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 #include "aom_util/aom_pthread.h"
25
26 #include "av1/common/alloccommon.h"
27 #include "av1/common/av1_common_int.h"
28 #include "av1/common/blockd.h"
29 #include "av1/common/entropymode.h"
30 #include "av1/common/enums.h"
31 #include "av1/common/reconintra.h"
32 #include "av1/common/resize.h"
33 #include "av1/common/thread_common.h"
34 #include "av1/common/timing.h"
35
36 #include "av1/encoder/aq_cyclicrefresh.h"
37 #include "av1/encoder/av1_quantize.h"
38 #include "av1/encoder/block.h"
39 #include "av1/encoder/context_tree.h"
40 #include "av1/encoder/enc_enums.h"
41 #include "av1/encoder/encodemb.h"
42 #include "av1/encoder/external_partition.h"
43 #include "av1/encoder/firstpass.h"
44 #include "av1/encoder/global_motion.h"
45 #include "av1/encoder/level.h"
46 #include "av1/encoder/lookahead.h"
47 #include "av1/encoder/mcomp.h"
48 #include "av1/encoder/pickcdef.h"
49 #include "av1/encoder/ratectrl.h"
50 #include "av1/encoder/rd.h"
51 #include "av1/encoder/speed_features.h"
52 #include "av1/encoder/svc_layercontext.h"
53 #include "av1/encoder/temporal_filter.h"
54 #if CONFIG_THREE_PASS
55 #include "av1/encoder/thirdpass.h"
56 #endif
57 #include "av1/encoder/tokenize.h"
58 #include "av1/encoder/tpl_model.h"
59 #include "av1/encoder/av1_noise_estimate.h"
60 #include "av1/encoder/bitstream.h"
61
62 #if CONFIG_INTERNAL_STATS
63 #include "aom_dsp/ssim.h"
64 #endif
65 #include "aom_dsp/variance.h"
66 #if CONFIG_DENOISE
67 #include "aom_dsp/noise_model.h"
68 #endif
69 #if CONFIG_TUNE_VMAF
70 #include "av1/encoder/tune_vmaf.h"
71 #endif
72 #if CONFIG_AV1_TEMPORAL_DENOISING
73 #include "av1/encoder/av1_temporal_denoiser.h"
74 #endif
75 #if CONFIG_TUNE_BUTTERAUGLI
76 #include "av1/encoder/tune_butteraugli.h"
77 #endif
78
79 #include "aom/internal/aom_codec_internal.h"
80
81 #ifdef __cplusplus
82 extern "C" {
83 #endif
84
85 // TODO(yunqing, any): Added suppression tag to quiet Doxygen warnings. Need to
86 // adjust it while we work on documentation.
87 /*!\cond */
88 // Number of frames required to test for scene cut detection
89 #define SCENE_CUT_KEY_TEST_INTERVAL 16
90
91 // Lookahead index threshold to enable temporal filtering for second arf.
92 #define TF_LOOKAHEAD_IDX_THR 7
93
94 #define HDR_QP_LEVELS 10
95 #define CHROMA_CB_QP_SCALE 1.04
96 #define CHROMA_CR_QP_SCALE 1.04
97 #define CHROMA_QP_SCALE -0.46
98 #define CHROMA_QP_OFFSET 9.26
99 #define QP_SCALE_FACTOR 2.0
100 #define DISABLE_HDR_LUMA_DELTAQ 1
101
102 // Rational number with an int64 numerator
103 // This structure holds a fractional value
104 typedef struct aom_rational64 {
105 int64_t num; // fraction numerator
106 int den; // fraction denominator
107 } aom_rational64_t; // alias for struct aom_rational
108
109 enum {
110 // Good Quality Fast Encoding. The encoder balances quality with the amount of
111 // time it takes to encode the output. Speed setting controls how fast.
112 GOOD,
113 // Realtime Fast Encoding. Will force some restrictions on bitrate
114 // constraints.
115 REALTIME,
116 // All intra mode. All the frames are coded as intra frames.
117 ALLINTRA
118 } UENUM1BYTE(MODE);
119
120 enum {
121 FRAMEFLAGS_KEY = 1 << 0,
122 FRAMEFLAGS_GOLDEN = 1 << 1,
123 FRAMEFLAGS_BWDREF = 1 << 2,
124 // TODO(zoeliu): To determine whether a frame flag is needed for ALTREF2_FRAME
125 FRAMEFLAGS_ALTREF = 1 << 3,
126 FRAMEFLAGS_INTRAONLY = 1 << 4,
127 FRAMEFLAGS_SWITCH = 1 << 5,
128 FRAMEFLAGS_ERROR_RESILIENT = 1 << 6,
129 } UENUM1BYTE(FRAMETYPE_FLAGS);
130
131 #if CONFIG_FPMT_TEST
132 enum {
133 PARALLEL_ENCODE = 0,
134 PARALLEL_SIMULATION_ENCODE,
135 NUM_FPMT_TEST_ENCODES
136 } UENUM1BYTE(FPMT_TEST_ENC_CFG);
137 #endif // CONFIG_FPMT_TEST
138 // 0 level frames are sometimes used for rate control purposes, but for
139 // reference mapping purposes, the minimum level should be 1.
140 #define MIN_PYR_LEVEL 1
get_true_pyr_level(int frame_level,int frame_order,int max_layer_depth)141 static inline int get_true_pyr_level(int frame_level, int frame_order,
142 int max_layer_depth) {
143 if (frame_order == 0) {
144 // Keyframe case
145 return MIN_PYR_LEVEL;
146 } else if (frame_level == MAX_ARF_LAYERS) {
147 // Leaves
148 return max_layer_depth;
149 } else if (frame_level == (MAX_ARF_LAYERS + 1)) {
150 // Altrefs
151 return MIN_PYR_LEVEL;
152 }
153 return AOMMAX(MIN_PYR_LEVEL, frame_level);
154 }
155
156 enum {
157 NO_AQ = 0,
158 VARIANCE_AQ = 1,
159 COMPLEXITY_AQ = 2,
160 CYCLIC_REFRESH_AQ = 3,
161 AQ_MODE_COUNT // This should always be the last member of the enum
162 } UENUM1BYTE(AQ_MODE);
163 enum {
164 NO_DELTA_Q = 0,
165 DELTA_Q_OBJECTIVE = 1, // Modulation to improve objective quality
166 DELTA_Q_PERCEPTUAL = 2, // Modulation to improve video perceptual quality
167 DELTA_Q_PERCEPTUAL_AI = 3, // Perceptual quality opt for all intra mode
168 DELTA_Q_USER_RATING_BASED = 4, // User rating based delta q mode
169 DELTA_Q_HDR = 5, // QP adjustment based on HDR block pixel average
170 DELTA_Q_VARIANCE_BOOST =
171 6, // Variance Boost style modulation for all intra mode
172 DELTA_Q_MODE_COUNT // This should always be the last member of the enum
173 } UENUM1BYTE(DELTAQ_MODE);
174
175 enum {
176 RESIZE_NONE = 0, // No frame resizing allowed.
177 RESIZE_FIXED = 1, // All frames are coded at the specified scale.
178 RESIZE_RANDOM = 2, // All frames are coded at a random scale.
179 RESIZE_DYNAMIC = 3, // Frames coded at lower scale based on rate control.
180 RESIZE_MODES
181 } UENUM1BYTE(RESIZE_MODE);
182
183 enum {
184 SS_CFG_SRC = 0,
185 SS_CFG_LOOKAHEAD = 1,
186 SS_CFG_FPF = 2,
187 SS_CFG_TOTAL = 3
188 } UENUM1BYTE(SS_CFG_OFFSET);
189
190 enum {
191 DISABLE_SCENECUT, // For LAP, lag_in_frames < 19
192 ENABLE_SCENECUT_MODE_1, // For LAP, lag_in_frames >=19 and < 33
193 ENABLE_SCENECUT_MODE_2 // For twopass and LAP - lag_in_frames >=33
194 } UENUM1BYTE(SCENECUT_MODE);
195
196 #define MAX_VBR_CORPUS_COMPLEXITY 10000
197
198 typedef enum {
199 MOD_FP, // First pass
200 MOD_TF, // Temporal filtering
201 MOD_TPL, // TPL
202 MOD_GME, // Global motion estimation
203 MOD_ENC, // Encode stage
204 MOD_LPF, // Deblocking loop filter
205 MOD_CDEF_SEARCH, // CDEF search
206 MOD_CDEF, // CDEF frame
207 MOD_LR, // Loop restoration filtering
208 MOD_PACK_BS, // Pack bitstream
209 MOD_FRAME_ENC, // Frame Parallel encode
210 MOD_AI, // All intra
211 NUM_MT_MODULES
212 } MULTI_THREADED_MODULES;
213
214 /*!\endcond */
215
216 /*!\enum COST_UPDATE_TYPE
217 * \brief This enum controls how often the entropy costs should be updated.
218 * \warning In case of any modifications/additions done to the enum
219 * COST_UPDATE_TYPE, the enum INTERNAL_COST_UPDATE_TYPE needs to be updated as
220 * well.
221 */
222 typedef enum {
223 COST_UPD_SB, /*!< Update every sb. */
224 COST_UPD_SBROW, /*!< Update every sb rows inside a tile. */
225 COST_UPD_TILE, /*!< Update every tile. */
226 COST_UPD_OFF, /*!< Turn off cost updates. */
227 NUM_COST_UPDATE_TYPES, /*!< Number of cost update types. */
228 } COST_UPDATE_TYPE;
229
230 /*!\enum LOOPFILTER_CONTROL
231 * \brief This enum controls to which frames loopfilter is applied.
232 */
233 typedef enum {
234 LOOPFILTER_NONE = 0, /*!< Disable loopfilter on all frames. */
235 LOOPFILTER_ALL = 1, /*!< Enable loopfilter for all frames. */
236 LOOPFILTER_REFERENCE = 2, /*!< Disable loopfilter on non reference frames. */
237 LOOPFILTER_SELECTIVELY =
238 3, /*!< Disable loopfilter on frames with low motion. */
239 } LOOPFILTER_CONTROL;
240
241 /*!\enum SKIP_APPLY_POSTPROC_FILTER
242 * \brief This enum controls the application of post-processing filters on a
243 * reconstructed frame.
244 */
245 typedef enum {
246 SKIP_APPLY_RESTORATION = 1 << 0,
247 SKIP_APPLY_SUPERRES = 1 << 1,
248 SKIP_APPLY_CDEF = 1 << 2,
249 SKIP_APPLY_LOOPFILTER = 1 << 3,
250 } SKIP_APPLY_POSTPROC_FILTER;
251
252 /*!
253 * \brief Encoder config related to resize.
254 */
255 typedef struct {
256 /*!
257 * Indicates the frame resize mode to be used by the encoder.
258 */
259 RESIZE_MODE resize_mode;
260 /*!
261 * Indicates the denominator for resize of inter frames, assuming 8 as the
262 * numerator. Its value ranges between 8-16.
263 */
264 uint8_t resize_scale_denominator;
265 /*!
266 * Indicates the denominator for resize of key frames, assuming 8 as the
267 * numerator. Its value ranges between 8-16.
268 */
269 uint8_t resize_kf_scale_denominator;
270 } ResizeCfg;
271
272 /*!
273 * \brief Encoder config for coding block partitioning.
274 */
275 typedef struct {
276 /*!
277 * Flag to indicate if rectanguar partitions should be enabled.
278 */
279 bool enable_rect_partitions;
280 /*!
281 * Flag to indicate if AB partitions should be enabled.
282 */
283 bool enable_ab_partitions;
284 /*!
285 * Flag to indicate if 1:4 / 4:1 partitions should be enabled.
286 */
287 bool enable_1to4_partitions;
288 /*!
289 * Indicates the minimum partition size that should be allowed. Both width and
290 * height of a partition cannot be smaller than the min_partition_size.
291 */
292 BLOCK_SIZE min_partition_size;
293 /*!
294 * Indicates the maximum partition size that should be allowed. Both width and
295 * height of a partition cannot be larger than the max_partition_size.
296 */
297 BLOCK_SIZE max_partition_size;
298 } PartitionCfg;
299
300 /*!
301 * \brief Encoder flags for intra prediction.
302 */
303 typedef struct {
304 /*!
305 * Flag to indicate if intra edge filtering process should be enabled.
306 */
307 bool enable_intra_edge_filter;
308 /*!
309 * Flag to indicate if recursive filtering based intra prediction should be
310 * enabled.
311 */
312 bool enable_filter_intra;
313 /*!
314 * Flag to indicate if smooth intra prediction modes should be enabled.
315 */
316 bool enable_smooth_intra;
317 /*!
318 * Flag to indicate if PAETH intra prediction mode should be enabled.
319 */
320 bool enable_paeth_intra;
321 /*!
322 * Flag to indicate if CFL uv intra mode should be enabled.
323 */
324 bool enable_cfl_intra;
325 /*!
326 * Flag to indicate if directional modes should be enabled.
327 */
328 bool enable_directional_intra;
329 /*!
330 * Flag to indicate if the subset of directional modes from D45 to D203 intra
331 * should be enabled. Has no effect if directional modes are disabled.
332 */
333 bool enable_diagonal_intra;
334 /*!
335 * Flag to indicate if delta angles for directional intra prediction should be
336 * enabled.
337 */
338 bool enable_angle_delta;
339 /*!
340 * Flag to indicate whether to automatically turn off several intral coding
341 * tools.
342 * This flag is only used when "--deltaq-mode=3" is true.
343 * When set to 1, the encoder will analyze the reconstruction quality
344 * as compared to the source image in the preprocessing pass.
345 * If the recontruction quality is considered high enough, we disable
346 * the following intra coding tools, for better encoding speed:
347 * "--enable_smooth_intra",
348 * "--enable_paeth_intra",
349 * "--enable_cfl_intra",
350 * "--enable_diagonal_intra".
351 */
352 bool auto_intra_tools_off;
353 } IntraModeCfg;
354
355 /*!
356 * \brief Encoder flags for transform sizes and types.
357 */
358 typedef struct {
359 /*!
360 * Flag to indicate if 64-pt transform should be enabled.
361 */
362 bool enable_tx64;
363 /*!
364 * Flag to indicate if flip and identity transform types should be enabled.
365 */
366 bool enable_flip_idtx;
367 /*!
368 * Flag to indicate if rectangular transform should be enabled.
369 */
370 bool enable_rect_tx;
371 /*!
372 * Flag to indicate whether or not to use a default reduced set for ext-tx
373 * rather than the potential full set of 16 transforms.
374 */
375 bool reduced_tx_type_set;
376 /*!
377 * Flag to indicate if transform type for intra blocks should be limited to
378 * DCT_DCT.
379 */
380 bool use_intra_dct_only;
381 /*!
382 * Flag to indicate if transform type for inter blocks should be limited to
383 * DCT_DCT.
384 */
385 bool use_inter_dct_only;
386 /*!
387 * Flag to indicate if intra blocks should use default transform type
388 * (mode-dependent) only.
389 */
390 bool use_intra_default_tx_only;
391 /*!
392 * Flag to indicate if transform size search should be enabled.
393 */
394 bool enable_tx_size_search;
395 } TxfmSizeTypeCfg;
396
397 /*!
398 * \brief Encoder flags for compound prediction modes.
399 */
400 typedef struct {
401 /*!
402 * Flag to indicate if distance-weighted compound type should be enabled.
403 */
404 bool enable_dist_wtd_comp;
405 /*!
406 * Flag to indicate if masked (wedge/diff-wtd) compound type should be
407 * enabled.
408 */
409 bool enable_masked_comp;
410 /*!
411 * Flag to indicate if smooth interintra mode should be enabled.
412 */
413 bool enable_smooth_interintra;
414 /*!
415 * Flag to indicate if difference-weighted compound type should be enabled.
416 */
417 bool enable_diff_wtd_comp;
418 /*!
419 * Flag to indicate if inter-inter wedge compound type should be enabled.
420 */
421 bool enable_interinter_wedge;
422 /*!
423 * Flag to indicate if inter-intra wedge compound type should be enabled.
424 */
425 bool enable_interintra_wedge;
426 } CompoundTypeCfg;
427
428 /*!
429 * \brief Encoder config related to frame super-resolution.
430 */
431 typedef struct {
432 /*!
433 * Indicates the qindex based threshold to be used when AOM_SUPERRES_QTHRESH
434 * mode is used for inter frames.
435 */
436 int superres_qthresh;
437 /*!
438 * Indicates the qindex based threshold to be used when AOM_SUPERRES_QTHRESH
439 * mode is used for key frames.
440 */
441 int superres_kf_qthresh;
442 /*!
443 * Indicates the denominator of the fraction that specifies the ratio between
444 * the superblock width before and after upscaling for inter frames. The
445 * numerator of this fraction is equal to the constant SCALE_NUMERATOR.
446 */
447 uint8_t superres_scale_denominator;
448 /*!
449 * Indicates the denominator of the fraction that specifies the ratio between
450 * the superblock width before and after upscaling for key frames. The
451 * numerator of this fraction is equal to the constant SCALE_NUMERATOR.
452 */
453 uint8_t superres_kf_scale_denominator;
454 /*!
455 * Indicates the Super-resolution mode to be used by the encoder.
456 */
457 aom_superres_mode superres_mode;
458 /*!
459 * Flag to indicate if super-resolution should be enabled for the sequence.
460 */
461 bool enable_superres;
462 } SuperResCfg;
463
464 /*!
465 * \brief Encoder config related to the coding of key frames.
466 */
467 typedef struct {
468 /*!
469 * Indicates the minimum distance to a key frame.
470 */
471 int key_freq_min;
472
473 /*!
474 * Indicates the maximum distance to a key frame.
475 */
476 int key_freq_max;
477
478 /*!
479 * Indicates if temporal filtering should be applied on keyframe.
480 */
481 int enable_keyframe_filtering;
482
483 /*!
484 * Indicates the number of frames after which a frame may be coded as an
485 * S-Frame.
486 */
487 int sframe_dist;
488
489 /*!
490 * Indicates how an S-Frame should be inserted.
491 * 1: the considered frame will be made into an S-Frame only if it is an
492 * altref frame. 2: the next altref frame will be made into an S-Frame.
493 */
494 int sframe_mode;
495
496 /*!
497 * Indicates if encoder should autodetect cut scenes and set the keyframes.
498 */
499 bool auto_key;
500
501 /*!
502 * Indicates the forward key frame distance.
503 */
504 int fwd_kf_dist;
505
506 /*!
507 * Indicates if forward keyframe reference should be enabled.
508 */
509 bool fwd_kf_enabled;
510
511 /*!
512 * Indicates if S-Frames should be enabled for the sequence.
513 */
514 bool enable_sframe;
515
516 /*!
517 * Indicates if intra block copy prediction mode should be enabled or not.
518 */
519 bool enable_intrabc;
520 } KeyFrameCfg;
521
522 /*!
