634 * \retval #AOM_CODEC_ERROR
3635 */
3636 int av1_get_compressed_data(AV1_COMP *cpi, AV1_COMP_DATA *const cpi_data);
3637
3638 /*!\brief Run 1-pass/2-pass encoding
3639 *
3640 * \ingroup high_level_algo
3641 * \callgraph
3642 * \callergraph
3643 */
3644 int av1_encode(AV1_COMP *const cpi, uint8_t *const dest,
3645 const EncodeFrameInput *const frame_input,
3646 const EncodeFrameParams *const frame_params,
3647 EncodeFrameResults *const frame_results);
3648
3649 /*!\cond */
3650 int av1_get_preview_raw_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *dest);
3651
3652 int av1_get_last_show_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *frame);
3653
3654 aom_codec_err_t av1_copy_new_frame_enc(AV1_COMMON *cm,
3655 YV12_BUFFER_CONFIG *new_frame,
3656 YV12_BUFFER_CONFIG *sd);
3657
3658 int av1_use_as_reference(int *ext_ref_frame_flags, int ref_frame_flags);
3659
3660 int av1_copy_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd);
3661
3662 int av1_set_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd);
3663
3664 int av1_set_size_literal(AV1_COMP *cpi, int width, int height);
3665
3666 void av1_set_frame_size(AV1_COMP *cpi, int width, int height);
3667
3668 void av1_set_mv_search_params(AV1_COMP *cpi);
3669
3670 int av1_set_active_map(AV1_COMP *cpi, unsigned char *map, int rows, int cols);
3671
3672 int av1_get_active_map(AV1_COMP *cpi, unsigned char *map, int rows, int cols);
3673
3674 int av1_set_internal_size(AV1EncoderConfig *const oxcf,
3675 ResizePendingParams *resize_pending_params,
3676 AOM_SCALING_MODE horiz_mode,
3677 AOM_SCALING_MODE vert_mode);
3678
3679 int av1_get_quantizer(struct AV1_COMP *cpi);
3680
3681 int av1_convert_sect5obus_to_annexb(uint8_t *buffer, size_t *input_size);
3682
3683 // Set screen content options.
3684 // This function estimates whether to use screen content tools, by counting
3685 // the portion of blocks that have few luma colors.
3686 // Modifies:
3687 // cpi->commom.features.allow_screen_content_tools
3688 // cpi->common.features.allow_intrabc
3689 // cpi->use_screen_content_tools
3690 // cpi->is_screen_content_type
3691 // However, the estimation is not accurate and may misclassify videos.
3692 // A slower but more accurate approach that determines whether to use screen
3693 // content tools is employed later. See av1_determine_sc_tools_with_encoding().
3694 void av1_set_screen_content_options(struct AV1_COMP *cpi,
3695 FeatureFlags *features);
3696
3697 void av1_update_frame_size(AV1_COMP *cpi);
3698
3699 typedef struct {
3700 int pyr_level;
3701 int disp_order;
3702 } RefFrameMapPair;
3703
init_ref_map_pair(AV1_COMP * cpi,RefFrameMapPair ref_frame_map_pairs[REF_FRAMES])3704 static INLINE void init_ref_map_pair(
3705 AV1_COMP *cpi, RefFrameMapPair ref_frame_map_pairs[REF_FRAMES]) {
3706 if (cpi->ppi->gf_group.update_type[cpi->gf_frame_index] == KF_UPDATE) {
3707 memset(ref_frame_map_pairs, -1, sizeof(*ref_frame_map_pairs) * REF_FRAMES);
3708 return;
3709 }
3710 memset(ref_frame_map_pairs, 0, sizeof(*ref_frame_map_pairs) * REF_FRAMES);
3711 for (int map_idx = 0; map_idx < REF_FRAMES; map_idx++) {
3712 // Get reference frame buffer.
3713 const RefCntBuffer *const buf = cpi->common.ref_frame_map[map_idx];
3714 if (ref_frame_map_pairs[map_idx].disp_order == -1) continue;
3715 if (buf == NULL) {
3716 ref_frame_map_pairs[map_idx].disp_order = -1;
3717 ref_frame_map_pairs[map_idx].pyr_level = -1;
3718 continue;
3719 } else if (buf->ref_count > 1) {
3720 // Once the keyframe is coded, the slots in ref_frame_map will all
3721 // point to the same frame. In that case, all subsequent pointers
3722 // matching the current are considered "free" slots. This will find
3723 // the next occurrence of the current pointer if ref_count indicates
3724 // there are multiple instances of it and mark it as free.