523 * \brief Encoder rate control configuration parameters
524 */
525 typedef struct {
526 /*!\cond */
527 // BUFFERING PARAMETERS
528 /*!\endcond */
529 /*!
530 * Indicates the amount of data that will be buffered by the decoding
531 * application prior to beginning playback, and is expressed in units of
532 * time(milliseconds).
533 */
534 int64_t starting_buffer_level_ms;
535 /*!
536 * Indicates the amount of data that the encoder should try to maintain in the
537 * decoder's buffer, and is expressed in units of time(milliseconds).
538 */
539 int64_t optimal_buffer_level_ms;
540 /*!
541 * Indicates the maximum amount of data that may be buffered by the decoding
542 * application, and is expressed in units of time(milliseconds).
543 */
544 int64_t maximum_buffer_size_ms;
545
546 /*!
547 * Indicates the bandwidth to be used in bits per second.
548 */
549 int64_t target_bandwidth;
550
551 /*!
552 * Indicates average complexity of the corpus in single pass vbr based on
553 * LAP. 0 indicates that corpus complexity vbr mode is disabled.
554 */
555 unsigned int vbr_corpus_complexity_lap;
556 /*!
557 * Indicates the maximum allowed bitrate for any intra frame as % of bitrate
558 * target.
559 */
560 unsigned int max_intra_bitrate_pct;
561 /*!
562 * Indicates the maximum allowed bitrate for any inter frame as % of bitrate
563 * target.
564 */
565 unsigned int max_inter_bitrate_pct;
566 /*!
567 * Indicates the percentage of rate boost for golden frame in CBR mode.
568 */
569 unsigned int gf_cbr_boost_pct;
570 /*!
571 * min_cr / 100 indicates the target minimum compression ratio for each
572 * frame.
573 */
574 unsigned int min_cr;
575 /*!
576 * Indicates the frame drop threshold.
577 */
578 int drop_frames_water_mark;
579 /*!
580 * under_shoot_pct indicates the tolerance of the VBR algorithm to
581 * undershoot and is used as a trigger threshold for more aggressive
582 * adaptation of Q. It's value can range from 0-100.
583 */
584 int under_shoot_pct;
585 /*!
586 * over_shoot_pct indicates the tolerance of the VBR algorithm to overshoot
587 * and is used as a trigger threshold for more aggressive adaptation of Q.
588 * It's value can range from 0-1000.
589 */
590 int over_shoot_pct;
591 /*!
592 * Indicates the maximum qindex that can be used by the quantizer i.e. the
593 * worst quality qindex.
594 */
595 int worst_allowed_q;
596 /*!
597 * Indicates the minimum qindex that can be used by the quantizer i.e. the
598 * best quality qindex.
599 */
600 int best_allowed_q;
601 /*!
602 * Indicates the Constant/Constrained Quality level.
603 */
604 int cq_level;
605 /*!
606 * Indicates if the encoding mode is vbr, cbr, constrained quality or
607 * constant quality.
608 */
609 enum aom_rc_mode mode;
610 /*!
611 * Indicates the bias (expressed on a scale of 0 to 100) for determining
612 * target size for the current frame. The value 0 indicates the optimal CBR
613 * mode value should be used, and 100 indicates the optimal VBR mode value
614 * should be used.
615 */
616 int vbrbias;
617 /*!
618 * Indicates the minimum bitrate to be used for a single frame as a percentage
619 * of the target bitrate.
620 */
621 int vbrmin_section;
622 /*!
623 * Indicates the maximum bitrate to be used for a single frame as a percentage
624 * of the target bitrate.
625 */
626 int vbrmax_section;
627
628 /*!
629 * Indicates the maximum consecutive amount of frame drops, in units of time
630 * (milliseconds). This is converted to frame units internally. Only used in
631 * CBR mode.
632 */
633 int max_consec_drop_ms;
634 } RateControlCfg;
635
636 /*!\cond */
637 typedef struct {
638 // Indicates the number of frames lag before encoding is started.
639 int lag_in_frames;
640 // Indicates the minimum gf/arf interval to be used.
641 int min_gf_interval;
642 // Indicates the maximum gf/arf interval to be used.
643 int max_gf_interval;
644 // Indicates the minimum height for GF group pyramid structure to be used.
645 int gf_min_pyr_height;
646 // Indicates the maximum height for GF group pyramid structure to be used.
647 int gf_max_pyr_height;
648 // Indicates if automatic set and use of altref frames should be enabled.
649 bool enable_auto_arf;
650 // Indicates if automatic set and use of (b)ackward (r)ef (f)rames should be
651 // enabled.
652 bool enable_auto_brf;
653 } GFConfig;
654
655 typedef struct {
656 // Indicates the number of tile groups.
657 unsigned int num_tile_groups;
658 // Indicates the MTU size for a tile group. If mtu is non-zero,
659 // num_tile_groups is set to DEFAULT_MAX_NUM_TG.
660 unsigned int mtu;
661 // Indicates the number of tile columns in log2.
662 int tile_columns;
663 // Indicates the number of tile rows in log2.
664 int tile_rows;
665 // Indicates the number of widths in the tile_widths[] array.
666 int tile_width_count;
667 // Indicates the number of heights in the tile_heights[] array.
668 int tile_height_count;
669 // Indicates the tile widths, and may be empty.
670 int tile_widths[MAX_TILE_COLS];
671 // Indicates the tile heights, and may be empty.
672 int tile_heights[MAX_TILE_ROWS];
673 // Indicates if large scale tile coding should be used.
674 bool enable_large_scale_tile;
675 // Indicates if single tile decoding mode should be enabled.
676 bool enable_single_tile_decoding;
677 // Indicates if EXT_TILE_DEBUG should be enabled.
678 bool enable_ext_tile_debug;
679 } TileConfig;
680
681 typedef struct {
682 // Indicates the width of the input frame.
683 int width;
684 // Indicates the height of the input frame.
685 int height;
686 // If forced_max_frame_width is non-zero then it is used to force the maximum
687 // frame width written in write_sequence_header().
688 int forced_max_frame_width;
689 // If forced_max_frame_width is non-zero then it is used to force the maximum
690 // frame height written in write_sequence_header().
691 int forced_max_frame_height;
692 // Indicates the frame width after applying both super-resolution and resize
693 // to the coded frame.
694 int render_width;
695 // Indicates the frame height after applying both super-resolution and resize
696 // to the coded frame.
697 int render_height;
698 } FrameDimensionCfg;
699
700 typedef struct {
701 // Indicates if warped motion should be enabled.
702 bool enable_warped_motion;
703 // Indicates if warped motion should be evaluated or not.
704 bool allow_warped_motion;
705 // Indicates if OBMC motion should be enabled.
706 bool enable_obmc;
707 } MotionModeCfg;
708
709 typedef struct {
710 // Timing info for each frame.
711 aom_timing_info_t timing_info;
712 // Indicates the number of time units of a decoding clock.
713 uint32_t num_units_in_decoding_tick;
714 // Indicates if decoder model information is present in the coded sequence
715 // header.
716 bool decoder_model_info_present_flag;
717 // Indicates if display model information is present in the coded sequence
718 // header.
719 bool display_model_info_present_flag;
720 // Indicates if timing info for each frame is present.
721 bool timing_info_present;
722 } DecoderModelCfg;
723
724 typedef struct {
725 // Indicates the update frequency for coeff costs.
726 COST_UPDATE_TYPE coeff;
727 // Indicates the update frequency for mode costs.
728 COST_UPDATE_TYPE mode;
729 // Indicates the update frequency for mv costs.
730 COST_UPDATE_TYPE mv;
731 // Indicates the update frequency for dv costs.
732 COST_UPDATE_TYPE dv;
733 } CostUpdateFreq;
734
735 typedef struct {
736 // Indicates the maximum number of reference frames allowed per frame.
737 unsigned int max_reference_frames;
738 // Indicates if the reduced set of references should be enabled.
739 bool enable_reduced_reference_set;
740 // Indicates if one-sided compound should be enabled.
741 bool enable_onesided_comp;
742 } RefFrameCfg;
743
744 typedef struct {
745 // Indicates the color space that should be used.
746 aom_color_primaries_t color_primaries;
747 // Indicates the characteristics of transfer function to be used.
748 aom_transfer_characteristics_t transfer_characteristics;
749 // Indicates the matrix coefficients to be used for the transfer function.
750 aom_matrix_coefficients_t matrix_coefficients;
751 // Indicates the chroma 4:2:0 sample position info.
752 aom_chroma_sample_position_t chroma_sample_position;
753 // Indicates if a limited color range or full color range should be used.
754 aom_color_range_t color_range;
755 } ColorCfg;
756
757 typedef struct {
758 // Indicates if extreme motion vector unit test should be enabled or not.
759 unsigned int motion_vector_unit_test;
760 // Indicates if superblock multipass unit test should be enabled or not.
761 unsigned int sb_multipass_unit_test;
762 } UnitTestCfg;
763
764 typedef struct {
765 // Indicates the file path to the VMAF model.
766 const char *vmaf_model_path;
767 // Indicates the path to the film grain parameters.
768 const char *film_grain_table_filename;
769 // Indicates the visual tuning metric.
770 aom_tune_metric tuning;
771 // Indicates if the current content is screen or default type.
772 aom_tune_content content;
773 // Indicates the film grain parameters.
774 int film_grain_test_vector;
775 // Indicates the in-block distortion metric to use.
776 aom_dist_metric dist_metric;
777 } TuneCfg;
778
779 typedef struct {
780 // Indicates the framerate of the input video.
781 double init_framerate;
782 // Indicates the bit-depth of the input video.
783 unsigned int input_bit_depth;
784 // Indicates the maximum number of frames to be encoded.
785 unsigned int limit;
786 // Indicates the chrome subsampling x value.
787 unsigned int chroma_subsampling_x;
788 // Indicates the chrome subsampling y value.
789 unsigned int chroma_subsampling_y;
790 } InputCfg;
791
792 typedef struct {
793 // If true, encoder will use fixed QP offsets, that are either:
794 // - Given by the user, and stored in 'fixed_qp_offsets' array, OR
795 // - Picked automatically from cq_level.
796 int use_fixed_qp_offsets;
797 // Indicates the minimum flatness of the quantization matrix.
798 int qm_minlevel;
799 // Indicates the maximum flatness of the quantization matrix.
800 int qm_maxlevel;
801 // Indicates if adaptive quantize_b should be enabled.
802 int quant_b_adapt;
803 // Indicates the Adaptive Quantization mode to be used.
804 AQ_MODE aq_mode;
805 // Indicates the delta q mode to be used.
806 DELTAQ_MODE deltaq_mode;
807 // Indicates the delta q mode strength.
808 DELTAQ_MODE deltaq_strength;
809 // Indicates if delta quantization should be enabled in chroma planes.
810 bool enable_chroma_deltaq;
811 // Indicates if delta quantization should be enabled for hdr video
812 bool enable_hdr_deltaq;
813 // Indicates if encoding with quantization matrices should be enabled.
814 bool using_qm;
815 } QuantizationCfg;
816
817 /*!\endcond */
818 /*!
819 * \brief Algorithm configuration parameters.
820 */
821 typedef struct {
822 /*!
823 * Controls the level at which rate-distortion optimization of transform
824 * coefficients favors sharpness in the block. Has no impact on RD when set
825 * to zero (default).
826 *
827 * For values 1-7, eob and skip block optimization are
828 * avoided and rdmult is adjusted in favor of block sharpness.
829 *
830 * In all-intra mode: it also sets the `loop_filter_sharpness` syntax element
831 * in the bitstream. Larger values increasingly reduce how much the filtering
832 * can change the sample values on block edges to favor perceived sharpness.
833 */
834 int sharpness;
835
836 /*!
837 * Indicates the trellis optimization mode of quantized coefficients.
838 * 0: disabled
839 * 1: enabled
840 * 2: enabled for rd search
841 * 3: true for estimate yrd search
842 */
843 int disable_trellis_quant;
844
845 /*!
846 * The maximum number of frames used to create an arf.
847 */
848 int arnr_max_frames;
849
850 /*!
851 * The temporal filter strength for arf used when creating ARFs.
852 */
853 int arnr_strength;
854
855 /*!
856 * Indicates the CDF update mode
857 * 0: no update
858 * 1: update on every frame(default)
859 * 2: selectively update
860 */
861 uint8_t cdf_update_mode;
862
863 /*!
864 * Indicates if RDO based on frame temporal dependency should be enabled.
865 */
866 bool enable_tpl_model;
867
868 /*!
869 * Indicates if coding of overlay frames for filtered ALTREF frames is
870 * enabled.
871 */
872 bool enable_overlay;
873
874 /*!
875 * Controls loop filtering
876 * 0: Loop filter is disabled for all frames
877 * 1: Loop filter is enabled for all frames
878 * 2: Loop filter is disabled for non-reference frames
879 * 3: Loop filter is disables for the frames with low motion
880 */
881 LOOPFILTER_CONTROL loopfilter_control;
882
883 /*!
884 * Indicates if the application of post-processing filters should be skipped
885 * on reconstructed frame.
886 */
887 bool skip_postproc_filtering;
888 } AlgoCfg;
889 /*!\cond */
890
891 typedef struct {
892 // Indicates the codec bit-depth.
893 aom_bit_depth_t bit_depth;
894 // Indicates the superblock size that should be used by the encoder.
895 aom_superblock_size_t superblock_size;
896 // Indicates if loopfilter modulation should be enabled.
897 bool enable_deltalf_mode;
898 // Indicates how CDEF should be applied.
899 CDEF_CONTROL cdef_control;
900 // Indicates if loop restoration filter should be enabled.
901 bool enable_restoration;
902 // When enabled, video mode should be used even for single frame input.
903 bool force_video_mode;
904 // Indicates if the error resiliency features should be enabled.
905 bool error_resilient_mode;
906 // Indicates if frame parallel decoding feature should be enabled.
907 bool frame_parallel_decoding_mode;
908 // Indicates if the input should be encoded as monochrome.
909 bool enable_monochrome;
910 // When enabled, the encoder will use a full header even for still pictures.
911 // When disabled, a reduced header is used for still pictures.
912 bool full_still_picture_hdr;
913 // Indicates if dual interpolation filters should be enabled.
914 bool enable_dual_filter;
915 // Indicates if frame order hint should be enabled or not.
916 bool enable_order_hint;
917 // Indicates if ref_frame_mvs should be enabled at the sequence level.
918 bool ref_frame_mvs_present;
919 // Indicates if ref_frame_mvs should be enabled at the frame level.
920 bool enable_ref_frame_mvs;
921 // Indicates if interintra compound mode is enabled.
922 bool enable_interintra_comp;
923 // Indicates if global motion should be enabled.
924 bool enable_global_motion;
925 // Indicates if palette should be enabled.
926 bool enable_palette;
927 } ToolCfg;
928
929 /*!\endcond */
930 /*!
931 * \brief Main encoder configuration data structure.
932 */
933 typedef struct AV1EncoderConfig {
934 /*!\cond */
935 // Configuration related to the input video.
936 InputCfg input_cfg;
937
938 // Configuration related to frame-dimensions.
939 FrameDimensionCfg frm_dim_cfg;
940
941 /*!\endcond */
942 /*!
943 * Encoder algorithm configuration.
944 */
945 AlgoCfg algo_cfg;
946
947 /*!
948 * Configuration related to key-frames.
949 */
950 KeyFrameCfg kf_cfg;
951
952 /*!
953 * Rate control configuration
954 */
955 RateControlCfg rc_cfg;
956 /*!\cond */
957
958 // Configuration related to Quantization.
959 QuantizationCfg q_cfg;
960
961 // Internal frame size scaling.
962 ResizeCfg resize_cfg;
963
964 // Frame Super-Resolution size scaling.
965 SuperResCfg superres_cfg;
966
967 /*!\endcond */
968 /*!
969 * stats_in buffer contains all of the stats packets produced in the first
970 * pass, concatenated.
971 */
972 aom_fixed_buf_t twopass_stats_in;
973 /*!\cond */
974
975 // Configuration related to encoder toolsets.
976 ToolCfg tool_cfg;
977
978 // Configuration related to Group of frames.
979 GFConfig gf_cfg;
980
981 // Tile related configuration parameters.
982 TileConfig tile_cfg;
983
984 // Configuration related to Tune.
985 TuneCfg tune_cfg;
986
987 // Configuration related to color.
988 ColorCfg color_cfg;
989
990 // Configuration related to decoder model.
991 DecoderModelCfg dec_model_cfg;
992
993 // Configuration related to reference frames.
994 RefFrameCfg ref_frm_cfg;
995
996 // Configuration related to unit tests.
997 UnitTestCfg unit_test_cfg;
998
999 // Flags related to motion mode.
1000 MotionModeCfg motion_mode_cfg;
1001
1002 // Flags related to intra mode search.
1003 IntraModeCfg intra_mode_cfg;
1004
1005 // Flags related to transform size/type.
1006 TxfmSizeTypeCfg txfm_cfg;
1007
1008 // Flags related to compound type.
1009 CompoundTypeCfg comp_type_cfg;
1010
1011 // Partition related information.
1012 PartitionCfg part_cfg;
1013
1014 // Configuration related to frequency of cost update.
1015 CostUpdateFreq cost_upd_freq;
1016
1017 #if CONFIG_DENOISE
1018 // Indicates the noise level.
1019 float noise_level;
1020 // Indicates the the denoisers block size.
1021 int noise_block_size;
1022 // Indicates whether to apply denoising to the frame to be encoded
1023 int enable_dnl_denoising;
1024 #endif
1025
1026 #if CONFIG_AV1_TEMPORAL_DENOISING
1027 // Noise sensitivity.
1028 int noise_sensitivity;
1029 #endif
1030 // Bit mask to specify which tier each of the 32 possible operating points
1031 // conforms to.
1032 unsigned int tier_mask;
1033
1034 // Indicates the number of pixels off the edge of a reference frame we're
1035 // allowed to go when forming an inter prediction.
1036 int border_in_pixels;
1037
1038 // Indicates the maximum number of threads that may be used by the encoder.
1039 int max_threads;
1040
1041 // Indicates the speed preset to be used.
1042 int speed;
1043
1044 // Indicates the target sequence level index for each operating point(OP).
1045 AV1_LEVEL target_seq_level_idx[MAX_NUM_OPERATING_POINTS];
1046
1047 // Indicates the bitstream profile to be used.
1048 BITSTREAM_PROFILE profile;
1049
1050 /*!\endcond */
1051 /*!
1052 * Indicates the current encoder pass :
1053 * AOM_RC_ONE_PASS = One pass encode,
1054 * AOM_RC_FIRST_PASS = First pass of multiple-pass
1055 * AOM_RC_SECOND_PASS = Second pass of multiple-pass
1056 * AOM_RC_THIRD_PASS = Third pass of multiple-pass
1057 */
1058 enum aom_enc_pass pass;
1059 /*!\cond */
1060
1061 // Total number of encoding passes.
1062 int passes;
1063
1064 // the name of the second pass output file when passes > 2
1065 const char *two_pass_output;
1066
1067 // the name of the second pass log file when passes > 2
1068 const char *second_pass_log;
1069
1070 // Indicates if the encoding is GOOD or REALTIME.
1071 MODE mode;
1072
1073 // Indicates if row-based multi-threading should be enabled or not.
1074 bool row_mt;
1075
1076 // Indicates if frame parallel multi-threading should be enabled or not.
1077 bool fp_mt;
1078
1079 // Indicates if 16bit frame buffers are to be used i.e., the content is >
1080 // 8-bit.
1081 bool use_highbitdepth;
1082
1083 // Indicates the bitstream syntax mode. 0 indicates bitstream is saved as
1084 // Section 5 bitstream, while 1 indicates the bitstream is saved in Annex - B
1085 // format.
1086 bool save_as_annexb;
1087
1088 // The path for partition stats reading and writing, used in the experiment
1089 // CONFIG_PARTITION_SEARCH_ORDER.
1090 const char *partition_info_path;
1091
1092 // The flag that indicates whether we use an external rate distribution to
1093 // guide adaptive quantization. It requires --deltaq-mode=3. The rate
1094 // distribution map file name is stored in |rate_distribution_info|.
1095 unsigned int enable_rate_guide_deltaq;
1096
1097 // The input file of rate distribution information used in all intra mode
1098 // to determine delta quantization.
1099 const char *rate_distribution_info;
1100
1101 // Exit the encoder when it fails to encode to a given level.
1102 int strict_level_conformance;
1103
1104 // Max depth for the GOP after a key frame
1105 int kf_max_pyr_height;
1106
1107 // A flag to control if we enable the superblock qp sweep for a given lambda
1108 int sb_qp_sweep;
1109 /*!\endcond */
1110 } AV1EncoderConfig;
1111
1112 /*!\cond */
is_lossless_requested(const RateControlCfg * const rc_cfg)1113 static inline int is_lossless_requested(const RateControlCfg *const rc_cfg) {
1114 return rc_cfg->best_allowed_q == 0 && rc_cfg->worst_allowed_q == 0;
1115 }
1116 /*!\endcond */
1117
1118 /*!
1119 * \brief Encoder-side probabilities for pruning of various AV1 tools
1120 */
1121 typedef struct {
1122 /*!
1123 * obmc_probs[i][j] is the probability of OBMC being the best motion mode for
1124 * jth block size and ith frame update type, averaged over past frames. If
1125 * obmc_probs[i][j] < thresh, then OBMC search is pruned.
1126 */
1127 int obmc_probs[FRAME_UPDATE_TYPES][BLOCK_SIZES_ALL];
1128
1129 /*!
1130 * warped_probs[i] is the probability of warped motion being the best motion
1131 * mode for ith frame update type, averaged over past frames. If
1132 * warped_probs[i] < thresh, then warped motion search is pruned.
1133 */
1134 int warped_probs[FRAME_UPDATE_TYPES];
1135
1136 /*!
1137 * tx_type_probs[i][j][k] is the probability of kth tx_type being the best
1138 * for jth transform size and ith frame update type, averaged over past
1139 * frames. If tx_type_probs[i][j][k] < thresh, then transform search for that
1140 * type is pruned.
1141 */
1142 int tx_type_probs[FRAME_UPDATE_TYPES][TX_SIZES_ALL][TX_TYPES];
1143
1144 /*!
1145 * switchable_interp_probs[i][j][k] is the probability of kth interpolation
1146 * filter being the best for jth filter context and ith frame update type,
1147 * averaged over past frames. If switchable_interp_probs[i][j][k] < thresh,
1148 * then interpolation filter search is pruned for that case.
1149 */
1150 int switchable_interp_probs[FRAME_UPDATE_TYPES][SWITCHABLE_FILTER_CONTEXTS]
1151 [SWITCHABLE_FILTERS];
1152 } FrameProbInfo;
1153
1154 /*!\cond */
1155
1156 typedef struct FRAME_COUNTS {
1157 // Note: This structure should only contain 'unsigned int' fields, or
1158 // aggregates built solely from 'unsigned int' fields/elements
1159 #if CONFIG_ENTROPY_STATS
1160 unsigned int kf_y_mode[KF_MODE_CONTEXTS][KF_MODE_CONTEXTS][INTRA_MODES];
1161 unsigned int angle_delta[DIRECTIONAL_MODES][2 * MAX_ANGLE_DELTA + 1];
1162 unsigned int y_mode[BLOCK_SIZE_GROUPS][INTRA_MODES];
1163 unsigned int uv_mode[CFL_ALLOWED_TYPES][INTRA_MODES][UV_INTRA_MODES];
1164 unsigned int cfl_sign[CFL_JOINT_SIGNS];
1165 unsigned int cfl_alpha[CFL_ALPHA_CONTEXTS][CFL_ALPHABET_SIZE];
1166 unsigned int palette_y_mode[PALATTE_BSIZE_CTXS][PALETTE_Y_MODE_CONTEXTS][2];
1167 unsigned int palette_uv_mode[PALETTE_UV_MODE_CONTEXTS][2];
1168 unsigned int palette_y_size[PALATTE_BSIZE_CTXS][PALETTE_SIZES];
1169 unsigned int palette_uv_size[PALATTE_BSIZE_CTXS][PALETTE_SIZES];
1170 unsigned int palette_y_color_index[PALETTE_SIZES]
1171 [PALETTE_COLOR_INDEX_CONTEXTS]
1172 [PALETTE_COLORS];
1173 unsigned int palette_uv_color_index[PALETTE_SIZES]
1174 [PALETTE_COLOR_INDEX_CONTEXTS]
1175 [PALETTE_COLORS];
1176 unsigned int partition[PARTITION_CONTEXTS][EXT_PARTITION_TYPES];
1177 unsigned int txb_skip[TOKEN_CDF_Q_CTXS][TX_SIZES][TXB_SKIP_CONTEXTS][2];
1178 unsigned int eob_extra[TOKEN_CDF_Q_CTXS][TX_SIZES][PLANE_TYPES]
1179 [EOB_COEF_CONTEXTS][2];
1180 unsigned int dc_sign[PLANE_TYPES][DC_SIGN_CONTEXTS][2];
1181 unsigned int coeff_lps[TX_SIZES][PLANE_TYPES][BR_CDF_SIZE - 1][LEVEL_CONTEXTS]
1182 [2];
1183 unsigned int eob_flag[TX_SIZES][PLANE_TYPES][EOB_COEF_CONTEXTS][2];
1184 unsigned int eob_multi16[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][5];
1185 unsigned int eob_multi32[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][6];
1186 unsigned int eob_multi64[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][7];
1187 unsigned int eob_multi128[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][8];
1188 unsigned int eob_multi256[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][9];
1189 unsigned int eob_multi512[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][10];
1190 unsigned int eob_multi1024[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][11];
1191 unsigned int coeff_lps_multi[TOKEN_CDF_Q_CTXS][TX_SIZES][PLANE_TYPES]
1192 [LEVEL_CONTEXTS][BR_CDF_SIZE];
1193 unsigned int coeff_base_multi[TOKEN_CDF_Q_CTXS][TX_SIZES][PLANE_TYPES]
1194 [SIG_COEF_CONTEXTS][NUM_BASE_LEVELS + 2];
1195 unsigned int coeff_base_eob_multi[TOKEN_CDF_Q_CTXS][TX_SIZES][PLANE_TYPES]
1196 [SIG_COEF_CONTEXTS_EOB][NUM_BASE_LEVELS + 1];
1197 unsigned int newmv_mode[NEWMV_MODE_CONTEXTS][2];
1198 unsigned int zeromv_mode[GLOBALMV_MODE_CONTEXTS][2];
1199 unsigned int refmv_mode[REFMV_MODE_CONTEXTS][2];
1200 unsigned int drl_mode[DRL_MODE_CONTEXTS][2];
1201 unsigned int inter_compound_mode[INTER_MODE_CONTEXTS][INTER_COMPOUND_MODES];
1202 unsigned int wedge_idx[BLOCK_SIZES_ALL][16];
1203 unsigned int interintra[BLOCK_SIZE_GROUPS][2];
1204 unsigned int interintra_mode[BLOCK_SIZE_GROUPS][INTERINTRA_MODES];
1205 unsigned int wedge_interintra[BLOCK_SIZES_ALL][2];
1206 unsigned int compound_type[BLOCK_SIZES_ALL][MASKED_COMPOUND_TYPES];
1207 unsigned int motion_mode[BLOCK_SIZES_ALL][MOTION_MODES];
1208 unsigned int obmc[BLOCK_SIZES_ALL][2];
1209 unsigned int intra_inter[INTRA_INTER_CONTEXTS][2];
1210 unsigned int comp_inter[COMP_INTER_CONTEXTS][2];
1211 unsigned int comp_ref_type[COMP_REF_TYPE_CONTEXTS][2];
1212 unsigned int uni_comp_ref[UNI_COMP_REF_CONTEXTS][UNIDIR_COMP_REFS - 1][2];
1213 unsigned int single_ref[REF_CONTEXTS][SINGLE_REFS - 1][2];
1214 unsigned int comp_ref[REF_CONTEXTS][FWD_REFS - 1][2];
1215 unsigned int comp_bwdref[REF_CONTEXTS][BWD_REFS - 1][2];
1216 unsigned int intrabc[2];
1217
1218 unsigned int txfm_partition[TXFM_PARTITION_CONTEXTS][2];
1219 unsigned int intra_tx_size[MAX_TX_CATS][TX_SIZE_CONTEXTS][MAX_TX_DEPTH + 1];
1220 unsigned int skip_mode[SKIP_MODE_CONTEXTS][2];
1221 unsigned int skip_txfm[SKIP_CONTEXTS][2];
1222 unsigned int compound_index[COMP_INDEX_CONTEXTS][2];
1223 unsigned int comp_group_idx[COMP_GROUP_IDX_CONTEXTS][2];
1224 unsigned int delta_q[DELTA_Q_PROBS][2];
1225 unsigned int delta_lf_multi[FRAME_LF_COUNT][DELTA_LF_PROBS][2];
1226 unsigned int delta_lf[DELTA_LF_PROBS][2];
1227
1228 unsigned int inter_ext_tx[EXT_TX_SETS_INTER][EXT_TX_SIZES][TX_TYPES];
1229 unsigned int intra_ext_tx[EXT_TX_SETS_INTRA][EXT_TX_SIZES][INTRA_MODES]
1230 [TX_TYPES];
1231 unsigned int filter_intra_mode[FILTER_INTRA_MODES];
1232 unsigned int filter_intra[BLOCK_SIZES_ALL][2];
1233 unsigned int switchable_restore[RESTORE_SWITCHABLE_TYPES];
1234 unsigned int wiener_restore[2];
1235 unsigned int sgrproj_restore[2];
1236 #endif // CONFIG_ENTROPY_STATS
1237
1238 unsigned int switchable_interp[SWITCHABLE_FILTER_CONTEXTS]
1239 [SWITCHABLE_FILTERS];
1240 } FRAME_COUNTS;
1241
1242 #define INTER_MODE_RD_DATA_OVERALL_SIZE 6400
1243
1244 typedef struct {
1245 int ready;
1246 double a;
1247 double b;
1248 double dist_mean;
1249 double ld_mean;
1250 double sse_mean;
1251 double sse_sse_mean;
1252 double sse_ld_mean;
1253 int num;
1254 double dist_sum;
1255 double ld_sum;
1256 double sse_sum;
1257 double sse_sse_sum;
1258 double sse_ld_sum;
1259 } InterModeRdModel;
1260
1261 typedef struct {
1262 int idx;
1263 int64_t rd;
1264 } RdIdxPair;
1265 // TODO(angiebird): This is an estimated size. We still need to figure what is
1266 // the maximum number of modes.
1267 #define MAX_INTER_MODES 1024
1268 // TODO(any): rename this struct to something else. There is already another
1269 // struct called inter_mode_info, which makes this terribly confusing.
1270 /*!\endcond */
1271 /*!
1272 * \brief Struct used to hold inter mode data for fast tx search.
1273 *
1274 * This struct is used to perform a full transform search only on winning
1275 * candidates searched with an estimate for transform coding RD.
1276 */
1277 typedef struct inter_modes_info {
1278 /*!
1279 * The number of inter modes for which data was stored in each of the
1280 * following arrays.
1281 */
1282 int num;
1283 /*!
1284 * Mode info struct for each of the candidate modes.
1285 */
1286 MB_MODE_INFO mbmi_arr[MAX_INTER_MODES];
1287 /*!
1288 * The rate for each of the candidate modes.
1289 */
1290 int mode_rate_arr[MAX_INTER_MODES];
1291 /*!
1292 * The sse of the predictor for each of the candidate modes.
1293 */
1294 int64_t sse_arr[MAX_INTER_MODES];
1295 /*!
1296 * The estimated rd of the predictor for each of the candidate modes.
1297 */
1298 int64_t est_rd_arr[MAX_INTER_MODES];
1299 /*!
1300 * The rate and mode index for each of the candidate modes.
1301 */
1302 RdIdxPair rd_idx_pair_arr[MAX_INTER_MODES];
1303 /*!
1304 * The full rd stats for each of the candidate modes.
1305 */
1306 RD_STATS rd_cost_arr[MAX_INTER_MODES];
1307 /*!
1308 * The full rd stats of luma only for each of the candidate modes.
1309 */
1310 RD_STATS rd_cost_y_arr[MAX_INTER_MODES];
1311 /*!
1312 * The full rd stats of chroma only for each of the candidate modes.
1313 */
1314 RD_STATS rd_cost_uv_arr[MAX_INTER_MODES];
1315 } InterModesInfo;
1316
1317 /*!\cond */
1318 typedef struct {
1319 // TODO(kyslov): consider changing to 64bit
1320
1321 // This struct is used for computing variance in choose_partitioning(), where
1322 // the max number of samples within a superblock is 32x32 (with 4x4 avg).
1323 // With 8bit bitdepth, uint32_t is enough for sum_square_error (2^8 * 2^8 * 32
1324 // * 32 = 2^26). For high bitdepth we need to consider changing this to 64 bit
1325 uint32_t sum_square_error;
1326 int32_t sum_error;
1327 int log2_count;
1328 int variance;
1329 } VPartVar;
1330
1331 typedef struct {
1332 VPartVar none;
1333 VPartVar horz[2];
1334 VPartVar vert[2];
1335 } VPVariance;
1336
1337 typedef struct {
1338 VPVariance part_variances;
1339 VPartVar split[4];
1340 } VP4x4;
1341
1342 typedef struct {
1343 VPVariance part_variances;
1344 VP4x4 split[4];
1345 } VP8x8;
1346
1347 typedef struct {
1348 VPVariance part_variances;
1349 VP8x8 split[4];
1350 } VP16x16;
1351
1352 typedef struct {
1353 VPVariance part_variances;
1354 VP16x16 split[4];
1355 } VP32x32;
1356
1357 typedef struct {
1358 VPVariance part_variances;
1359 VP32x32 split[4];
1360 } VP64x64;
1361
1362 typedef struct {
1363 VPVariance part_variances;
1364 VP64x64 *split;
1365 } VP128x128;
1366
1367 /*!\endcond */
1368
1369 /*!
1370 * \brief Thresholds for variance based partitioning.
1371 */
1372 typedef struct {
1373 /*!
1374 * If block variance > threshold, then that block is forced to split.
1375 * thresholds[0] - threshold for 128x128;
1376 * thresholds[1] - threshold for 64x64;
1377 * thresholds[2] - threshold for 32x32;
1378 * thresholds[3] - threshold for 16x16;
1379 * thresholds[4] - threshold for 8x8;
1380 */
1381 int64_t thresholds[5];
1382
1383 /*!
1384 * MinMax variance threshold for 8x8 sub blocks of a 16x16 block. If actual
1385 * minmax > threshold_minmax, the 16x16 is forced to split.
1386 */
1387 int64_t threshold_minmax;
1388 } VarBasedPartitionInfo;
1389
1390 /*!
1391 * \brief Encoder parameters for synchronization of row based multi-threading
1392 */
1393 typedef struct {
1394 #if CONFIG_MULTITHREAD
1395 /**
1396 * \name Synchronization objects for top-right dependency.
1397 */
1398 /**@{*/
1399 pthread_mutex_t *mutex_; /*!< Mutex lock object */
1400 pthread_cond_t *cond_; /*!< Condition variable */
1401 /**@}*/
1402 #endif // CONFIG_MULTITHREAD
1403 /*!
1404 * Buffer to store the superblock whose encoding is complete.
1405 * num_finished_cols[i] stores the number of superblocks which finished
1406 * encoding in the ith superblock row.
1407 */
1408 int *num_finished_cols;
1409 /*!
1410 * Denotes the superblock interval at which conditional signalling should
1411 * happen. Also denotes the minimum number of extra superblocks of the top row
1412 * to be complete to start encoding the current superblock. A value of 1
1413 * indicates top-right dependency.
1414 */
1415 int sync_range;
1416 /*!
1417 * Denotes the additional number of superblocks in the previous row to be
1418 * complete to start encoding the current superblock when intraBC tool is
1419 * enabled. This additional top-right delay is required to satisfy the
1420 * hardware constraints for intraBC tool when row multithreading is enabled.
1421 */
1422 int intrabc_extra_top_right_sb_delay;
1423 /*!
1424 * Number of superblock rows.
1425 */
1426 int rows;
1427 /*!
1428 * The superblock row (in units of MI blocks) to be processed next.
1429 */
1430 int next_mi_row;
1431 /*!
1432 * Number of threads processing the current tile.
1433 */
1434 int num_threads_working;
1435 } AV1EncRowMultiThreadSync;
1436
1437 /*!\cond */
1438
1439 // TODO(jingning) All spatially adaptive variables should go to TileDataEnc.