3725 for (int idx2 = map_idx + 1; idx2 < REF_FRAMES; ++idx2) {
3726 const RefCntBuffer *const buf2 = cpi->common.ref_frame_map[idx2];
3727 if (buf2 == buf) {
3728 ref_frame_map_pairs[idx2].disp_order = -1;
3729 ref_frame_map_pairs[idx2].pyr_level = -1;
3730 }
3731 }
3732 }
3733 ref_frame_map_pairs[map_idx].disp_order = (int)buf->display_order_hint;
3734 ref_frame_map_pairs[map_idx].pyr_level = buf->pyramid_level;
3735 }
3736 }
3737
3738 #if CONFIG_FPMT_TEST
calc_frame_data_update_flag(GF_GROUP * const gf_group,int gf_frame_index,bool * const do_frame_data_update)3739 static AOM_INLINE void calc_frame_data_update_flag(
3740 GF_GROUP *const gf_group, int gf_frame_index,
3741 bool *const do_frame_data_update) {
3742 *do_frame_data_update = true;
3743 // Set the flag to false for all frames in a given parallel encode set except
3744 // the last frame in the set with frame_parallel_level = 2.
3745 if (gf_group->frame_parallel_level[gf_frame_index] == 1) {
3746 *do_frame_data_update = false;
3747 } else if (gf_group->frame_parallel_level[gf_frame_index] == 2) {
3748 // Check if this is the last frame in the set with frame_parallel_level = 2.
3749 for (int i = gf_frame_index + 1; i < gf_group->size; i++) {
3750 if ((gf_group->frame_parallel_level[i] == 0 &&
3751 (gf_group->update_type[i] == ARF_UPDATE ||
3752 gf_group->update_type[i] == INTNL_ARF_UPDATE)) ||
3753 gf_group->frame_parallel_level[i] == 1) {
3754 break;
3755 } else if (gf_group->frame_parallel_level[i] == 2) {
3756 *do_frame_data_update = false;
3757 break;
3758 }
3759 }
3760 }
3761 }
3762 #endif
3763
3764 // av1 uses 10,000,000 ticks/second as time stamp
3765 #define TICKS_PER_SEC 10000000LL
3766
3767 static INLINE int64_t
timebase_units_to_ticks(const aom_rational64_t * timestamp_ratio,int64_t n)3768 timebase_units_to_ticks(const aom_rational64_t *timestamp_ratio, int64_t n) {
3769 return n * timestamp_ratio->num / timestamp_ratio->den;
3770 }
3771
3772 static INLINE int64_t
ticks_to_timebase_units(const aom_rational64_t * timestamp_ratio,int64_t n)3773 ticks_to_timebase_units(const aom_rational64_t *timestamp_ratio, int64_t n) {
3774 int64_t round = timestamp_ratio->num / 2;
3775 if (round > 0) --round;
3776 return (n * timestamp_ratio->den + round) / timestamp_ratio->num;
3777 }
3778
frame_is_kf_gf_arf(const AV1_COMP * cpi)3779 static INLINE int frame_is_kf_gf_arf(const AV1_COMP *cpi) {
3780 const GF_GROUP *const gf_group = &cpi->ppi->gf_group;
3781 const FRAME_UPDATE_TYPE update_type =
3782 gf_group->update_type[cpi->gf_frame_index];
3783
3784 return frame_is_intra_only(&cpi->common) || update_type == ARF_UPDATE ||
3785 update_type == GF_UPDATE;
3786 }
3787
3788 // TODO(huisu@google.com, youzhou@microsoft.com): enable hash-me for HBD.
av1_use_hash_me(const AV1_COMP * const cpi)3789 static INLINE int av1_use_hash_me(const AV1_COMP *const cpi) {
3790 return (cpi->common.features.allow_screen_content_tools &&
3791 cpi->common.features.allow_intrabc &&
3792 frame_is_intra_only(&cpi->common));
3793 }
3794
get_ref_frame_yv12_buf(const AV1_COMMON * const cm,MV_REFERENCE_FRAME ref_frame)3795 static INLINE const YV12_BUFFER_CONFIG *get_ref_frame_yv12_buf(
3796 const AV1_COMMON *const cm, MV_REFERENCE_FRAME ref_frame) {
3797 const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame);
3798 return buf != NULL ? &buf->buf : NULL;
3799 }
3800
alloc_frame_mvs(AV1_COMMON * const cm,RefCntBuffer * buf)3801 static INLINE void alloc_frame_mvs(AV1_COMMON *const cm, RefCntBuffer *buf) {
3802 assert(buf != NULL);
3803 ensure_mv_buffer(buf, cm);
3804 buf->width = cm->width;
3805 buf->height = cm->height;
3806 }
3807
3808 // Get the allocated token size for a tile. It does the same calculation as in
3809 // the frame token allocation.