1440 typedef struct TileDataEnc {
1441 TileInfo tile_info;
1442 DECLARE_ALIGNED(16, FRAME_CONTEXT, tctx);
1443 FRAME_CONTEXT *row_ctx;
1444 uint64_t abs_sum_level;
1445 uint8_t allow_update_cdf;
1446 InterModeRdModel inter_mode_rd_models[BLOCK_SIZES_ALL];
1447 AV1EncRowMultiThreadSync row_mt_sync;
1448 MV firstpass_top_mv;
1449 } TileDataEnc;
1450
1451 typedef struct RD_COUNTS {
1452 int compound_ref_used_flag;
1453 int skip_mode_used_flag;
1454 int tx_type_used[TX_SIZES_ALL][TX_TYPES];
1455 int obmc_used[BLOCK_SIZES_ALL][2];
1456 int warped_used[2];
1457 int newmv_or_intra_blocks;
1458 uint64_t seg_tmp_pred_cost[2];
1459 } RD_COUNTS;
1460
1461 typedef struct ThreadData {
1462 MACROBLOCK mb;
1463 MvCosts *mv_costs_alloc;
1464 IntraBCMVCosts *dv_costs_alloc;
1465 RD_COUNTS rd_counts;
1466 FRAME_COUNTS *counts;
1467 PC_TREE_SHARED_BUFFERS shared_coeff_buf;
1468 SIMPLE_MOTION_DATA_TREE *sms_tree;
1469 SIMPLE_MOTION_DATA_TREE *sms_root;
1470 uint32_t *hash_value_buffer[2][2];
1471 OBMCBuffer obmc_buffer;
1472 PALETTE_BUFFER *palette_buffer;
1473 CompoundTypeRdBuffers comp_rd_buffer;
1474 CONV_BUF_TYPE *tmp_conv_dst;
1475 uint64_t abs_sum_level;
1476 uint8_t *tmp_pred_bufs[2];
1477 uint8_t *wiener_tmp_pred_buf;
1478 int intrabc_used;
1479 int deltaq_used;
1480 int coefficient_size;
1481 int max_mv_magnitude;
1482 int interp_filter_selected[SWITCHABLE];
1483 FRAME_CONTEXT *tctx;
1484 VP64x64 *vt64x64;
1485 int32_t num_64x64_blocks;
1486 PICK_MODE_CONTEXT *firstpass_ctx;
1487 TemporalFilterData tf_data;
1488 TplBuffers tpl_tmp_buffers;
1489 TplTxfmStats tpl_txfm_stats;
1490 GlobalMotionData gm_data;
1491 // Pointer to the array of structures to store gradient information of each
1492 // pixel in a superblock. The buffer constitutes of MAX_SB_SQUARE pixel level
1493 // structures for each of the plane types (PLANE_TYPE_Y and PLANE_TYPE_UV).
1494 PixelLevelGradientInfo *pixel_gradient_info;
1495 // Pointer to the array of structures to store source variance information of
1496 // each 4x4 sub-block in a superblock. Block4x4VarInfo structure is used to
1497 // store source variance and log of source variance of each 4x4 sub-block
1498 // for subsequent retrieval.
1499 Block4x4VarInfo *src_var_info_of_4x4_sub_blocks;
1500 // Pointer to pc tree root.
1501 PC_TREE *pc_root;
1502 } ThreadData;
1503
1504 struct EncWorkerData;
1505
1506 /*!\endcond */
1507
1508 /*!
1509 * \brief Encoder data related to row-based multi-threading
1510 */
1511 typedef struct {
1512 /*!
1513 * Number of tile rows for which row synchronization memory is allocated.
1514 */
1515 int allocated_tile_rows;
1516 /*!
1517 * Number of tile cols for which row synchronization memory is allocated.
1518 */
1519 int allocated_tile_cols;
1520 /*!
1521 * Number of rows for which row synchronization memory is allocated
1522 * per tile. During first-pass/look-ahead stage this equals the
1523 * maximum number of macroblock rows in a tile. During encode stage,
1524 * this equals the maximum number of superblock rows in a tile.
1525 */
1526 int allocated_rows;
1527 /*!
1528 * Number of columns for which entropy context memory is allocated
1529 * per tile. During encode stage, this equals the maximum number of
1530 * superblock columns in a tile minus 1. The entropy context memory
1531 * is not allocated during first-pass/look-ahead stage.
1532 */
1533 int allocated_cols;
1534
1535 /*!
1536 * thread_id_to_tile_id[i] indicates the tile id assigned to the ith thread.
1537 */
1538 int thread_id_to_tile_id[MAX_NUM_THREADS];
1539
1540 /*!
1541 * num_tile_cols_done[i] indicates the number of tile columns whose encoding
1542 * is complete in the ith superblock row.
1543 */
1544 int *num_tile_cols_done;
1545
1546 /*!
1547 * Number of superblock rows in a frame for which 'num_tile_cols_done' is
1548 * allocated.
1549 */
1550 int allocated_sb_rows;
1551
1552 /*!
1553 * Initialized to false, set to true by the worker thread that encounters an
1554 * error in order to abort the processing of other worker threads.
1555 */
1556 bool row_mt_exit;
1557
1558 /*!
1559 * Initialized to false, set to true during first pass encoding by the worker
1560 * thread that encounters an error in order to abort the processing of other
1561 * worker threads.
1562 */
1563 bool firstpass_mt_exit;
1564
1565 /*!
1566 * Initialized to false, set to true in cal_mb_wiener_var_hook() by the worker
1567 * thread that encounters an error in order to abort the processing of other
1568 * worker threads.
1569 */
1570 bool mb_wiener_mt_exit;
1571
1572 #if CONFIG_MULTITHREAD
1573 /*!
1574 * Mutex lock used while dispatching jobs.
1575 */
1576 pthread_mutex_t *mutex_;
1577 /*!
1578 * Condition variable used to dispatch loopfilter jobs.
1579 */
1580 pthread_cond_t *cond_;
1581 #endif
1582
1583 /**
1584 * \name Row synchronization related function pointers.
1585 */
1586 /**@{*/
1587 /*!
1588 * Reader.
1589 */
1590 void (*sync_read_ptr)(AV1EncRowMultiThreadSync *const, int, int);
1591 /*!
1592 * Writer.
1593 */
1594 void (*sync_write_ptr)(AV1EncRowMultiThreadSync *const, int, int, int);
1595 /**@}*/
1596 } AV1EncRowMultiThreadInfo;
1597
1598 /*!
1599 * \brief Encoder data related to multi-threading for allintra deltaq-mode=3
1600 */
1601 typedef struct {
1602 #if CONFIG_MULTITHREAD
1603 /*!
1604 * Mutex lock used while dispatching jobs.
1605 */
1606 pthread_mutex_t *mutex_;
1607 /*!
1608 * Condition variable used to dispatch loopfilter jobs.
1609 */
1610 pthread_cond_t *cond_;
1611 #endif
1612
1613 /**
1614 * \name Row synchronization related function pointers for all intra mode
1615 */
1616 /**@{*/
1617 /*!
1618 * Reader.
1619 */
1620 void (*intra_sync_read_ptr)(AV1EncRowMultiThreadSync *const, int, int);
1621 /*!
1622 * Writer.
1623 */
1624 void (*intra_sync_write_ptr)(AV1EncRowMultiThreadSync *const, int, int, int);
1625 /**@}*/
1626 } AV1EncAllIntraMultiThreadInfo;
1627
1628 /*!
1629 * \brief Max number of recodes used to track the frame probabilities.
1630 */
1631 #define NUM_RECODES_PER_FRAME 10
1632
1633 /*!
1634 * \brief Max number of frames that can be encoded in a parallel encode set.
1635 */
1636 #define MAX_PARALLEL_FRAMES 4
1637
1638 /*!
1639 * \brief Buffers to be backed up during parallel encode set to be restored
1640 * later.
1641 */
1642 typedef struct RestoreStateBuffers {
1643 /*!
1644 * Backup of original CDEF srcbuf.
1645 */
1646 uint16_t *cdef_srcbuf;
1647
1648 /*!
1649 * Backup of original CDEF colbuf.
1650 */
1651 uint16_t *cdef_colbuf[MAX_MB_PLANE];
1652
1653 /*!
1654 * Backup of original LR rst_tmpbuf.
1655 */
1656 int32_t *rst_tmpbuf;
1657
1658 /*!
1659 * Backup of original LR rlbs.
1660 */
1661 RestorationLineBuffers *rlbs;
1662 } RestoreStateBuffers;
1663
1664 /*!
1665 * \brief Parameters related to restoration types.
1666 */
1667 typedef struct {
1668 /*!
1669 * Stores the best coefficients for Wiener restoration.
1670 */
1671 WienerInfo wiener;
1672
1673 /*!
1674 * Stores the best coefficients for Sgrproj restoration.
1675 */
1676 SgrprojInfo sgrproj;
1677
1678 /*!
1679 * The rtype to use for this unit given a frame rtype as index. Indices:
1680 * WIENER, SGRPROJ, SWITCHABLE.
1681 */
1682 RestorationType best_rtype[RESTORE_TYPES - 1];
1683 } RestUnitSearchInfo;
1684
1685 /*!
1686 * \brief Structure to hold search parameter per restoration unit and
1687 * intermediate buffer of Wiener filter used in pick filter stage of Loop
1688 * restoration.
1689 */
1690 typedef struct {
1691 /*!
1692 * Array of pointers to 'RestUnitSearchInfo' which holds data related to
1693 * restoration types.
1694 */
1695 RestUnitSearchInfo *rusi[MAX_MB_PLANE];
1696
1697 /*!
1698 * Buffer used to hold dgd-avg data during SIMD call of Wiener filter.
1699 */
1700 int16_t *dgd_avg;
1701 } AV1LrPickStruct;
1702
1703 /*!
1704 * \brief Primary Encoder parameters related to multi-threading.
1705 */
1706 typedef struct PrimaryMultiThreadInfo {
1707 /*!
1708 * Number of workers created for multi-threading.
1709 */
1710 int num_workers;
1711
1712 /*!
1713 * Number of workers used for different MT modules.
1714 */
1715 int num_mod_workers[NUM_MT_MODULES];
1716
1717 /*!
1718 * Synchronization object used to launch job in the worker thread.
1719 */
1720 AVxWorker *workers;
1721
1722 /*!
1723 * Data specific to each worker in encoder multi-threading.
1724 * tile_thr_data[i] stores the worker data of the ith thread.
1725 */
1726 struct EncWorkerData *tile_thr_data;
1727
1728 /*!
1729 * CDEF row multi-threading data.
1730 */
1731 AV1CdefWorkerData *cdef_worker;
1732
1733 /*!
1734 * Primary(Level 1) Synchronization object used to launch job in the worker
1735 * thread.
1736 */
1737 AVxWorker *p_workers[MAX_PARALLEL_FRAMES];
1738
1739 /*!
1740 * Number of primary workers created for multi-threading.
1741 */
1742 int p_num_workers;
1743
1744 /*!
1745 * Tracks the number of workers in encode stage multi-threading.
1746 */
1747 int prev_num_enc_workers;
1748 } PrimaryMultiThreadInfo;
1749
1750 /*!
1751 * \brief Encoder parameters related to multi-threading.
1752 */
1753 typedef struct MultiThreadInfo {
1754 /*!
1755 * Number of workers created for multi-threading.
1756 */
1757 int num_workers;
1758
1759 /*!
1760 * Number of workers used for different MT modules.
1761 */
1762 int num_mod_workers[NUM_MT_MODULES];
1763
1764 /*!
1765 * Synchronization object used to launch job in the worker thread.
1766 */
1767 AVxWorker *workers;
1768
1769 /*!
1770 * Data specific to each worker in encoder multi-threading.
1771 * tile_thr_data[i] stores the worker data of the ith thread.
1772 */
1773 struct EncWorkerData *tile_thr_data;
1774
1775 /*!
1776 * When set, indicates that row based multi-threading of the encoder is
1777 * enabled.
1778 */
1779 bool row_mt_enabled;
1780
1781 /*!
1782 * When set, indicates that multi-threading for bitstream packing is enabled.
1783 */
1784 bool pack_bs_mt_enabled;
1785
1786 /*!
1787 * Encoder row multi-threading data.
1788 */
1789 AV1EncRowMultiThreadInfo enc_row_mt;
1790
1791 /*!
1792 * Encoder multi-threading data for allintra mode in the preprocessing stage
1793 * when --deltaq-mode=3.
1794 */
1795 AV1EncAllIntraMultiThreadInfo intra_mt;
1796
1797 /*!
1798 * Tpl row multi-threading data.
1799 */
1800 AV1TplRowMultiThreadInfo tpl_row_mt;
1801
1802 /*!
1803 * Loop Filter multi-threading object.
1804 */
1805 AV1LfSync lf_row_sync;
1806
1807 /*!
1808 * Loop Restoration multi-threading object.
1809 */
1810 AV1LrSync lr_row_sync;
1811
1812 /*!
1813 * Pack bitstream multi-threading object.
1814 */
1815 AV1EncPackBSSync pack_bs_sync;
1816
1817 /*!
1818 * Global Motion multi-threading object.
1819 */
1820 AV1GlobalMotionSync gm_sync;
1821
1822 /*!
1823 * Temporal Filter multi-threading object.
1824 */
1825 AV1TemporalFilterSync tf_sync;
1826
1827 /*!
1828 * CDEF search multi-threading object.
1829 */
1830 AV1CdefSync cdef_sync;
1831
1832 /*!
1833 * Pointer to CDEF row multi-threading data for the frame.
1834 */
1835 AV1CdefWorkerData *cdef_worker;
1836
1837 /*!
1838 * Buffers to be stored/restored before/after parallel encode.
1839 */
1840 RestoreStateBuffers restore_state_buf;
1841
1842 /*!
1843 * In multi-threaded realtime encoding with row-mt enabled, pipeline
1844 * loop-filtering after encoding.
1845 */
1846 int pipeline_lpf_mt_with_enc;
1847 } MultiThreadInfo;
1848
1849 /*!\cond */
1850
1851 typedef struct ActiveMap {
1852 int enabled;
1853 int update;
1854 unsigned char *map;
1855 } ActiveMap;
1856
1857 /*!\endcond */
1858
1859 /*!
1860 * \brief Encoder info used for decision on forcing integer motion vectors.
1861 */
1862 typedef struct {
1863 /*!
1864 * cs_rate_array[i] is the fraction of blocks in a frame which either match
1865 * with the collocated block or are smooth, where i is the rate_index.
1866 */
1867 double cs_rate_array[32];
1868 /*!
1869 * rate_index is used to index cs_rate_array.
1870 */
1871 int rate_index;
1872 /*!
1873 * rate_size is the total number of entries populated in cs_rate_array.
1874 */
1875 int rate_size;
1876 } ForceIntegerMVInfo;
1877
1878 /*!\cond */
1879
1880 #if CONFIG_INTERNAL_STATS
1881 // types of stats
1882 enum {
1883 STAT_Y,
1884 STAT_U,
1885 STAT_V,
1886 STAT_ALL,
1887 NUM_STAT_TYPES // This should always be the last member of the enum
1888 } UENUM1BYTE(StatType);
1889
1890 typedef struct IMAGE_STAT {
1891 double stat[NUM_STAT_TYPES];
1892 double worst;
1893 } ImageStat;
1894 #endif // CONFIG_INTERNAL_STATS
1895
1896 typedef struct {
1897 int ref_count;
1898 YV12_BUFFER_CONFIG buf;
1899 } EncRefCntBuffer;
1900
1901 /*!\endcond */
1902
1903 /*!
1904 * \brief Buffer to store mode information at mi_alloc_bsize (4x4 or 8x8) level
1905 *
1906 * This is used for bitstream preparation.
1907 */
1908 typedef struct {
1909 /*!
1910 * frame_base[mi_row * stride + mi_col] stores the mode information of
1911 * block (mi_row,mi_col).
1912 */
1913 MB_MODE_INFO_EXT_FRAME *frame_base;
1914 /*!
1915 * Size of frame_base buffer.
1916 */
1917 int alloc_size;
1918 /*!
1919 * Stride of frame_base buffer.
1920 */
1921 int stride;
1922 } MBMIExtFrameBufferInfo;
1923
1924 /*!\cond */
1925
1926 #if CONFIG_COLLECT_PARTITION_STATS
1927 typedef struct FramePartitionTimingStats {
1928 int partition_decisions[6][EXT_PARTITION_TYPES];
1929 int partition_attempts[6][EXT_PARTITION_TYPES];
1930 int64_t partition_times[6][EXT_PARTITION_TYPES];
1931
1932 int partition_redo;
1933 } FramePartitionTimingStats;
1934 #endif // CONFIG_COLLECT_PARTITION_STATS
1935
1936 #if CONFIG_COLLECT_COMPONENT_TIMING
1937 #include "aom_ports/aom_timer.h"
1938 // Adjust the following to add new components.
1939 enum {
1940 av1_encode_strategy_time,
1941 av1_get_one_pass_rt_params_time,
1942 av1_get_second_pass_params_time,
1943 denoise_and_encode_time,
1944 apply_filtering_time,
1945 av1_tpl_setup_stats_time,
1946 encode_frame_to_data_rate_time,
1947 encode_with_or_without_recode_time,
1948 loop_filter_time,
1949 cdef_time,
1950 loop_restoration_time,
1951 av1_pack_bitstream_final_time,
1952 av1_encode_frame_time,
1953 av1_compute_global_motion_time,
1954 av1_setup_motion_field_time,
1955 encode_sb_row_time,
1956
1957 rd_pick_partition_time,
1958 rd_use_partition_time,
1959 choose_var_based_partitioning_time,
1960 av1_prune_partitions_time,
1961 none_partition_search_time,
1962 split_partition_search_time,
1963 rectangular_partition_search_time,
1964 ab_partitions_search_time,
1965 rd_pick_4partition_time,
1966 encode_sb_time,
1967
1968 rd_pick_sb_modes_time,
1969 av1_rd_pick_intra_mode_sb_time,
1970 av1_rd_pick_inter_mode_sb_time,
1971 set_params_rd_pick_inter_mode_time,
1972 skip_inter_mode_time,
1973 handle_inter_mode_time,
1974 evaluate_motion_mode_for_winner_candidates_time,
1975 do_tx_search_time,
1976 handle_intra_mode_time,
1977 refine_winner_mode_tx_time,
1978 av1_search_palette_mode_time,
1979 handle_newmv_time,
1980 compound_type_rd_time,
1981 interpolation_filter_search_time,
1982 motion_mode_rd_time,
1983
1984 nonrd_use_partition_time,
1985 pick_sb_modes_nonrd_time,
1986 hybrid_intra_mode_search_time,
1987 nonrd_pick_inter_mode_sb_time,
1988 encode_b_nonrd_time,
1989
1990 kTimingComponents,
1991 } UENUM1BYTE(TIMING_COMPONENT);
1992
get_component_name(int index)1993 static inline char const *get_component_name(int index) {
1994 switch (index) {
1995 case av1_encode_strategy_time: return "av1_encode_strategy_time";
1996 case av1_get_one_pass_rt_params_time:
1997 return "av1_get_one_pass_rt_params_time";
1998 case av1_get_second_pass_params_time:
1999 return "av1_get_second_pass_params_time";
2000 case denoise_and_encode_time: return "denoise_and_encode_time";
2001 case apply_filtering_time: return "apply_filtering_time";
2002 case av1_tpl_setup_stats_time: return "av1_tpl_setup_stats_time";
2003 case encode_frame_to_data_rate_time:
2004 return "encode_frame_to_data_rate_time";
2005 case encode_with_or_without_recode_time:
2006 return "encode_with_or_without_recode_time";
2007 case loop_filter_time: return "loop_filter_time";
2008 case cdef_time: return "cdef_time";
2009 case loop_restoration_time: return "loop_restoration_time";
2010 case av1_pack_bitstream_final_time: return "av1_pack_bitstream_final_time";
2011 case av1_encode_frame_time: return "av1_encode_frame_time";
2012 case av1_compute_global_motion_time:
2013 return "av1_compute_global_motion_time";
2014 case av1_setup_motion_field_time: return "av1_setup_motion_field_time";
2015 case encode_sb_row_time: return "encode_sb_row_time";
2016
2017 case rd_pick_partition_time: return "rd_pick_partition_time";
2018 case rd_use_partition_time: return "rd_use_partition_time";
2019 case choose_var_based_partitioning_time:
2020 return "choose_var_based_partitioning_time";
2021 case av1_prune_partitions_time: return "av1_prune_partitions_time";
2022 case none_partition_search_time: return "none_partition_search_time";
2023 case split_partition_search_time: return "split_partition_search_time";
2024 case rectangular_partition_search_time:
2025 return "rectangular_partition_search_time";
2026 case ab_partitions_search_time: return "ab_partitions_search_time";
2027 case rd_pick_4partition_time: return "rd_pick_4partition_time";
2028 case encode_sb_time: return "encode_sb_time";
2029
2030 case rd_pick_sb_modes_time: return "rd_pick_sb_modes_time";
2031 case av1_rd_pick_intra_mode_sb_time:
2032 return "av1_rd_pick_intra_mode_sb_time";
2033 case av1_rd_pick_inter_mode_sb_time:
2034 return "av1_rd_pick_inter_mode_sb_time";
2035 case set_params_rd_pick_inter_mode_time:
2036 return "set_params_rd_pick_inter_mode_time";
2037 case skip_inter_mode_time: return "skip_inter_mode_time";
2038 case handle_inter_mode_time: return "handle_inter_mode_time";
2039 case evaluate_motion_mode_for_winner_candidates_time:
2040 return "evaluate_motion_mode_for_winner_candidates_time";
2041 case do_tx_search_time: return "do_tx_search_time";
2042 case handle_intra_mode_time: return "handle_intra_mode_time";
2043 case refine_winner_mode_tx_time: return "refine_winner_mode_tx_time";
2044 case av1_search_palette_mode_time: return "av1_search_palette_mode_time";
2045 case handle_newmv_time: return "handle_newmv_time";
2046 case compound_type_rd_time: return "compound_type_rd_time";
2047 case interpolation_filter_search_time:
2048 return "interpolation_filter_search_time";
2049 case motion_mode_rd_time: return "motion_mode_rd_time";
2050
2051 case nonrd_use_partition_time: return "nonrd_use_partition_time";
2052 case pick_sb_modes_nonrd_time: return "pick_sb_modes_nonrd_time";
2053 case hybrid_intra_mode_search_time: return "hybrid_intra_mode_search_time";
2054 case nonrd_pick_inter_mode_sb_time: return "nonrd_pick_inter_mode_sb_time";
2055 case encode_b_nonrd_time: return "encode_b_nonrd_time";
2056
2057 default: assert(0);
2058 }
2059 return "error";
2060 }
2061 #endif
2062
2063 // The maximum number of internal ARFs except ALTREF_FRAME
2064 #define MAX_INTERNAL_ARFS (REF_FRAMES - BWDREF_FRAME - 1)
2065
2066 /*!\endcond */
2067
2068 /*!
2069 * \brief Parameters related to global motion search
2070 */
2071 typedef struct {
2072 /*!
2073 * Flag to indicate if global motion search needs to be rerun.
2074 */
2075 bool search_done;
2076
2077 /*!
2078 * Array of pointers to the frame buffers holding the reference frames.
2079 * ref_buf[i] stores the pointer to the reference frame of the ith
2080 * reference frame type.
2081 */
2082 YV12_BUFFER_CONFIG *ref_buf[REF_FRAMES];
2083
2084 /*!
2085 * Holds the number of valid reference frames in past and future directions
2086 * w.r.t. the current frame. num_ref_frames[i] stores the total number of
2087 * valid reference frames in 'i' direction.
2088 */
2089 int num_ref_frames[MAX_DIRECTIONS];
2090
2091 /*!
2092 * Array of structure which stores the valid reference frames in past and
2093 * future directions and their corresponding distance from the source frame.
2094 * reference_frames[i][j] holds the jth valid reference frame type in the
2095 * direction 'i' and its temporal distance from the source frame .
2096 */
2097 FrameDistPair reference_frames[MAX_DIRECTIONS][REF_FRAMES - 1];
2098
2099 /**
2100 * \name Dimensions for which segment map is allocated.
2101 */
2102 /**@{*/
2103 int segment_map_w; /*!< segment map width */
2104 int segment_map_h; /*!< segment map height */
2105 /**@}*/
2106 } GlobalMotionInfo;
2107
2108 /*!
2109 * \brief Flags related to interpolation filter search
2110 */
2111 typedef struct {
2112 /*!
2113 * Stores the default value of skip flag depending on chroma format
2114 * Set as 1 for monochrome and 3 for other color formats
2115 */
2116 int default_interp_skip_flags;
2117 /*!
2118 * Filter mask to allow certain interp_filter type.
2119 */
2120 uint16_t interp_filter_search_mask;
2121 } InterpSearchFlags;
2122
2123 /*!
2124 * \brief Parameters for motion vector search process
2125 */
2126 typedef struct {
2127 /*!
2128 * Largest MV component used in a frame.
2129 * The value from the previous frame is used to set the full pixel search
2130 * range for the current frame.
2131 */
2132 int max_mv_magnitude;
2133 /*!
2134 * Parameter indicating initial search window to be used in full-pixel search.
2135 * Range [0, MAX_MVSEARCH_STEPS-2]. Lower value indicates larger window.
2136 */
2137 int mv_step_param;
2138 /*!
2139 * Pointer to sub-pixel search function.
2140 * In encoder: av1_find_best_sub_pixel_tree
2141 * av1_find_best_sub_pixel_tree_pruned
2142 * av1_find_best_sub_pixel_tree_pruned_more
2143 * In MV unit test: av1_return_max_sub_pixel_mv
2144 * av1_return_min_sub_pixel_mv
2145 */
2146 fractional_mv_step_fp *find_fractional_mv_step;
2147 /*!
2148 * Search site configuration for full-pel MV search.
2149 * search_site_cfg[SS_CFG_SRC]: Used in tpl, rd/non-rd inter mode loop, simple
2150 * motion search. search_site_cfg[SS_CFG_LOOKAHEAD]: Used in intraBC, temporal
2151 * filter search_site_cfg[SS_CFG_FPF]: Used during first pass and lookahead
2152 */
2153 search_site_config search_site_cfg[SS_CFG_TOTAL][NUM_DISTINCT_SEARCH_METHODS];
2154 } MotionVectorSearchParams;
2155
2156 /*!
2157 * \brief Refresh frame flags for different type of frames.
2158 *
2159 * If the refresh flag is true for a particular reference frame, after the
2160 * current frame is encoded, the reference frame gets refreshed (updated) to
2161 * be the current frame. Note: Usually at most one flag will be set to true at
2162 * a time. But, for key-frames, all flags are set to true at once.
2163 */
2164 typedef struct {
2165 bool golden_frame; /*!< Refresh flag for golden frame */
2166 bool bwd_ref_frame; /*!< Refresh flag for bwd-ref frame */
2167 bool alt_ref_frame; /*!< Refresh flag for alt-ref frame */
2168 } RefreshFrameInfo;
2169
2170 /*!
2171 * \brief Desired dimensions for an externally triggered resize.
2172 *
2173 * When resize is triggered externally, the desired dimensions are stored in
2174 * this struct until used in the next frame to be coded. These values are
2175 * effective only for one frame and are reset after they are used.
2176 */
2177 typedef struct {
2178 int width; /*!< Desired resized width */
2179 int height; /*!< Desired resized height */
2180 } ResizePendingParams;
2181
2182 /*!
2183 * \brief Refrence frame distance related variables.
2184 */
2185 typedef struct {
2186 /*!
2187 * True relative distance of reference frames w.r.t. the current frame.
2188 */
2189 int ref_relative_dist[INTER_REFS_PER_FRAME];
2190 /*!
2191 * The nearest reference w.r.t. current frame in the past.
2192 */
2193 int8_t nearest_past_ref;
2194 /*!
2195 * The nearest reference w.r.t. current frame in the future.
2196 */
2197 int8_t nearest_future_ref;
2198 } RefFrameDistanceInfo;
2199
2200 /*!
2201 * \brief Parameters used for winner mode processing.
2202 *
2203 * This is a basic two pass approach: in the first pass, we reduce the number of
2204 * transform searches based on some thresholds during the rdopt process to find
2205 * the "winner mode". In the second pass, we perform a more through tx search
2206 * on the winner mode.
2207 * There are some arrays in the struct, and their indices are used in the
2208 * following manner:
2209 * Index 0: Default mode evaluation, Winner mode processing is not applicable
2210 * (Eg : IntraBc).
2211 * Index 1: Mode evaluation.
2212 * Index 2: Winner mode evaluation
2213 * Index 1 and 2 are only used when the respective speed feature is on.
2214 */
2215 typedef struct {
2216 /*!
2217 * Threshold to determine if trellis optimization is to be enabled
2218 * based on :
2219 * 0 : dist threshold
2220 * 1 : satd threshold
2221 * Corresponds to enable_winner_mode_for_coeff_opt speed feature.
2222 */
2223 unsigned int coeff_opt_thresholds[MODE_EVAL_TYPES][2];
2224
2225 /*!
2226 * Determines the tx size search method during rdopt.
2227 * Corresponds to enable_winner_mode_for_tx_size_srch speed feature.
2228 */
2229 TX_SIZE_SEARCH_METHOD tx_size_search_methods[MODE_EVAL_TYPES];
2230
2231 /*!
2232 * Controls how often we should approximate prediction error with tx
2233 * coefficients. If it's 0, then never. If 1, then it's during the tx_type
2234 * search only. If 2, then always.
2235 * Corresponds to tx_domain_dist_level speed feature.
2236 */
2237 unsigned int use_transform_domain_distortion[MODE_EVAL_TYPES];
2238
2239 /*!
2240 * Threshold to approximate pixel domain distortion with transform domain
2241 * distortion. This is only used if use_transform_domain_distortion is on.
2242 * Corresponds to enable_winner_mode_for_use_tx_domain_dist speed feature.
2243 */
2244 unsigned int tx_domain_dist_threshold[MODE_EVAL_TYPES];
2245
2246 /*!
2247 * Controls how often we should try to skip the transform process based on
2248 * result from dct.
2249 * Corresponds to use_skip_flag_prediction speed feature.
2250 */
2251 unsigned int skip_txfm_level[MODE_EVAL_TYPES];
2252
2253 /*!
2254 * Predict DC only txfm blocks for default, mode and winner mode evaluation.
2255 * Index 0: Default mode evaluation, Winner mode processing is not applicable.
2256 * Index 1: Mode evaluation, Index 2: Winner mode evaluation
2257 */
2258 unsigned int predict_dc_level[MODE_EVAL_TYPES];
2259 } WinnerModeParams;
2260
2261 /*!
2262 * \brief Frame refresh flags set by the external interface.
2263 *
2264 * Flags set by external interface to determine which reference buffers are
2265 * refreshed by this frame. When set, the encoder will update the particular
2266 * reference frame buffer with the contents of the current frame.
2267 */
2268 typedef struct {
2269 bool last_frame; /*!< Refresh flag for last frame */
2270 bool golden_frame; /*!< Refresh flag for golden frame */
2271 bool bwd_ref_frame; /*!< Refresh flag for bwd-ref frame */
2272 bool alt2_ref_frame; /*!< Refresh flag for alt2-ref frame */
2273 bool alt_ref_frame; /*!< Refresh flag for alt-ref frame */
2274 /*!
2275 * Flag indicating if the update of refresh frame flags is pending.
2276 */
2277 bool update_pending;
2278 } ExtRefreshFrameFlagsInfo;
2279
2280 /*!
2281 * \brief Flags signalled by the external interface at frame level.
2282 */
2283 typedef struct {
2284 /*!
2285 * Bit mask to disable certain reference frame types.
2286 */
2287 int ref_frame_flags;
2288
2289 /*!
2290 * Frame refresh flags set by the external interface.
2291 */
2292 ExtRefreshFrameFlagsInfo refresh_frame;
2293
2294 /*!
2295 * Flag to enable the update of frame contexts at the end of a frame decode.
2296 */
2297 bool refresh_frame_context;
2298
2299 /*!
2300 * Flag to indicate that update of refresh_frame_context from external
2301 * interface is pending.
2302 */
2303 bool refresh_frame_context_pending;
2304
2305 /*!
2306 * Flag to enable temporal MV prediction.
2307 */
2308 bool use_ref_frame_mvs;
2309
2310 /*!
2311 * Indicates whether the current frame is to be coded as error resilient.
2312 */
2313 bool use_error_resilient;
2314
2315 /*!
2316 * Indicates whether the current frame is to be coded as s-frame.
2317 */
2318 bool use_s_frame;
2319
2320 /*!
2321 * Indicates whether the current frame's primary_ref_frame is set to
2322 * PRIMARY_REF_NONE.
2323 */
2324 bool use_primary_ref_none;
2325 } ExternalFlags;
2326
2327 /*!\cond */
2328
2329 typedef struct {
2330 // Some misc info
2331 int high_prec;
2332 int q;
2333 int order;
2334
2335 // MV counters
2336 int inter_count;
2337 int intra_count;
2338 int default_mvs;
2339 int mv_joint_count[4];
2340 int last_bit_zero;
2341 int last_bit_nonzero;
2342
2343 // Keep track of the rates
2344 int total_mv_rate;
2345 int hp_total_mv_rate;
2346 int lp_total_mv_rate;
2347
2348 // Texture info
2349 int horz_text;
2350 int vert_text;
2351 int diag_text;
2352
2353 // Whether the current struct contains valid data
2354 int valid;
2355 } MV_STATS;
2356
2357 typedef struct WeberStats {
2358 int64_t mb_wiener_variance;
2359 int64_t src_variance;
2360 int64_t rec_variance;
2361 int16_t src_pix_max;
2362 int16_t rec_pix_max;
2363 int64_t distortion;
2364 int64_t satd;
2365 double max_scale;
2366 } WeberStats;
2367
2368 typedef struct {
2369 struct loopfilter lf;
2370 CdefInfo cdef_info;
2371 YV12_BUFFER_CONFIG copy_buffer;
2372 RATE_CONTROL rc;
2373 MV_STATS mv_stats;
2374 } CODING_CONTEXT;
2375
2376 typedef struct {
2377 int frame_width;
2378 int frame_height;
2379 int mi_rows;
2380 int mi_cols;
2381 int mb_rows;
2382 int mb_cols;
2383 int num_mbs;
2384 aom_bit_depth_t bit_depth;
2385 int subsampling_x;
2386 int subsampling_y;
2387 } FRAME_INFO;
2388
2389 /*!
2390 * \brief This structure stores different types of frame indices.
2391 */
2392 typedef struct {
2393 int show_frame_count;
2394 } FRAME_INDEX_SET;
2395
2396 /*!\endcond */
2397
2398 /*!
2399 * \brief Segmentation related information for the current frame.
2400 */
2401 typedef struct {
2402 /*!
2403 * 3-bit number containing the segment affiliation for each 4x4 block in the
2404 * frame. map[y * stride + x] contains the segment id of the 4x4 block at
2405 * (x,y) position.
2406 */
2407 uint8_t *map;
2408 /*!
2409 * Flag to indicate if current frame has lossless segments or not.
2410 * 1: frame has at least one lossless segment.
2411 * 0: frame has no lossless segments.
2412 */
2413 bool has_lossless_segment;
2414 } EncSegmentationInfo;
2415
2416 /*!
2417 * \brief Frame time stamps.
2418 */
2419 typedef struct {
2420 /*!
2421 * Start time stamp of the previous frame
2422 */
2423 int64_t prev_ts_start;
2424 /*!
2425 * End time stamp of the previous frame
2426 */
2427 int64_t prev_ts_end;
2428 /*!
2429 * Start time stamp of the first frame
2430 */
2431 int64_t first_ts_start;
2432 } TimeStamps;
2433
2434 /*!
2435 * Pointers to the memory allocated for frame level transform coeff related
2436 * info.
2437 */
2438 typedef struct {
2439 /*!
2440 * Pointer to the transformed coefficients buffer.
2441 */
2442 tran_low_t *tcoeff;
2443 /*!
2444 * Pointer to the eobs buffer.
2445 */
2446 uint16_t *eobs;
2447 /*!
2448 * Pointer to the entropy_ctx buffer.
2449 */
2450 uint8_t *entropy_ctx;
2451 } CoeffBufferPool;
2452
2453 #if !CONFIG_REALTIME_ONLY
2454 /*!\cond */
2455 // DUCKY_ENCODE_FRAME_MODE is c version of EncodeFrameMode
2456 enum {
2457 DUCKY_ENCODE_FRAME_MODE_NONE, // Let native AV1 determine q index and rdmult
2458 DUCKY_ENCODE_FRAME_MODE_QINDEX, // DuckyEncode determines q index and AV1
2459 // determines rdmult
2460 DUCKY_ENCODE_FRAME_MODE_QINDEX_RDMULT, // DuckyEncode determines q index and
2461 // rdmult
2462 } UENUM1BYTE(DUCKY_ENCODE_FRAME_MODE);
2463
2464 enum {
2465 DUCKY_ENCODE_GOP_MODE_NONE, // native AV1 decides GOP
2466 DUCKY_ENCODE_GOP_MODE_RCL, // rate control lib decides GOP
2467 } UENUM1BYTE(DUCKY_ENCODE_GOP_MODE);
2468
2469 typedef struct DuckyEncodeFrameInfo {
2470 DUCKY_ENCODE_FRAME_MODE qp_mode;
2471 DUCKY_ENCODE_GOP_MODE gop_mode;
2472 int q_index;
2473 int rdmult;
2474 // These two arrays are equivalent to std::vector<SuperblockEncodeParameters>
2475 int *superblock_encode_qindex;
2476 int *superblock_encode_rdmult;
2477 int delta_q_enabled;
2478 } DuckyEncodeFrameInfo;
2479
2480 typedef struct DuckyEncodeFrameResult {
2481 int global_order_idx;
2482 int q_index;
2483 int rdmult;
2484 int rate;
2485 int64_t dist;
2486 double psnr;
2487 } DuckyEncodeFrameResult;
2488
2489 typedef struct DuckyEncodeInfo {
2490 DuckyEncodeFrameInfo frame_info;
2491 DuckyEncodeFrameResult frame_result;
2492 } DuckyEncodeInfo;
2493 /*!\endcond */
2494 #endif
2495
2496 /*!\cond */
2497 typedef struct RTC_REF {
2498 /*!