allocated_tokens(const TileInfo * tile,int sb_size_log2,int num_planes)3810 static INLINE unsigned int allocated_tokens(const TileInfo *tile,
3811 int sb_size_log2, int num_planes) {
3812 int tile_mb_rows =
3813 ROUND_POWER_OF_TWO(tile->mi_row_end - tile->mi_row_start, 2);
3814 int tile_mb_cols =
3815 ROUND_POWER_OF_TWO(tile->mi_col_end - tile->mi_col_start, 2);
3816
3817 return get_token_alloc(tile_mb_rows, tile_mb_cols, sb_size_log2, num_planes);
3818 }
3819
get_start_tok(AV1_COMP * cpi,int tile_row,int tile_col,int mi_row,TokenExtra ** tok,int sb_size_log2,int num_planes)3820 static INLINE void get_start_tok(AV1_COMP *cpi, int tile_row, int tile_col,
3821 int mi_row, TokenExtra **tok, int sb_size_log2,
3822 int num_planes) {
3823 AV1_COMMON *const cm = &cpi->common;
3824 const int tile_cols = cm->tiles.cols;
3825 TileDataEnc *this_tile = &cpi->tile_data[tile_row * tile_cols + tile_col];
3826 const TileInfo *const tile_info = &this_tile->tile_info;
3827
3828 const int tile_mb_cols =
3829 (tile_info->mi_col_end - tile_info->mi_col_start + 2) >> 2;
3830 const int tile_mb_row = (mi_row - tile_info->mi_row_start + 2) >> 2;
3831
3832 *tok = cpi->token_info.tile_tok[tile_row][tile_col] +
3833 get_token_alloc(tile_mb_row, tile_mb_cols, sb_size_log2, num_planes);
3834 }
3835
3836 void av1_apply_encoding_flags(AV1_COMP *cpi, aom_enc_frame_flags_t flags);
3837
3838 #define ALT_MIN_LAG 3
is_altref_enabled(int lag_in_frames,bool enable_auto_arf)3839 static INLINE int is_altref_enabled(int lag_in_frames, bool enable_auto_arf) {
3840 return lag_in_frames >= ALT_MIN_LAG && enable_auto_arf;
3841 }
3842
can_disable_altref(const GFConfig * gf_cfg)3843 static AOM_INLINE int can_disable_altref(const GFConfig *gf_cfg) {
3844 return is_altref_enabled(gf_cfg->lag_in_frames, gf_cfg->enable_auto_arf) &&
3845 (gf_cfg->gf_min_pyr_height == 0);
3846 }
3847
3848 // Helper function to compute number of blocks on either side of the frame.
get_num_blocks(const int frame_length,const int mb_length)3849 static INLINE int get_num_blocks(const int frame_length, const int mb_length) {
3850 return (frame_length + mb_length - 1) / mb_length;
3851 }
3852
3853 // Check if statistics generation stage
is_stat_generation_stage(const AV1_COMP * const cpi)3854 static INLINE int is_stat_generation_stage(const AV1_COMP *const cpi) {
3855 assert(IMPLIES(cpi->compressor_stage == LAP_STAGE,
3856 cpi->oxcf.pass == AOM_RC_ONE_PASS && cpi->ppi->lap_enabled));
3857 return (cpi->oxcf.pass == AOM_RC_FIRST_PASS ||
3858 (cpi->compressor_stage == LAP_STAGE));
3859 }
3860 // Check if statistics consumption stage
is_stat_consumption_stage_twopass(const AV1_COMP * const cpi)3861 static INLINE int is_stat_consumption_stage_twopass(const AV1_COMP *const cpi) {
3862 return (cpi->oxcf.pass >= AOM_RC_SECOND_PASS);
3863 }
3864
3865 // Check if statistics consumption stage
is_stat_consumption_stage(const AV1_COMP * const cpi)3866 static INLINE int is_stat_consumption_stage(const AV1_COMP *const cpi) {
3867 return (is_stat_consumption_stage_twopass(cpi) ||
3868 (cpi->oxcf.pass == AOM_RC_ONE_PASS &&
3869 (cpi->compressor_stage == ENCODE_STAGE) && cpi->ppi->lap_enabled));
3870 }
3871
3872 // Decide whether 'dv_costs' need to be allocated/stored during the encoding.
av1_need_dv_costs(const AV1_COMP * const cpi)3873 static AOM_INLINE bool av1_need_dv_costs(const AV1_COMP *const cpi) {
3874 return !cpi->sf.rt_sf.use_nonrd_pick_mode &&
3875 av1_allow_intrabc(&cpi->common) && !is_stat_generation_stage(cpi);
3876 }
3877
3878 /*!\endcond */
3879 /*!\brief Check if the current stage has statistics
3880 *
3881 *\ingroup two_pass_algo
3882 *
3883 * \param[in] cpi Top - level encoder instance structure
3884 *
3885 * \return 0 if no stats for current stage else 1
3886 */
has_no_stats_stage(const AV1_COMP * const cpi)3887 static INLINE int has_no_stats_stage(const AV1_COMP *const cpi) {
3888 assert(
3889 IMPLIES(!cpi->ppi->lap_enabled, cpi->compressor_stage == ENCODE_STAGE));
3890 return (cpi->oxcf.pass == AOM_RC_ONE_PASS && !cpi->ppi->lap_enabled);
3891 }
3892
3893 /*!\cond */
3894
is_one_pass_rt_params(const AV1_COMP * cpi)3895 static INLINE int is_one_pass_rt_params(const AV1_COMP *cpi) {
3896 return has_no_stats_stage(cpi) && cpi->oxcf.mode == REALTIME &&
3897 cpi->oxcf.gf_cfg.lag_in_frames == 0;
3898 }
3899
3900 // Use default/internal reference structure for single-layer RTC.
use_rtc_reference_structure_one_layer(const AV1_COMP * cpi)3901 static INLINE int use_rtc_reference_structure_one_layer(const AV1_COMP *cpi) {
3902 return is_one_pass_rt_params(cpi) && cpi->ppi->number_spatial_layers == 1 &&
3903 cpi->ppi->number_temporal_layers == 1 &&
3904 !cpi->ppi->rtc_ref.set_ref_frame_config;
3905 }
3906
3907 // Function return size of frame stats buffer
get_stats_buf_size(int num_lap_buffer,int num_lag_buffer)3908 static INLINE int get_stats_buf_size(int num_lap_buffer, int num_lag_buffer) {
3909 /* if lookahead is enabled return num_lap_buffers else num_lag_buffers */
3910 return (num_lap_buffer > 0 ? num_lap_buffer + 1 : num_lag_buffer);
3911 }
3912
3913 // TODO(zoeliu): To set up cpi->oxcf.gf_cfg.enable_auto_brf
3914
set_ref_ptrs(const AV1_COMMON * cm,MACROBLOCKD * xd,MV_REFERENCE_FRAME ref0,MV_REFERENCE_FRAME ref1)3915 static INLINE void set_ref_ptrs(const AV1_COMMON *cm, MACROBLOCKD *xd,
3916 MV_REFERENCE_FRAME ref0,
3917 MV_REFERENCE_FRAME ref1) {
3918 xd->block_ref_scale_factors[0] =
3919 get_ref_scale_factors_const(cm, ref0 >= LAST_FRAME ? ref0 : 1);
3920 xd->block_ref_scale_factors[1] =
3921 get_ref_scale_factors_const(cm, ref1 >= LAST_FRAME ? ref1 : 1);
3922 }
3923
get_chessboard_index(int frame_index)3924 static INLINE int get_chessboard_index(int frame_index) {
3925 return frame_index & 0x1;
3926 }
3927
cond_cost_list_const(const struct AV1_COMP * cpi,const int * cost_list)3928 static INLINE const int *cond_cost_list_const(const struct AV1_COMP *cpi,
3929 const int *cost_list) {
3930 const int use_cost_list = cpi->sf.mv_sf.subpel_search_method != SUBPEL_TREE &&
3931 cpi->sf.mv_sf.use_fullpel_costlist;
3932 return use_cost_list ? cost_list : NULL;
3933 }
3934
cond_cost_list(const struct AV1_COMP * cpi,int * cost_list)3935 static INLINE int *cond_cost_list(const struct AV1_COMP *cpi, int *cost_list) {
3936 const int use_cost_list = cpi->sf.mv_sf.subpel_search_method != SUBPEL_TREE &&
3937 cpi->sf.mv_sf.use_fullpel_costlist;
3938 return use_cost_list ? cost_list : NULL;
3939 }
3940
3941 // Compression ratio of current frame.
3942 double av1_get_compression_ratio(const AV1_COMMON *const cm,
3943 size_t encoded_frame_size);
3944
3945 void av1_new_framerate(AV1_COMP *cpi, double framerate);
3946
3947 void av1_setup_frame_size(AV1_COMP *cpi);
3948
3949 #define LAYER_IDS_TO_IDX(sl, tl, num_tl) ((sl) * (num_tl) + (tl))
3950
3951 // Returns 1 if a frame is scaled and 0 otherwise.
av1_resize_scaled(const AV1_COMMON * cm)3952 static INLINE int av1_resize_scaled(const AV1_COMMON *cm) {
3953 return cm->superres_upscaled_width != cm->render_width ||
3954 cm->superres_upscaled_height != cm->render_height;
3955 }
3956
av1_frame_scaled(const AV1_COMMON * cm)3957 static INLINE int av1_frame_scaled(const AV1_COMMON *cm) {
3958 return av1_superres_scaled(cm) || av1_resize_scaled(cm);
3959 }
3960
3961 // Don't allow a show_existing_frame to coincide with an error resilient
3962 // frame. An exception can be made for a forward keyframe since it has no
3963 // previous dependencies.