2499 * LAST_FRAME (0), LAST2_FRAME(1), LAST3_FRAME(2), GOLDEN_FRAME(3),
2500 * BWDREF_FRAME(4), ALTREF2_FRAME(5), ALTREF_FRAME(6).
2501 */
2502 int reference[INTER_REFS_PER_FRAME];
2503 int ref_idx[INTER_REFS_PER_FRAME];
2504 int refresh[REF_FRAMES];
2505 int set_ref_frame_config;
2506 int non_reference_frame;
2507 int ref_frame_comp[3];
2508 int gld_idx_1layer;
2509 /*!
2510 * Frame number of the last frame that refreshed the buffer slot.
2511 */
2512 unsigned int buffer_time_index[REF_FRAMES];
2513 /*!
2514 * Spatial layer id of the last frame that refreshed the buffer slot.
2515 */
2516 unsigned char buffer_spatial_layer[REF_FRAMES];
2517 /*!
2518 * Flag to indicate whether closest reference was the previous frame.
2519 */
2520 bool reference_was_previous_frame;
2521 /*!
2522 * Flag to indicate this frame is based on longer term reference only,
2523 * for recovery from past loss, and it should be biased for improved coding.
2524 */
2525 bool bias_recovery_frame;
2526 } RTC_REF;
2527 /*!\endcond */
2528
2529 /*!
2530 * \brief Structure to hold data corresponding to an encoded frame.
2531 */
2532 typedef struct AV1_COMP_DATA {
2533 /*!
2534 * Buffer to store packed bitstream data of a frame.
2535 */
2536 unsigned char *cx_data;
2537
2538 /*!
2539 * Allocated size of the cx_data buffer.
2540 */
2541 size_t cx_data_sz;
2542
2543 /*!
2544 * Size of data written in the cx_data buffer.
2545 */
2546 size_t frame_size;
2547
2548 /*!
2549 * Flags for the frame.
2550 */
2551 unsigned int lib_flags;
2552
2553 /*!
2554 * Time stamp for start of frame.
2555 */
2556 int64_t ts_frame_start;
2557
2558 /*!
2559 * Time stamp for end of frame.
2560 */
2561 int64_t ts_frame_end;
2562
2563 /*!
2564 * Flag to indicate flush call.
2565 */
2566 int flush;
2567
2568 /*!
2569 * Time base for sequence.
2570 */
2571 const aom_rational64_t *timestamp_ratio;
2572
2573 /*!
2574 * Decide to pop the source for this frame from input buffer queue.
2575 */
2576 int pop_lookahead;
2577 } AV1_COMP_DATA;
2578
2579 /*!
2580 * \brief Top level primary encoder structure
2581 */
2582 typedef struct AV1_PRIMARY {
2583 /*!
2584 * Array of frame level encoder stage top level structures
2585 */
2586 struct AV1_COMP *parallel_cpi[MAX_PARALLEL_FRAMES];
2587
2588 /*!
2589 * Array of structures to hold data of frames encoded in a given parallel
2590 * encode set.
2591 */
2592 struct AV1_COMP_DATA parallel_frames_data[MAX_PARALLEL_FRAMES - 1];
2593 #if CONFIG_FPMT_TEST
2594 /*!
2595 * Flag which enables/disables simulation path for fpmt unit test.
2596 * 0 - FPMT integration
2597 * 1 - FPMT simulation
2598 */
2599 FPMT_TEST_ENC_CFG fpmt_unit_test_cfg;
2600
2601 /*!
2602 * Temporary variable simulating the delayed frame_probability update.
2603 */
2604 FrameProbInfo temp_frame_probs;
2605
2606 /*!
2607 * Temporary variable holding the updated frame probability across
2608 * frames. Copy its value to temp_frame_probs for frame_parallel_level 0
2609 * frames or last frame in parallel encode set.
2610 */
2611 FrameProbInfo temp_frame_probs_simulation;
2612
2613 /*!
2614 * Temporary variable simulating the delayed update of valid global motion
2615 * model across frames.
2616 */
2617 int temp_valid_gm_model_found[FRAME_UPDATE_TYPES];
2618 #endif // CONFIG_FPMT_TEST
2619 /*!
2620 * Copy of cm->ref_frame_map maintained to facilitate sequential update of
2621 * ref_frame_map by lower layer depth frames encoded ahead of time in a
2622 * parallel encode set.
2623 */
2624 RefCntBuffer *ref_frame_map_copy[REF_FRAMES];
2625
2626 /*!
2627 * Start time stamp of the last encoded show frame
2628 */
2629 int64_t ts_start_last_show_frame;
2630
2631 /*!
2632 * End time stamp of the last encoded show frame
2633 */
2634 int64_t ts_end_last_show_frame;
2635
2636 /*!
2637 * Number of frame level contexts(cpis)
2638 */
2639 int num_fp_contexts;
2640
2641 /*!
2642 * Loopfilter levels of the previous encoded frame.
2643 */
2644 int filter_level[2];
2645
2646 /*!
2647 * Chrominance component loopfilter level of the previous encoded frame.
2648 */
2649 int filter_level_u;
2650
2651 /*!
2652 * Chrominance component loopfilter level of the previous encoded frame.
2653 */
2654 int filter_level_v;
2655
2656 /*!
2657 * Encode stage top level structure
2658 * During frame parallel encode, this is the same as parallel_cpi[0]
2659 */
2660 struct AV1_COMP *cpi;
2661
2662 /*!
2663 * Lookahead processing stage top level structure
2664 */
2665 struct AV1_COMP *cpi_lap;
2666
2667 /*!
2668 * Look-ahead context.
2669 */
2670 struct lookahead_ctx *lookahead;
2671
2672 /*!
2673 * Sequence parameters have been transmitted already and locked
2674 * or not. Once locked av1_change_config cannot change the seq
2675 * parameters.
2676 */
2677 int seq_params_locked;
2678
2679 /*!
2680 * Pointer to internal utility functions that manipulate aom_codec_* data
2681 * structures.
2682 */
2683 struct aom_codec_pkt_list *output_pkt_list;
2684
2685 /*!
2686 * When set, indicates that internal ARFs are enabled.
2687 */
2688 int internal_altref_allowed;
2689
2690 /*!
2691 * Tell if OVERLAY frame shows existing alt_ref frame.
2692 */
2693 int show_existing_alt_ref;
2694
2695 /*!
2696 * Information related to a gf group.
2697 */
2698 GF_GROUP gf_group;
2699
2700 /*!
2701 * Track prior gf group state.
2702 */
2703 GF_STATE gf_state;
2704
2705 /*!
2706 * Flag indicating whether look ahead processing (LAP) is enabled.
2707 */
2708 int lap_enabled;
2709
2710 /*!
2711 * Parameters for AV1 bitstream levels.
2712 */
2713 AV1LevelParams level_params;
2714
2715 /*!
2716 * Calculates PSNR on each frame when set to 1.
2717 */
2718 int b_calculate_psnr;
2719
2720 /*!
2721 * Number of frames left to be encoded, is 0 if limit is not set.
2722 */
2723 int frames_left;
2724
2725 /*!
2726 * Information related to two pass encoding.
2727 */
2728 TWO_PASS twopass;
2729
2730 /*!
2731 * Rate control related parameters.
2732 */
2733 PRIMARY_RATE_CONTROL p_rc;
2734
2735 /*!
2736 * Info and resources used by temporal filtering.
2737 */
2738 TEMPORAL_FILTER_INFO tf_info;
2739 /*!
2740 * Elements part of the sequence header, that are applicable for all the
2741 * frames in the video.
2742 */
2743 SequenceHeader seq_params;
2744
2745 /*!
2746 * Indicates whether to use SVC.
2747 */
2748 int use_svc;
2749
2750 /*!
2751 * If true, buffer removal times are present.
2752 */
2753 bool buffer_removal_time_present;
2754
2755 /*!
2756 * Number of temporal layers: may be > 1 for SVC (scalable vector coding).
2757 */
2758 unsigned int number_temporal_layers;
2759
2760 /*!
2761 * Number of spatial layers: may be > 1 for SVC (scalable vector coding).
2762 */
2763 unsigned int number_spatial_layers;
2764
2765 /*!
2766 * Code and details about current error status.
2767 */
2768 struct aom_internal_error_info error;
2769
2770 /*!
2771 * Function pointers to variants of sse/sad/variance computation functions.
2772 * fn_ptr[i] indicates the list of function pointers corresponding to block
2773 * size i.
2774 */
2775 aom_variance_fn_ptr_t fn_ptr[BLOCK_SIZES_ALL];
2776
2777 /*!
2778 * tpl_sb_rdmult_scaling_factors[i] stores the RD multiplier scaling factor of
2779 * the ith 16 x 16 block in raster scan order.
2780 */
2781 double *tpl_sb_rdmult_scaling_factors;
2782
2783 /*!
2784 * Parameters related to tpl.
2785 */
2786 TplParams tpl_data;
2787
2788 /*!
2789 * Motion vector stats of the previous encoded frame.
2790 */
2791 MV_STATS mv_stats;
2792
2793 #if CONFIG_INTERNAL_STATS
2794 /*!\cond */
2795 uint64_t total_time_receive_data;
2796 uint64_t total_time_compress_data;
2797
2798 unsigned int total_mode_chosen_counts[MAX_MODES];
2799
2800 int count[2];
2801 uint64_t total_sq_error[2];
2802 uint64_t total_samples[2];
2803 ImageStat psnr[2];
2804
2805 double total_blockiness;
2806 double worst_blockiness;
2807
2808 uint64_t total_bytes;
2809 double summed_quality;
2810 double summed_weights;
2811 double summed_quality_hbd;
2812 double summed_weights_hbd;
2813 unsigned int total_recode_hits;
2814 double worst_ssim;
2815 double worst_ssim_hbd;
2816
2817 ImageStat fastssim;
2818 ImageStat psnrhvs;
2819
2820 int b_calculate_blockiness;
2821 int b_calculate_consistency;
2822
2823 double total_inconsistency;
2824 double worst_consistency;
2825 Ssimv *ssim_vars;
2826 Metrics metrics;
2827 /*!\endcond */
2828 #endif
2829
2830 #if CONFIG_ENTROPY_STATS
2831 /*!
2832 * Aggregates frame counts for the sequence.
2833 */
2834 FRAME_COUNTS aggregate_fc;
2835 #endif // CONFIG_ENTROPY_STATS
2836
2837 /*!
2838 * For each type of reference frame, this contains the index of a reference
2839 * frame buffer for a reference frame of the same type. We use this to
2840 * choose our primary reference frame (which is the most recent reference
2841 * frame of the same type as the current frame).
2842 */
2843 int fb_of_context_type[REF_FRAMES];
2844
2845 /*!
2846 * Primary Multi-threading parameters.
2847 */
2848 PrimaryMultiThreadInfo p_mt_info;
2849
2850 /*!
2851 * Probabilities for pruning of various AV1 tools.
2852 */
2853 FrameProbInfo frame_probs;
2854
2855 /*!
2856 * Indicates if a valid global motion model has been found in the different
2857 * frame update types of a GF group.
2858 * valid_gm_model_found[i] indicates if valid global motion model has been
2859 * found in the frame update type with enum value equal to i
2860 */
2861 int valid_gm_model_found[FRAME_UPDATE_TYPES];
2862
2863 /*!
2864 * Struct for the reference structure for RTC.
2865 */
2866 RTC_REF rtc_ref;
2867
2868 /*!
2869 * Struct for all intra mode row multi threading in the preprocess stage
2870 * when --deltaq-mode=3.
2871 */
2872 AV1EncRowMultiThreadSync intra_row_mt_sync;
2873 } AV1_PRIMARY;
2874
2875 /*!
2876 * \brief Top level encoder structure.
2877 */
2878 typedef struct AV1_COMP {
2879 /*!
2880 * Pointer to top level primary encoder structure
2881 */
2882 AV1_PRIMARY *ppi;
2883
2884 /*!
2885 * Quantization and dequantization parameters for internal quantizer setup
2886 * in the encoder.
2887 */
2888 EncQuantDequantParams enc_quant_dequant_params;
2889
2890 /*!
2891 * Structure holding thread specific variables.
2892 */
2893 ThreadData td;
2894
2895 /*!
2896 * Statistics collected at frame level.
2897 */
2898 FRAME_COUNTS counts;
2899
2900 /*!
2901 * Holds buffer storing mode information at 4x4/8x8 level.
2902 */
2903 MBMIExtFrameBufferInfo mbmi_ext_info;
2904
2905 /*!
2906 * Buffer holding the transform block related information.
2907 * coeff_buffer_base[i] stores the transform block related information of the
2908 * ith superblock in raster scan order.
2909 */
2910 CB_COEFF_BUFFER *coeff_buffer_base;
2911
2912 /*!
2913 * Structure holding pointers to frame level memory allocated for transform
2914 * block related information.
2915 */
2916 CoeffBufferPool coeff_buffer_pool;
2917
2918 /*!
2919 * Structure holding variables common to encoder and decoder.
2920 */
2921 AV1_COMMON common;
2922
2923 /*!
2924 * Encoder configuration related parameters.
2925 */
2926 AV1EncoderConfig oxcf;
2927
2928 /*!
2929 * Stores the trellis optimization type at segment level.
2930 * optimize_seg_arr[i] stores the trellis opt type for ith segment.
2931 */
2932 TRELLIS_OPT_TYPE optimize_seg_arr[MAX_SEGMENTS];
2933
2934 /*!
2935 * Pointer to the frame buffer holding the source frame to be used during the
2936 * current stage of encoding. It can be the raw input, temporally filtered
2937 * input or scaled input.
2938 */
2939 YV12_BUFFER_CONFIG *source;
2940
2941 /*!
2942 * Pointer to the frame buffer holding the last raw source frame.
2943 * last_source is NULL for the following cases:
2944 * 1) First frame
2945 * 2) Alt-ref frames
2946 * 3) All frames for all-intra frame encoding.
2947 */
2948 YV12_BUFFER_CONFIG *last_source;
2949
2950 /*!
2951 * Pointer to the frame buffer holding the unscaled source frame.
2952 * It can be either the raw input or temporally filtered input.
2953 */
2954 YV12_BUFFER_CONFIG *unscaled_source;
2955
2956 /*!
2957 * Frame buffer holding the resized source frame (cropping / superres).
2958 */
2959 YV12_BUFFER_CONFIG scaled_source;
2960
2961 /*!
2962 * Pointer to the frame buffer holding the unscaled last source frame.
2963 */
2964 YV12_BUFFER_CONFIG *unscaled_last_source;
2965
2966 /*!
2967 * Frame buffer holding the resized last source frame.
2968 */
2969 YV12_BUFFER_CONFIG scaled_last_source;
2970
2971 /*!
2972 * Pointer to the original source frame. This is used to determine if the
2973 * content is screen.
2974 */
2975 YV12_BUFFER_CONFIG *unfiltered_source;
2976
2977 /*!
2978 * Frame buffer holding the orig source frame for PSNR calculation in rtc tf
2979 * case.
2980 */
2981 YV12_BUFFER_CONFIG orig_source;
2982
2983 /*!
2984 * Skip tpl setup when tpl data from gop length decision can be reused.
2985 */
2986 int skip_tpl_setup_stats;
2987
2988 /*!
2989 * Scaling factors used in the RD multiplier modulation.
2990 * TODO(sdeng): consider merge the following arrays.
2991 * tpl_rdmult_scaling_factors is a temporary buffer used to store the
2992 * intermediate scaling factors which are used in the calculation of
2993 * tpl_sb_rdmult_scaling_factors. tpl_rdmult_scaling_factors[i] stores the
2994 * intermediate scaling factor of the ith 16 x 16 block in raster scan order.
2995 */
2996 double *tpl_rdmult_scaling_factors;
2997
2998 /*!
2999 * Temporal filter context.
3000 */
3001 TemporalFilterCtx tf_ctx;
3002
3003 /*!
3004 * Pointer to CDEF search context.
3005 */
3006 CdefSearchCtx *cdef_search_ctx;
3007
3008 /*!
3009 * Variables related to forcing integer mv decisions for the current frame.
3010 */
3011 ForceIntegerMVInfo force_intpel_info;
3012
3013 /*!
3014 * Pointer to the buffer holding the scaled reference frames.
3015 * scaled_ref_buf[i] holds the scaled reference frame of type i.
3016 */
3017 RefCntBuffer *scaled_ref_buf[INTER_REFS_PER_FRAME];
3018
3019 /*!
3020 * Pointer to the buffer holding the last show frame.
3021 */
3022 RefCntBuffer *last_show_frame_buf;
3023
3024 /*!
3025 * Refresh frame flags for golden, bwd-ref and alt-ref frames.
3026 */
3027 RefreshFrameInfo refresh_frame;
3028
3029 /*!
3030 * Flag to reduce the number of reference frame buffers used in rt.
3031 */
3032 int rt_reduce_num_ref_buffers;
3033
3034 /*!
3035 * Flags signalled by the external interface at frame level.
3036 */
3037 ExternalFlags ext_flags;
3038
3039 /*!
3040 * Temporary frame buffer used to store the non-loop filtered reconstructed
3041 * frame during the search of loop filter level.
3042 */
3043 YV12_BUFFER_CONFIG last_frame_uf;
3044
3045 /*!
3046 * Temporary frame buffer used to store the loop restored frame during loop
3047 * restoration search.
3048 */
3049 YV12_BUFFER_CONFIG trial_frame_rst;
3050
3051 /*!
3052 * Ambient reconstruction err target for force key frames.
3053 */
3054 int64_t ambient_err;
3055
3056 /*!
3057 * Parameters related to rate distortion optimization.
3058 */
3059 RD_OPT rd;
3060
3061 /*!
3062 * Temporary coding context used to save and restore when encoding with and
3063 * without super-resolution.
3064 */
3065 CODING_CONTEXT coding_context;
3066
3067 /*!
3068 * Parameters related to global motion search.
3069 */
3070 GlobalMotionInfo gm_info;
3071
3072 /*!
3073 * Parameters related to winner mode processing.
3074 */
3075 WinnerModeParams winner_mode_params;
3076
3077 /*!
3078 * Frame time stamps.
3079 */
3080 TimeStamps time_stamps;
3081
3082 /*!
3083 * Rate control related parameters.
3084 */
3085 RATE_CONTROL rc;
3086
3087 /*!
3088 * Frame rate of the video.
3089 */
3090 double framerate;
3091
3092 /*!
3093 * Bitmask indicating which reference buffers may be referenced by this frame.
3094 */
3095 int ref_frame_flags;
3096
3097 /*!
3098 * speed is passed as a per-frame parameter into the encoder.
3099 */
3100 int speed;
3101
3102 /*!
3103 * sf contains fine-grained config set internally based on speed.
3104 */
3105 SPEED_FEATURES sf;
3106
3107 /*!
3108 * Parameters for motion vector search process.