encode_show_existing_frame(const AV1_COMMON * cm)3964 static INLINE int encode_show_existing_frame(const AV1_COMMON *cm) {
3965 return cm->show_existing_frame && (!cm->features.error_resilient_mode ||
3966 cm->current_frame.frame_type == KEY_FRAME);
3967 }
3968
3969 // Get index into the 'cpi->mbmi_ext_info.frame_base' array for the given
3970 // 'mi_row' and 'mi_col'.
get_mi_ext_idx(const int mi_row,const int mi_col,const BLOCK_SIZE mi_alloc_bsize,const int mbmi_ext_stride)3971 static INLINE int get_mi_ext_idx(const int mi_row, const int mi_col,
3972 const BLOCK_SIZE mi_alloc_bsize,
3973 const int mbmi_ext_stride) {
3974 const int mi_ext_size_1d = mi_size_wide[mi_alloc_bsize];
3975 const int mi_ext_row = mi_row / mi_ext_size_1d;
3976 const int mi_ext_col = mi_col / mi_ext_size_1d;
3977 return mi_ext_row * mbmi_ext_stride + mi_ext_col;
3978 }
3979
3980 // Lighter version of set_offsets that only sets the mode info
3981 // pointers.
set_mode_info_offsets(const CommonModeInfoParams * const mi_params,const MBMIExtFrameBufferInfo * const mbmi_ext_info,MACROBLOCK * const x,MACROBLOCKD * const xd,int mi_row,int mi_col)3982 static INLINE void set_mode_info_offsets(
3983 const CommonModeInfoParams *const mi_params,
3984 const MBMIExtFrameBufferInfo *const mbmi_ext_info, MACROBLOCK *const x,
3985 MACROBLOCKD *const xd, int mi_row, int mi_col) {
3986 set_mi_offsets(mi_params, xd, mi_row, mi_col);
3987 const int ext_idx = get_mi_ext_idx(mi_row, mi_col, mi_params->mi_alloc_bsize,
3988 mbmi_ext_info->stride);
3989 x->mbmi_ext_frame = mbmi_ext_info->frame_base + ext_idx;
3990 }
3991
3992 // Check to see if the given partition size is allowed for a specified number
3993 // of mi block rows and columns remaining in the image.
3994 // If not then return the largest allowed partition size
find_partition_size(BLOCK_SIZE bsize,int rows_left,int cols_left,int * bh,int * bw)3995 static INLINE BLOCK_SIZE find_partition_size(BLOCK_SIZE bsize, int rows_left,
3996 int cols_left, int *bh, int *bw) {
3997 int int_size = (int)bsize;
3998 if (rows_left <= 0 || cols_left <= 0) {
3999 return AOMMIN(bsize, BLOCK_8X8);
4000 } else {
4001 for (; int_size > 0; int_size -= 3) {
4002 *bh = mi_size_high[int_size];
4003 *bw = mi_size_wide[int_size];
4004 if ((*bh <= rows_left) && (*bw <= cols_left)) {
4005 break;
4006 }
4007 }
4008 }
4009 return (BLOCK_SIZE)int_size;
4010 }
4011
4012 static const uint8_t av1_ref_frame_flag_list[REF_FRAMES] = { 0,
4013 AOM_LAST_FLAG,
4014 AOM_LAST2_FLAG,
4015 AOM_LAST3_FLAG,
4016 AOM_GOLD_FLAG,
4017 AOM_BWD_FLAG,
4018 AOM_ALT2_FLAG,
4019 AOM_ALT_FLAG };
4020
4021 // When more than 'max_allowed_refs' are available, we reduce the number of
4022 // reference frames one at a time based on this order.
4023 static const MV_REFERENCE_FRAME disable_order[] = {
4024 LAST3_FRAME,
4025 LAST2_FRAME,
4026 ALTREF2_FRAME,
4027 BWDREF_FRAME,
4028 };
4029
4030 static const MV_REFERENCE_FRAME
4031 ref_frame_priority_order[INTER_REFS_PER_FRAME] = {
4032 LAST_FRAME, ALTREF_FRAME, BWDREF_FRAME, GOLDEN_FRAME,
4033 ALTREF2_FRAME, LAST2_FRAME, LAST3_FRAME,
4034 };
4035
get_ref_frame_flags(const SPEED_FEATURES * const sf,const int use_one_pass_rt_params,const YV12_BUFFER_CONFIG ** ref_frames,const int ext_ref_frame_flags)4036 static INLINE int get_ref_frame_flags(const SPEED_FEATURES *const sf,
4037 const int use_one_pass_rt_params,
4038 const YV12_BUFFER_CONFIG **ref_frames,
4039 const int ext_ref_frame_flags) {
4040 // cpi->ext_flags.ref_frame_flags allows certain reference types to be
4041 // disabled by the external interface. These are set by
4042 // av1_apply_encoding_flags(). Start with what the external interface allows,
4043 // then suppress any reference types which we have found to be duplicates.
4044 int flags = ext_ref_frame_flags;
4045
4046 for (int i = 1; i < INTER_REFS_PER_FRAME; ++i) {
4047 const YV12_BUFFER_CONFIG *const this_ref = ref_frames[i];
4048 // If this_ref has appeared before, mark the corresponding ref frame as
4049 // invalid. For one_pass_rt mode, only disable GOLDEN_FRAME if it's the
4050 // same as LAST_FRAME or ALTREF_FRAME (if ALTREF is being used in nonrd).
4051 int index =
4052 (use_one_pass_rt_params && ref_frame_priority_order[i] == GOLDEN_FRAME)
4053 ? (1 + sf->rt_sf.use_nonrd_altref_frame)
4054 : i;
4055 for (int j = 0; j < index; ++j) {
4056 // If this_ref has appeared before (same as the reference corresponding
4057 // to lower index j), remove it as a reference only if that reference
4058 // (for index j) is actually used as a reference.
4059 if (this_ref == ref_frames[j] &&
4060 (flags & (1 << (ref_frame_priority_order[j] - 1)))) {
4061 flags &= ~(1 << (ref_frame_priority_order[i] - 1));
4062 break;
4063 }
4064 }
4065 }
4066 return flags;
4067 }
4068
4069 // Returns a Sequence Header OBU stored in an aom_fixed_buf_t, or NULL upon
4070 // failure. When a non-NULL aom_fixed_buf_t pointer is returned by this
4071 // function, the memory must be freed by the caller. Both the buf member of the
4072 // aom_fixed_buf_t, and the aom_fixed_buf_t pointer itself must be freed. Memory
4073 // returned must be freed via call to free().
4074 //
4075 // Note: The OBU returned is in Low Overhead Bitstream Format. Specifically,
4076 // the obu_has_size_field bit is set, and the buffer contains the obu_size
4077 // field.
4078 aom_fixed_buf_t *av1_get_global_headers(AV1_PRIMARY *ppi);
4079
4080 #define MAX_GFUBOOST_FACTOR 10.0
4081 #define MIN_GFUBOOST_FACTOR 4.0
4082
is_frame_tpl_eligible(const GF_GROUP * const gf_group,uint8_t index)4083 static INLINE int is_frame_tpl_eligible(const GF_GROUP *const gf_group,
4084 uint8_t index) {
4085 const FRAME_UPDATE_TYPE update_type = gf_group->update_type[index];
4086 return update_type == ARF_UPDATE || update_type == GF_UPDATE ||
4087 update_type == KF_UPDATE;
4088 }
4089
is_frame_eligible_for_ref_pruning(const GF_GROUP * gf_group,int selective_ref_frame,int prune_ref_frames,int gf_index)4090 static INLINE int is_frame_eligible_for_ref_pruning(const GF_GROUP *gf_group,
4091 int selective_ref_frame,
4092 int prune_ref_frames,
4093 int gf_index) {
4094 return (selective_ref_frame > 0) && (prune_ref_frames > 0) &&
4095 !is_frame_tpl_eligible(gf_group, gf_index);
4096 }
4097
4098 // Get update type of the current frame.
get_frame_update_type(const GF_GROUP * gf_group,int gf_frame_index)4099 static INLINE FRAME_UPDATE_TYPE get_frame_update_type(const GF_GROUP *gf_group,
4100 int gf_frame_index) {
4101 return gf_group->update_type[gf_frame_index];
4102 }
4103
av1_pixels_to_mi(int pixels)4104 static INLINE int av1_pixels_to_mi(int pixels) {
4105 return ALIGN_POWER_OF_TWO(pixels, 3) >> MI_SIZE_LOG2;
4106 }
4107
is_psnr_calc_enabled(const AV1_COMP * cpi)4108 static AOM_INLINE int is_psnr_calc_enabled(const AV1_COMP *cpi) {
4109 const AV1_COMMON *const cm = &cpi->common;
4110
4111 return cpi->ppi->b_calculate_psnr && !is_stat_generation_stage(cpi) &&
4112 cm->show_frame;
4113 }
4114
is_frame_resize_pending(const AV1_COMP * const cpi)4115 static INLINE int is_frame_resize_pending(const AV1_COMP *const cpi) {
4116 const ResizePendingParams *const resize_pending_params =
4117 &cpi->resize_pending_params;
4118 return (resize_pending_params->width && resize_pending_params->height &&
4119 (cpi->common.width != resize_pending_params->width ||
4120 cpi->common.height != resize_pending_params->height));
4121 }
4122
4123 // Check if loop filter is used.
is_loopfilter_used(const AV1_COMMON * const cm)4124 static INLINE int is_loopfilter_used(const AV1_COMMON *const cm) {
4125 return !cm->features.coded_lossless && !cm->tiles.large_scale;
4126 }
4127
4128 // Check if CDEF is used.
is_cdef_used(const AV1_COMMON * const cm)4129 static INLINE int is_cdef_used(const AV1_COMMON *const cm) {
4130 return cm->seq_params->enable_cdef && !cm->features.coded_lossless &&
4131 !cm->tiles.large_scale;
4132 }
4133
4134 // Check if loop restoration filter is used.
is_restoration_used(const AV1_COMMON * const cm)4135 static INLINE int is_restoration_used(const AV1_COMMON *const cm) {
4136 return cm->seq_params->enable_restoration && !cm->features.all_lossless &&
4137 !cm->tiles.large_scale;
4138 }
4139
4140 // Checks if post-processing filters need to be applied.