3109 */
3110 MotionVectorSearchParams mv_search_params;
3111
3112 /*!
3113 * When set, indicates that all reference frames are forward references,
3114 * i.e., all the reference frames are output before the current frame.
3115 */
3116 int all_one_sided_refs;
3117
3118 /*!
3119 * Segmentation related information for current frame.
3120 */
3121 EncSegmentationInfo enc_seg;
3122
3123 /*!
3124 * Parameters related to cyclic refresh aq-mode.
3125 */
3126 CYCLIC_REFRESH *cyclic_refresh;
3127 /*!
3128 * Parameters related to active map. Active maps indicate
3129 * if there is any activity on a 4x4 block basis.
3130 */
3131 ActiveMap active_map;
3132
3133 /*!
3134 * The frame processing order within a GOP.
3135 */
3136 unsigned char gf_frame_index;
3137
3138 #if CONFIG_INTERNAL_STATS
3139 /*!\cond */
3140 uint64_t time_compress_data;
3141
3142 unsigned int mode_chosen_counts[MAX_MODES];
3143 int bytes;
3144 unsigned int frame_recode_hits;
3145 /*!\endcond */
3146 #endif
3147
3148 #if CONFIG_SPEED_STATS
3149 /*!
3150 * For debugging: number of transform searches we have performed.
3151 */
3152 unsigned int tx_search_count;
3153 #endif // CONFIG_SPEED_STATS
3154
3155 /*!
3156 * When set, indicates that the frame is droppable, i.e., this frame
3157 * does not update any reference buffers.
3158 */
3159 int droppable;
3160
3161 /*!
3162 * Stores the frame parameters during encoder initialization.
3163 */
3164 FRAME_INFO frame_info;
3165
3166 /*!
3167 * Stores different types of frame indices.
3168 */
3169 FRAME_INDEX_SET frame_index_set;
3170
3171 /*!
3172 * Stores the cm->width in the last call of alloc_compressor_data(). Helps
3173 * determine whether compressor data should be reallocated when cm->width
3174 * changes.
3175 */
3176 int data_alloc_width;
3177
3178 /*!
3179 * Stores the cm->height in the last call of alloc_compressor_data(). Helps
3180 * determine whether compressor data should be reallocated when cm->height
3181 * changes.
3182 */
3183 int data_alloc_height;
3184
3185 /*!
3186 * Number of MBs in the full-size frame; to be used to
3187 * normalize the firstpass stats. This will differ from the
3188 * number of MBs in the current frame when the frame is
3189 * scaled.
3190 */
3191 int initial_mbs;
3192
3193 /*!
3194 * Flag to indicate whether the frame size inforamation has been
3195 * setup and propagated to associated allocations.
3196 */
3197 bool frame_size_related_setup_done;
3198
3199 /*!
3200 * The width of the frame that is lastly encoded.
3201 * It is updated in the function "encoder_encode()".
3202 */
3203 int last_coded_width;
3204
3205 /*!
3206 * The height of the frame that is lastly encoded.
3207 * It is updated in the function "encoder_encode()".
3208 */
3209 int last_coded_height;
3210
3211 /*!
3212 * Resize related parameters.
3213 */
3214 ResizePendingParams resize_pending_params;
3215
3216 /*!
3217 * Pointer to struct holding adaptive data/contexts/models for the tile during
3218 * encoding.
3219 */
3220 TileDataEnc *tile_data;
3221 /*!
3222 * Number of tiles for which memory has been allocated for tile_data.
3223 */
3224 int allocated_tiles;
3225
3226 /*!
3227 * Structure to store the palette token related information.
3228 */
3229 TokenInfo token_info;
3230
3231 /*!
3232 * VARIANCE_AQ segment map refresh.
3233 */
3234 int vaq_refresh;
3235
3236 /*!
3237 * Thresholds for variance based partitioning.
3238 */
3239 VarBasedPartitionInfo vbp_info;
3240
3241 /*!
3242 * Number of recodes in the frame.
3243 */
3244 int num_frame_recode;
3245
3246 /*!
3247 * Current frame probability of parallel frames, across recodes.
3248 */
3249 FrameProbInfo frame_new_probs[NUM_RECODES_PER_FRAME];
3250
3251 /*!
3252 * Retain condition for transform type frame_probability calculation
3253 */
3254 int do_update_frame_probs_txtype[NUM_RECODES_PER_FRAME];
3255
3256 /*!
3257 * Retain condition for obmc frame_probability calculation
3258 */
3259 int do_update_frame_probs_obmc[NUM_RECODES_PER_FRAME];
3260
3261 /*!
3262 * Retain condition for warped motion frame_probability calculation
3263 */
3264 int do_update_frame_probs_warp[NUM_RECODES_PER_FRAME];
3265
3266 /*!
3267 * Retain condition for interpolation filter frame_probability calculation
3268 */
3269 int do_update_frame_probs_interpfilter[NUM_RECODES_PER_FRAME];
3270
3271 #if CONFIG_FPMT_TEST
3272 /*!
3273 * Temporary variable for simulation.
3274 * Previous frame's framerate.
3275 */
3276 double temp_framerate;
3277 #endif
3278 /*!
3279 * Updated framerate for the current parallel frame.
3280 * cpi->framerate is updated with new_framerate during
3281 * post encode updates for parallel frames.
3282 */
3283 double new_framerate;
3284
3285 /*!
3286 * Retain condition for fast_extra_bits calculation.
3287 */
3288 int do_update_vbr_bits_off_target_fast;
3289
3290 /*!
3291 * Multi-threading parameters.
3292 */
3293 MultiThreadInfo mt_info;
3294
3295 /*!
3296 * Specifies the frame to be output. It is valid only if show_existing_frame
3297 * is 1. When show_existing_frame is 0, existing_fb_idx_to_show is set to
3298 * INVALID_IDX.
3299 */
3300 int existing_fb_idx_to_show;
3301
3302 /*!
3303 * A flag to indicate if intrabc is ever used in current frame.
3304 */
3305 int intrabc_used;
3306
3307 /*!
3308 * Mark which ref frames can be skipped for encoding current frame during RDO.
3309 */
3310 int prune_ref_frame_mask;
3311
3312 /*!
3313 * Loop Restoration context.
3314 */
3315 AV1LrStruct lr_ctxt;
3316
3317 /*!
3318 * Loop Restoration context used during pick stage.
3319 */
3320 AV1LrPickStruct pick_lr_ctxt;
3321
3322 /*!
3323 * Pointer to list of tables with film grain parameters.
3324 */
3325 aom_film_grain_table_t *film_grain_table;
3326
3327 #if CONFIG_DENOISE
3328 /*!
3329 * Pointer to structure holding the denoised image buffers and the helper
3330 * noise models.
3331 */
3332 struct aom_denoise_and_model_t *denoise_and_model;
3333 #endif
3334
3335 /*!
3336 * Flags related to interpolation filter search.
3337 */
3338 InterpSearchFlags interp_search_flags;
3339
3340 /*!
3341 * Turn on screen content tools flag.
3342 * Note that some videos are not screen content videos, but
3343 * screen content tools could also improve coding efficiency.
3344 * For example, videos with large flat regions, gaming videos that look
3345 * like natural videos.
3346 */
3347 int use_screen_content_tools;
3348
3349 /*!
3350 * A flag to indicate "real" screen content videos.
3351 * For example, screen shares, screen editing.
3352 * This type is true indicates |use_screen_content_tools| must be true.
3353 * In addition, rate control strategy is adjusted when this flag is true.
3354 */
3355 int is_screen_content_type;
3356
3357 #if CONFIG_COLLECT_PARTITION_STATS
3358 /*!
3359 * Accumulates the partition timing stat over the whole frame.
3360 */
3361 FramePartitionTimingStats partition_stats;
3362 #endif // CONFIG_COLLECT_PARTITION_STATS
3363
3364 #if CONFIG_COLLECT_COMPONENT_TIMING
3365 /*!
3366 * component_time[] are initialized to zero while encoder starts.
3367 */
3368 uint64_t component_time[kTimingComponents];
3369 /*!
3370 * Stores timing for individual components between calls of start_timing()
3371 * and end_timing().
3372 */
3373 struct aom_usec_timer component_timer[kTimingComponents];
3374 /*!
3375 * frame_component_time[] are initialized to zero at beginning of each frame.
3376 */
3377 uint64_t frame_component_time[kTimingComponents];
3378 #endif
3379
3380 /*!
3381 * Count the number of OBU_FRAME and OBU_FRAME_HEADER for level calculation.
3382 */
3383 int frame_header_count;
3384
3385 /*!
3386 * Whether any no-zero delta_q was actually used.
3387 */
3388 int deltaq_used;
3389
3390 /*!
3391 * Refrence frame distance related variables.
3392 */
3393 RefFrameDistanceInfo ref_frame_dist_info;
3394
3395 /*!
3396 * ssim_rdmult_scaling_factors[i] stores the RD multiplier scaling factor of
3397 * the ith 16 x 16 block in raster scan order. This scaling factor is used for
3398 * RD multiplier modulation when SSIM tuning is enabled.
3399 */
3400 double *ssim_rdmult_scaling_factors;
3401
3402 #if CONFIG_TUNE_VMAF
3403 /*!
3404 * Parameters for VMAF tuning.
3405 */
3406 TuneVMAFInfo vmaf_info;
3407 #endif
3408
3409 #if CONFIG_TUNE_BUTTERAUGLI
3410 /*!
3411 * Parameters for Butteraugli tuning.
3412 */
3413 TuneButteraugliInfo butteraugli_info;
3414 #endif
3415
3416 /*!
3417 * Parameters for scalable video coding.
3418 */
3419 SVC svc;
3420
3421 /*!
3422 * Indicates whether current processing stage is encode stage or LAP stage.
3423 */
3424 COMPRESSOR_STAGE compressor_stage;
3425
3426 /*!
3427 * Frame type of the last frame. May be used in some heuristics for speeding
3428 * up the encoding.
3429 */
3430 FRAME_TYPE last_frame_type;
3431
3432 /*!
3433 * Number of tile-groups.
3434 */
3435 int num_tg;
3436
3437 /*!
3438 * Super-resolution mode currently being used by the encoder.
3439 * This may / may not be same as user-supplied mode in oxcf->superres_mode
3440 * (when we are recoding to try multiple options for example).
3441 */
3442 aom_superres_mode superres_mode;
3443
3444 /*!
3445 * First pass related data.
3446 */
3447 FirstPassData firstpass_data;
3448
3449 /*!
3450 * Temporal Noise Estimate
3451 */
3452 NOISE_ESTIMATE noise_estimate;
3453
3454 #if CONFIG_AV1_TEMPORAL_DENOISING
3455 /*!
3456 * Temporal Denoiser
3457 */
3458 AV1_DENOISER denoiser;
3459 #endif
3460
3461 /*!
3462 * Count on how many consecutive times a block uses small/zeromv for encoding
3463 * in a scale of 8x8 block.
3464 */
3465 uint8_t *consec_zero_mv;
3466
3467 /*!
3468 * Allocated memory size for |consec_zero_mv|.
3469 */
3470 int consec_zero_mv_alloc_size;
3471
3472 /*!
3473 * Block size of first pass encoding
3474 */
3475 BLOCK_SIZE fp_block_size;
3476
3477 /*!
3478 * The counter of encoded super block, used to differentiate block names.
3479 * This number starts from 0 and increases whenever a super block is encoded.
3480 */
3481 int sb_counter;
3482
3483 /*!
3484 * Available bitstream buffer size in bytes
3485 */
3486 size_t available_bs_size;
3487
3488 /*!
3489 * The controller of the external partition model.
3490 * It is used to do partition type selection based on external models.
3491 */
3492 ExtPartController ext_part_controller;
3493
3494 /*!
3495 * Motion vector stats of the current encoded frame, used to update the
3496 * ppi->mv_stats during postencode.
3497 */
3498 MV_STATS mv_stats;
3499 /*!
3500 * Stores the reference refresh index for the current frame.
3501 */
3502 int ref_refresh_index;
3503
3504 /*!
3505 * A flag to indicate if the reference refresh index is available for the
3506 * current frame.
3507 */
3508 bool refresh_idx_available;
3509
3510 /*!
3511 * Reference frame index corresponding to the frame to be excluded from being
3512 * used as a reference by frame_parallel_level 2 frame in a parallel
3513 * encode set of lower layer frames.
3514 */
3515 int ref_idx_to_skip;
3516 #if CONFIG_FPMT_TEST
3517 /*!
3518 * Stores the wanted frame buffer index for choosing primary ref frame by a
3519 * frame_parallel_level 2 frame in a parallel encode set of lower layer
3520 * frames.
3521 */
3522
3523 int wanted_fb;
3524 #endif // CONFIG_FPMT_TEST
3525
3526 /*!
3527 * A flag to indicate frames that will update their data to the primary
3528 * context at the end of the encode. It is set for non-parallel frames and the
3529 * last frame in encode order in a given parallel encode set.
3530 */
3531 bool do_frame_data_update;
3532
3533 #if CONFIG_RD_COMMAND
3534 /*!
3535 * A structure for assigning external q_index / rdmult for experiments
3536 */
3537 RD_COMMAND rd_command;
3538 #endif // CONFIG_RD_COMMAND
3539
3540 /*!
3541 * Buffer to store MB variance after Wiener filter.
3542 */
3543 WeberStats *mb_weber_stats;
3544
3545 /*!
3546 * Buffer to store rate cost estimates for each macro block (8x8) in the
3547 * preprocessing stage used in allintra mode.
3548 */
3549 int *prep_rate_estimates;
3550
3551 /*!
3552 * Buffer to store rate cost estimates for each 16x16 block read
3553 * from an external file, used in allintra mode.
3554 */
3555 double *ext_rate_distribution;
3556
3557 /*!
3558 * The scale that equals sum_rate_uniform_quantizer / sum_ext_rate.
3559 */
3560 double ext_rate_scale;
3561
3562 /*!
3563 * Buffer to store MB variance after Wiener filter.
3564 */
3565 BLOCK_SIZE weber_bsize;
3566
3567 /*!
3568 * Frame level Wiener filter normalization.
3569 */
3570 int64_t norm_wiener_variance;
3571
3572 /*!
3573 * Buffer to store delta-q values for delta-q mode 4.
3574 */
3575 int *mb_delta_q;
3576
3577 /*!
3578 * Flag to indicate that current frame is dropped.
3579 */
3580 bool is_dropped_frame;
3581
3582 #if CONFIG_BITRATE_ACCURACY
3583 /*!
3584 * Structure stores information needed for bitrate accuracy experiment.
3585 */
3586 VBR_RATECTRL_INFO vbr_rc_info;
3587 #endif
3588
3589 #if CONFIG_RATECTRL_LOG
3590 /*!
3591 * Structure stores information of rate control decisions.
3592 */
3593 RATECTRL_LOG rc_log;
3594 #endif // CONFIG_RATECTRL_LOG
3595
3596 /*!
3597 * Frame level twopass status and control data
3598 */
3599 TWO_PASS_FRAME twopass_frame;
3600
3601 #if CONFIG_THREE_PASS
3602 /*!
3603 * Context needed for third pass encoding.
3604 */
3605 THIRD_PASS_DEC_CTX *third_pass_ctx;
3606 #endif
3607
3608 /*!
3609 * File pointer to second pass log
3610 */
3611 FILE *second_pass_log_stream;
3612
3613 /*!
3614 * Buffer to store 64x64 SAD
3615 */
3616 uint64_t *src_sad_blk_64x64;
3617
3618 /*!
3619 * SSE between the current frame and the reconstructed last frame
3620 * It is only used for CBR mode.
3621 * It is not used if the reference frame has a different frame size.
3622 */
3623 uint64_t rec_sse;
3624
3625 /*!
3626 * A flag to indicate whether the encoder is controlled by DuckyEncode or not.
3627 * 1:yes 0:no
3628 */
3629 int use_ducky_encode;
3630
3631 #if !CONFIG_REALTIME_ONLY
3632 /*! A structure that facilitates the communication between DuckyEncode and AV1
3633 * encoder.
3634 */
3635 DuckyEncodeInfo ducky_encode_info;
3636 #endif // CONFIG_REALTIME_ONLY
3637 //
3638 /*!
3639 * Frames since last frame with cdf update.
3640 */
3641 int frames_since_last_update;
3642
3643 /*!
3644 * Block level thresholds to force zeromv-skip at partition level.
3645 */
3646 unsigned int zeromv_skip_thresh_exit_part[BLOCK_SIZES_ALL];
3647
3648 /*!
3649 * Should we allocate a downsampling pyramid for each frame buffer?
3650 * This is currently only used for global motion
3651 */
3652 bool alloc_pyramid;
3653
3654 #if CONFIG_SALIENCY_MAP
3655 /*!
3656 * Pixel level saliency map for each frame.
3657 */
3658 uint8_t *saliency_map;
3659
3660 /*!
3661 * Superblock level rdmult scaling factor driven by saliency map.
3662 */
3663 double *sm_scaling_factor;
3664 #endif
3665
3666 /*!
3667 * Number of pixels that choose palette mode for luma in the
3668 * fast encoding pass in av1_determine_sc_tools_with_encoding().
3669 */
3670 int palette_pixel_num;
3671
3672 /*!
3673 * Flag to indicate scaled_last_source is available,
3674 * so scaling is not needed for last_source.
3675 */
3676 int scaled_last_source_available;
3677 } AV1_COMP;
3678
3679 /*!
3680 * \brief Input frames and last input frame
3681 */
3682 typedef struct EncodeFrameInput {
3683 /*!\cond */
3684 YV12_BUFFER_CONFIG *source;
3685 YV12_BUFFER_CONFIG *last_source;
3686 int64_t ts_duration;
3687 /*!\endcond */
3688 } EncodeFrameInput;
3689
3690 /*!
3691 * \brief contains per-frame encoding parameters decided upon by
3692 * av1_encode_strategy() and passed down to av1_encode().
3693 */
3694 typedef struct EncodeFrameParams {
3695 /*!
3696 * Is error resilient mode enabled
3697 */
3698 int error_resilient_mode;
3699 /*!
3700 * Frame type (eg KF vs inter frame etc)
3701 */
3702 FRAME_TYPE frame_type;
3703
3704 /*!\cond */
3705 int primary_ref_frame;
3706 int order_offset;
3707
3708 /*!\endcond */
3709 /*!
3710 * Should the current frame be displayed after being decoded
3711 */
3712 int show_frame;
3713
3714 /*!\cond */
3715 int refresh_frame_flags;
3716
3717 int show_existing_frame;
3718 int existing_fb_idx_to_show;
3719
3720 /*!\endcond */
3721 /*!
3722 * Bitmask of which reference buffers may be referenced by this frame.
3723 */
3724 int ref_frame_flags;
3725
3726 /*!
3727 * Reference buffer assignment for this frame.
3728 */
3729 int remapped_ref_idx[REF_FRAMES];
3730
3731 /*!