4141 // NOTE: This function decides if the application of different post-processing
4142 // filters on the reconstructed frame can be skipped at the encoder side.
4143 // However the computation of different filter parameters that are signaled in
4144 // the bitstream is still required.
derive_skip_apply_postproc_filters(const AV1_COMP * cpi,int use_loopfilter,int use_cdef,int use_superres,int use_restoration)4145 static INLINE unsigned int derive_skip_apply_postproc_filters(
4146 const AV1_COMP *cpi, int use_loopfilter, int use_cdef, int use_superres,
4147 int use_restoration) {
4148 // Though CDEF parameter selection should be dependent on
4149 // deblocked/loop-filtered pixels for cdef_pick_method <=
4150 // CDEF_FAST_SEARCH_LVL5, CDEF strength values are calculated based on the
4151 // pixel values that are not loop-filtered in svc real-time encoding mode.
4152 // Hence this case is handled separately using the condition below.
4153 if (cpi->ppi->rtc_ref.non_reference_frame)
4154 return (SKIP_APPLY_LOOPFILTER | SKIP_APPLY_CDEF);
4155
4156 if (!cpi->oxcf.algo_cfg.skip_postproc_filtering || cpi->ppi->b_calculate_psnr)
4157 return 0;
4158 assert(cpi->oxcf.mode == ALLINTRA);
4159
4160 // The post-processing filters are applied one after the other in the
4161 // following order: deblocking->cdef->superres->restoration. In case of
4162 // ALLINTRA encoding, the reconstructed frame is not used as a reference
4163 // frame. Hence, the application of these filters can be skipped when
4164 // 1. filter parameters of the subsequent stages are not dependent on the
4165 // filtered output of the current stage or
4166 // 2. subsequent filtering stages are disabled
4167 if (use_restoration) return SKIP_APPLY_RESTORATION;
4168 if (use_superres) return SKIP_APPLY_SUPERRES;
4169 if (use_cdef) {
4170 // CDEF parameter selection is not dependent on the deblocked frame if
4171 // cdef_pick_method is CDEF_PICK_FROM_Q. Hence the application of deblocking
4172 // filters and cdef filters can be skipped in this case.
4173 return (cpi->sf.lpf_sf.cdef_pick_method == CDEF_PICK_FROM_Q &&
4174 use_loopfilter)
4175 ? (SKIP_APPLY_LOOPFILTER | SKIP_APPLY_CDEF)
4176 : SKIP_APPLY_CDEF;
4177 }
4178 if (use_loopfilter) return SKIP_APPLY_LOOPFILTER;
4179
4180 return 0; // All post-processing stages disabled.
4181 }
4182
set_postproc_filter_default_params(AV1_COMMON * cm)4183 static INLINE void set_postproc_filter_default_params(AV1_COMMON *cm) {
4184 struct loopfilter *const lf = &cm->lf;
4185 CdefInfo *const cdef_info = &cm->cdef_info;
4186 RestorationInfo *const rst_info = cm->rst_info;
4187
4188 lf->filter_level[0] = 0;
4189 lf->filter_level[1] = 0;
4190 cdef_info->cdef_bits = 0;
4191 cdef_info->cdef_strengths[0] = 0;
4192 cdef_info->nb_cdef_strengths = 1;
4193 cdef_info->cdef_uv_strengths[0] = 0;
4194 rst_info[0].frame_restoration_type = RESTORE_NONE;
4195 rst_info[1].frame_restoration_type = RESTORE_NONE;
4196 rst_info[2].frame_restoration_type = RESTORE_NONE;
4197 }
4198
is_inter_tx_size_search_level_one(const TX_SPEED_FEATURES * tx_sf)4199 static INLINE int is_inter_tx_size_search_level_one(
4200 const TX_SPEED_FEATURES *tx_sf) {
4201 return (tx_sf->inter_tx_size_search_init_depth_rect >= 1 &&
4202 tx_sf->inter_tx_size_search_init_depth_sqr >= 1);
4203 }
4204
get_lpf_opt_level(const SPEED_FEATURES * sf)4205 static INLINE int get_lpf_opt_level(const SPEED_FEATURES *sf) {
4206 int lpf_opt_level = 0;
4207 if (is_inter_tx_size_search_level_one(&sf->tx_sf))
4208 lpf_opt_level = (sf->lpf_sf.lpf_pick == LPF_PICK_FROM_Q) ? 2 : 1;
4209 return lpf_opt_level;
4210 }
4211
4212 // Enable switchable motion mode only if warp and OBMC tools are allowed
is_switchable_motion_mode_allowed(bool allow_warped_motion,bool enable_obmc)4213 static INLINE bool is_switchable_motion_mode_allowed(bool allow_warped_motion,
4214 bool enable_obmc) {
4215 return (allow_warped_motion || enable_obmc);
4216 }
4217
4218 #if CONFIG_AV1_TEMPORAL_DENOISING
denoise_svc(const struct AV1_COMP * const cpi)4219 static INLINE int denoise_svc(const struct AV1_COMP *const cpi) {
4220 return (!cpi->ppi->use_svc ||
4221 (cpi->ppi->use_svc &&
4222 cpi->svc.spatial_layer_id >= cpi->svc.first_layer_denoise));
4223 }
4224 #endif
4225
4226 #if CONFIG_COLLECT_PARTITION_STATS == 2
av1_print_fr_partition_timing_stats(const FramePartitionTimingStats * part_stats,const char * filename)4227 static INLINE void av1_print_fr_partition_timing_stats(
4228 const FramePartitionTimingStats *part_stats, const char *filename) {
4229 FILE *f = fopen(filename, "w");
4230 if (!f) {
4231 return;
4232 }
4233
4234 fprintf(f, "bsize,redo,");
4235 for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
4236 fprintf(f, "decision_%d,", part);
4237 }
4238 for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
4239 fprintf(f, "attempt_%d,", part);
4240 }
4241 for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
4242 fprintf(f, "time_%d,", part);
4243 }
4244 fprintf(f, "\n");
4245
4246 static const int bsizes[6] = { 128, 64, 32, 16, 8, 4 };
4247
4248 for (int bsize_idx = 0; bsize_idx < 6; bsize_idx++) {
4249 fprintf(f, "%d,%d,", bsizes[bsize_idx], part_stats->partition_redo);
4250 for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
4251 fprintf(f, "%d,", part_stats->partition_decisions[bsize_idx][part]);
4252 }
4253 for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
4254 fprintf(f, "%d,", part_stats->partition_attempts[bsize_idx][part]);
4255 }
4256 for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
4257 fprintf(f, "%ld,", part_stats->partition_times[bsize_idx][part]);
4258 }
4259 fprintf(f, "\n");
4260 }
4261 fclose(f);
4262 }
4263 #endif // CONFIG_COLLECT_PARTITION_STATS == 2
4264
4265 #if CONFIG_COLLECT_PARTITION_STATS
av1_get_bsize_idx_for_part_stats(BLOCK_SIZE bsize)4266 static INLINE int av1_get_bsize_idx_for_part_stats(BLOCK_SIZE bsize) {
4267 assert(bsize == BLOCK_128X128 || bsize == BLOCK_64X64 ||
4268 bsize == BLOCK_32X32 || bsize == BLOCK_16X16 || bsize == BLOCK_8X8 ||
4269 bsize == BLOCK_4X4);
4270 switch (bsize) {
4271 case BLOCK_128X128: return 0;
4272 case BLOCK_64X64: return 1;
4273 case BLOCK_32X32: return 2;
4274 case BLOCK_16X16: return 3;
4275 case BLOCK_8X8: return 4;
4276 case BLOCK_4X4: return 5;
4277 default: assert(0 && "Invalid bsize for partition_stats."); return -1;
4278 }
4279 }
4280 #endif // CONFIG_COLLECT_PARTITION_STATS
4281
4282 #if CONFIG_COLLECT_COMPONENT_TIMING
start_timing(AV1_COMP * cpi,int component)4283 static INLINE void start_timing(AV1_COMP *cpi, int component) {
4284 aom_usec_timer_start(&cpi->component_timer[component]);
4285 }
end_timing(AV1_COMP * cpi,int component)4286 static INLINE void end_timing(AV1_COMP *cpi, int component) {
4287 aom_usec_timer_mark(&cpi->component_timer[component]);
4288 cpi->frame_component_time[component] +=
4289 aom_usec_timer_elapsed(&cpi->component_timer[component]);
4290 }
get_frame_type_enum(int type)4291 static INLINE char const *get_frame_type_enum(int type) {
4292 switch (type) {
4293 case 0: return "KEY_FRAME";
4294 case 1: return "INTER_FRAME";
4295 case 2: return "INTRA_ONLY_FRAME";
4296 case 3: return "S_FRAME";
4297 default: assert(0);
4298 }
4299 return "error";
4300 }
4301 #endif
4302
4303 /*!\endcond */
4304
4305 #ifdef __cplusplus
4306 } // extern "C"
4307 #endif
4308
4309 #endif // AOM_AV1_ENCODER_ENCODER_H_
4310