3732 * Flags which determine which reference buffers are refreshed by this
3733 * frame.
3734 */
3735 RefreshFrameInfo refresh_frame;
3736
3737 /*!
3738 * Speed level to use for this frame: Bigger number means faster.
3739 */
3740 int speed;
3741 } EncodeFrameParams;
3742
3743 /*!\cond */
3744
3745 void av1_initialize_enc(unsigned int usage, enum aom_rc_mode end_usage);
3746
3747 struct AV1_COMP *av1_create_compressor(AV1_PRIMARY *ppi,
3748 const AV1EncoderConfig *oxcf,
3749 BufferPool *const pool,
3750 COMPRESSOR_STAGE stage,
3751 int lap_lag_in_frames);
3752
3753 struct AV1_PRIMARY *av1_create_primary_compressor(
3754 struct aom_codec_pkt_list *pkt_list_head, int num_lap_buffers,
3755 const AV1EncoderConfig *oxcf);
3756
3757 void av1_remove_compressor(AV1_COMP *cpi);
3758
3759 void av1_remove_primary_compressor(AV1_PRIMARY *ppi);
3760
3761 #if CONFIG_ENTROPY_STATS
3762 void print_entropy_stats(AV1_PRIMARY *const ppi);
3763 #endif
3764 #if CONFIG_INTERNAL_STATS
3765 void print_internal_stats(AV1_PRIMARY *ppi);
3766 #endif
3767
3768 void av1_change_config_seq(AV1_PRIMARY *ppi, const AV1EncoderConfig *oxcf,
3769 bool *sb_size_changed);
3770
3771 void av1_change_config(AV1_COMP *cpi, const AV1EncoderConfig *oxcf,
3772 bool sb_size_changed);
3773
3774 aom_codec_err_t av1_check_initial_width(AV1_COMP *cpi, int use_highbitdepth,
3775 int subsampling_x, int subsampling_y);
3776
3777 void av1_post_encode_updates(AV1_COMP *const cpi,
3778 const AV1_COMP_DATA *const cpi_data);
3779
3780 void av1_release_scaled_references_fpmt(AV1_COMP *cpi);
3781
3782 void av1_decrement_ref_counts_fpmt(BufferPool *buffer_pool,
3783 int ref_buffers_used_map);
3784
3785 void av1_init_sc_decisions(AV1_PRIMARY *const ppi);
3786
3787 AV1_COMP *av1_get_parallel_frame_enc_data(AV1_PRIMARY *const ppi,
3788 AV1_COMP_DATA *const first_cpi_data);
3789
3790 int av1_init_parallel_frame_context(const AV1_COMP_DATA *const first_cpi_data,
3791 AV1_PRIMARY *const ppi,
3792 int *ref_buffers_used_map);
3793
3794 /*!\endcond */
3795
3796 /*!\brief Obtain the raw frame data
3797 *
3798 * \ingroup high_level_algo
3799 * This function receives the raw frame data from input.
3800 *
3801 * \param[in] cpi Top-level encoder structure
3802 * \param[in] frame_flags Flags to decide how to encoding the frame
3803 * \param[in,out] sd Contain raw frame data
3804 * \param[in] time_stamp Time stamp of the frame
3805 * \param[in] end_time_stamp End time stamp
3806 *
3807 * \return Returns a value to indicate if the frame data is received
3808 * successfully.
3809 * \note The caller can assume that a copy of this frame is made and not just a
3810 * copy of the pointer.
3811 */
3812 int av1_receive_raw_frame(AV1_COMP *cpi, aom_enc_frame_flags_t frame_flags,
3813 const YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
3814 int64_t end_time_stamp);
3815
3816 /*!\brief Encode a frame
3817 *
3818 * \ingroup high_level_algo
3819 * \callgraph
3820 * \callergraph
3821 * This function encodes the raw frame data, and outputs the frame bit stream
3822 * to the designated buffer. The caller should use the output parameters
3823 * cpi_data->ts_frame_start and cpi_data->ts_frame_end only when this function
3824 * returns AOM_CODEC_OK.
3825 *
3826 * \param[in] cpi Top-level encoder structure
3827 * \param[in,out] cpi_data Data corresponding to a frame encode
3828 *
3829 * \return Returns a value to indicate if the encoding is done successfully.
3830 * \retval #AOM_CODEC_OK
3831 * \retval -1
3832 * No frame encoded; more input is required.
3833 * \retval "A nonzero (positive) aom_codec_err_t code"
3834 * The encoding failed with the error. Sets the error code and error message
3835 * in \c cpi->common.error.
3836 */
3837 int av1_get_compressed_data(AV1_COMP *cpi, AV1_COMP_DATA *const cpi_data);
3838
3839 /*!\brief Run 1-pass/2-pass encoding
3840 *
3841 * \ingroup high_level_algo
3842 * \callgraph
3843 * \callergraph
3844 */
3845 int av1_encode(AV1_COMP *const cpi, uint8_t *const dest, size_t dest_size,
3846 const EncodeFrameInput *const frame_input,
3847 const EncodeFrameParams *const frame_params,
3848 size_t *const frame_size);
3849
3850 /*!\cond */
3851 int av1_get_preview_raw_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *dest);
3852
3853 int av1_get_last_show_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *frame);
3854
3855 aom_codec_err_t av1_copy_new_frame_enc(AV1_COMMON *cm,
3856 YV12_BUFFER_CONFIG *new_frame,
3857 YV12_BUFFER_CONFIG *sd);
3858
3859 int av1_use_as_reference(int *ext_ref_frame_flags, int ref_frame_flags);
3860
3861 int av1_copy_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd);
3862
3863 int av1_set_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd);
3864
3865 void av1_set_frame_size(AV1_COMP *cpi, int width, int height);
3866
3867 void av1_set_mv_search_params(AV1_COMP *cpi);
3868
3869 int av1_set_active_map(AV1_COMP *cpi, unsigned char *map, int rows, int cols);
3870
3871 int av1_get_active_map(AV1_COMP *cpi, unsigned char *map, int rows, int cols);
3872
3873 int av1_set_internal_size(AV1EncoderConfig *const oxcf,
3874 ResizePendingParams *resize_pending_params,
3875 AOM_SCALING_MODE horiz_mode,
3876 AOM_SCALING_MODE vert_mode);
3877
3878 int av1_get_quantizer(struct AV1_COMP *cpi);
3879
3880 // This function assumes that the input buffer contains valid OBUs. It should
3881 // not be called on untrusted input.
3882 int av1_convert_sect5obus_to_annexb(uint8_t *buffer, size_t buffer_size,
3883 size_t *input_size);
3884
3885 void av1_alloc_mb_wiener_var_pred_buf(AV1_COMMON *cm, ThreadData *td);
3886
3887 void av1_dealloc_mb_wiener_var_pred_buf(ThreadData *td);
3888
3889 // Set screen content options.
3890 // This function estimates whether to use screen content tools, by counting
3891 // the portion of blocks that have few luma colors.
3892 // Modifies:
3893 // cpi->commom.features.allow_screen_content_tools
3894 // cpi->common.features.allow_intrabc
3895 // cpi->use_screen_content_tools
3896 // cpi->is_screen_content_type
3897 // However, the estimation is not accurate and may misclassify videos.
3898 // A slower but more accurate approach that determines whether to use screen
3899 // content tools is employed later. See av1_determine_sc_tools_with_encoding().
3900 void av1_set_screen_content_options(struct AV1_COMP *cpi,
3901 FeatureFlags *features);
3902
3903 void av1_update_frame_size(AV1_COMP *cpi);
3904
3905 typedef struct {
3906 int pyr_level;
3907 int disp_order;
3908 } RefFrameMapPair;
3909
init_ref_map_pair(AV1_COMP * cpi,RefFrameMapPair ref_frame_map_pairs[REF_FRAMES])3910 static inline void init_ref_map_pair(
3911 AV1_COMP *cpi, RefFrameMapPair ref_frame_map_pairs[REF_FRAMES]) {
3912 if (cpi->ppi->gf_group.update_type[cpi->gf_frame_index] == KF_UPDATE) {
3913 memset(ref_frame_map_pairs, -1, sizeof(*ref_frame_map_pairs) * REF_FRAMES);
3914 return;
3915 }
3916 memset(ref_frame_map_pairs, 0, sizeof(*ref_frame_map_pairs) * REF_FRAMES);
3917 for (int map_idx = 0; map_idx < REF_FRAMES; map_idx++) {
3918 // Get reference frame buffer.
3919 const RefCntBuffer *const buf = cpi->common.ref_frame_map[map_idx];
3920 if (ref_frame_map_pairs[map_idx].disp_order == -1) continue;
3921 if (buf == NULL) {
3922 ref_frame_map_pairs[map_idx].disp_order = -1;
3923 ref_frame_map_pairs[map_idx].pyr_level = -1;
3924 continue;
3925 } else if (buf->ref_count > 1) {
3926 // Once the keyframe is coded, the slots in ref_frame_map will all
3927 // point to the same frame. In that case, all subsequent pointers
3928 // matching the current are considered "free" slots. This will find
3929 // the next occurrence of the current pointer if ref_count indicates
3930 // there are multiple instances of it and mark it as free.
3931 for (int idx2 = map_idx + 1; idx2 < REF_FRAMES; ++idx2) {
3932 const RefCntBuffer *const buf2 = cpi->common.ref_frame_map[idx2];
3933 if (buf2 == buf) {
3934 ref_frame_map_pairs[idx2].disp_order = -1;
3935 ref_frame_map_pairs[idx2].pyr_level = -1;
3936 }
3937 }
3938 }
3939 ref_frame_map_pairs[map_idx].disp_order = (int)buf->display_order_hint;
3940 ref_frame_map_pairs[map_idx].pyr_level = buf->pyramid_level;
3941 }
3942 }
3943
3944 #if CONFIG_FPMT_TEST
calc_frame_data_update_flag(GF_GROUP * const gf_group,int gf_frame_index,bool * const do_frame_data_update)3945 static inline void calc_frame_data_update_flag(
3946 GF_GROUP *const gf_group, int gf_frame_index,
3947 bool *const do_frame_data_update) {
3948 *do_frame_data_update = true;
3949 // Set the flag to false for all frames in a given parallel encode set except
3950 // the last frame in the set with frame_parallel_level = 2.
3951 if (gf_group->frame_parallel_level[gf_frame_index] == 1) {
3952 *do_frame_data_update = false;
3953 } else if (gf_group->frame_parallel_level[gf_frame_index] == 2) {
3954 // Check if this is the last frame in the set with frame_parallel_level = 2.
3955 for (int i = gf_frame_index + 1; i < gf_group->size; i++) {
3956 if ((gf_group->frame_parallel_level[i] == 0 &&
3957 (gf_group->update_type[i] == ARF_UPDATE ||
3958 gf_group->update_type[i] == INTNL_ARF_UPDATE)) ||
3959 gf_group->frame_parallel_level[i] == 1) {
3960 break;
3961 } else if (gf_group->frame_parallel_level[i] == 2) {
3962 *do_frame_data_update = false;
3963 break;
3964 }
3965 }
3966 }
3967 }
3968 #endif
3969
3970 // av1 uses 10,000,000 ticks/second as time stamp
3971 #define TICKS_PER_SEC 10000000LL
3972
timebase_units_to_ticks(const aom_rational64_t * timestamp_ratio,int64_t n)3973 static inline int64_t timebase_units_to_ticks(
3974 const aom_rational64_t *timestamp_ratio, int64_t n) {
3975 return n * timestamp_ratio->num / timestamp_ratio->den;
3976 }
3977
ticks_to_timebase_units(const aom_rational64_t * timestamp_ratio,int64_t n)3978 static inline int64_t ticks_to_timebase_units(
3979 const aom_rational64_t *timestamp_ratio, int64_t n) {
3980 int64_t round = timestamp_ratio->num / 2;
3981 if (round > 0) --round;
3982 return (n * timestamp_ratio->den + round) / timestamp_ratio->num;
3983 }
3984
frame_is_kf_gf_arf(const AV1_COMP * cpi)3985 static inline int frame_is_kf_gf_arf(const AV1_COMP *cpi) {
3986 const GF_GROUP *const gf_group = &cpi->ppi->gf_group;
3987 const FRAME_UPDATE_TYPE update_type =
3988 gf_group->update_type[cpi->gf_frame_index];
3989
3990 return frame_is_intra_only(&cpi->common) || update_type == ARF_UPDATE ||
3991 update_type == GF_UPDATE;
3992 }
3993
3994 // TODO(huisu@google.com, youzhou@microsoft.com): enable hash-me for HBD.
av1_use_hash_me(const AV1_COMP * const cpi)3995 static inline int av1_use_hash_me(const AV1_COMP *const cpi) {
3996 return (cpi->common.features.allow_screen_content_tools &&
3997 cpi->common.features.allow_intrabc &&
3998 frame_is_intra_only(&cpi->common));
3999 }
4000
get_ref_frame_yv12_buf(const AV1_COMMON * const cm,MV_REFERENCE_FRAME ref_frame)4001 static inline const YV12_BUFFER_CONFIG *get_ref_frame_yv12_buf(
4002 const AV1_COMMON *const cm, MV_REFERENCE_FRAME ref_frame) {
4003 const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame);
4004 return buf != NULL ? &buf->buf : NULL;
4005 }
4006
alloc_frame_mvs(AV1_COMMON * const cm,RefCntBuffer * buf)4007 static inline void alloc_frame_mvs(AV1_COMMON *const cm, RefCntBuffer *buf) {
4008 assert(buf != NULL);
4009 ensure_mv_buffer(buf, cm);
4010 buf->width = cm->width;
4011 buf->height = cm->height;
4012 }
4013
4014 // Get the allocated token size for a tile. It does the same calculation as in
4015 // the frame token allocation.
allocated_tokens(const TileInfo * tile,int sb_size_log2,int num_planes)4016 static inline unsigned int allocated_tokens(const TileInfo *tile,
4017 int sb_size_log2, int num_planes) {
4018 int tile_mb_rows =
4019 ROUND_POWER_OF_TWO(tile->mi_row_end - tile->mi_row_start, 2);
4020 int tile_mb_cols =
4021 ROUND_POWER_OF_TWO(tile->mi_col_end - tile->mi_col_start, 2);
4022
4023 return get_token_alloc(tile_mb_rows, tile_mb_cols, sb_size_log2, num_planes);
4024 }
4025
get_start_tok(AV1_COMP * cpi,int tile_row,int tile_col,int mi_row,TokenExtra ** tok,int sb_size_log2,int num_planes)4026 static inline void get_start_tok(AV1_COMP *cpi, int tile_row, int tile_col,
4027 int mi_row, TokenExtra **tok, int sb_size_log2,
4028 int num_planes) {
4029 AV1_COMMON *const cm = &cpi->common;
4030 const int tile_cols = cm->tiles.cols;
4031 TileDataEnc *this_tile = &cpi->tile_data[tile_row * tile_cols + tile_col];
4032 const TileInfo *const tile_info = &this_tile->tile_info;
4033
4034 const int tile_mb_cols =
4035 (tile_info->mi_col_end - tile_info->mi_col_start + 2) >> 2;
4036 const int tile_mb_row = (mi_row - tile_info->mi_row_start + 2) >> 2;
4037
4038 *tok = cpi->token_info.tile_tok[tile_row][tile_col] +
4039 get_token_alloc(tile_mb_row, tile_mb_cols, sb_size_log2, num_planes);
4040 }
4041
4042 void av1_apply_encoding_flags(AV1_COMP *cpi, aom_enc_frame_flags_t flags);
4043
4044 #define ALT_MIN_LAG 3
is_altref_enabled(int lag_in_frames,bool enable_auto_arf)4045 static inline int is_altref_enabled(int lag_in_frames, bool enable_auto_arf) {
4046 return lag_in_frames >= ALT_MIN_LAG && enable_auto_arf;
4047 }
4048
can_disable_altref(const GFConfig * gf_cfg)4049 static inline int can_disable_altref(const GFConfig *gf_cfg) {
4050 return is_altref_enabled(gf_cfg->lag_in_frames, gf_cfg->enable_auto_arf) &&
4051 (gf_cfg->gf_min_pyr_height == 0);
4052 }
4053
4054 // Helper function to compute number of blocks on either side of the frame.
get_num_blocks(const int frame_length,const int mb_length)4055 static inline int get_num_blocks(const int frame_length, const int mb_length) {
4056 return (frame_length + mb_length - 1) / mb_length;
4057 }
4058
4059 // Check if statistics generation stage
is_stat_generation_stage(const AV1_COMP * const cpi)4060 static inline int is_stat_generation_stage(const AV1_COMP *const cpi) {
4061 assert(IMPLIES(cpi->compressor_stage == LAP_STAGE,
4062 cpi->oxcf.pass == AOM_RC_ONE_PASS && cpi->ppi->lap_enabled));
4063 return (cpi->oxcf.pass == AOM_RC_FIRST_PASS ||
4064 (cpi->compressor_stage == LAP_STAGE));
4065 }
4066 // Check if statistics consumption stage
is_stat_consumption_stage_twopass(const AV1_COMP * const cpi)4067 static inline int is_stat_consumption_stage_twopass(const AV1_COMP *const cpi) {
4068 return (cpi->oxcf.pass >= AOM_RC_SECOND_PASS);
4069 }
4070
4071 // Check if statistics consumption stage
is_stat_consumption_stage(const AV1_COMP * const cpi)4072 static inline int is_stat_consumption_stage(const AV1_COMP *const cpi) {
4073 return (is_stat_consumption_stage_twopass(cpi) ||
4074 (cpi->oxcf.pass == AOM_RC_ONE_PASS &&
4075 (cpi->compressor_stage == ENCODE_STAGE) && cpi->ppi->lap_enabled));
4076 }
4077
4078 // Decide whether 'dv_costs' need to be allocated/stored during the encoding.
av1_need_dv_costs(const AV1_COMP * const cpi)4079 static inline bool av1_need_dv_costs(const AV1_COMP *const cpi) {
4080 return !cpi->sf.rt_sf.use_nonrd_pick_mode &&
4081 av1_allow_intrabc(&cpi->common) && !is_stat_generation_stage(cpi);
4082 }
4083
4084 /*!\endcond */
4085 /*!\brief Check if the current stage has statistics
4086 *
4087 *\ingroup two_pass_algo
4088 *
4089 * \param[in] cpi Top - level encoder instance structure
4090 *
4091 * \return 0 if no stats for current stage else 1
4092 */
has_no_stats_stage(const AV1_COMP * const cpi)4093 static inline int has_no_stats_stage(const AV1_COMP *const cpi) {
4094 assert(
4095 IMPLIES(!cpi->ppi->lap_enabled, cpi->compressor_stage == ENCODE_STAGE));
4096 return (cpi->oxcf.pass == AOM_RC_ONE_PASS && !cpi->ppi->lap_enabled);
4097 }
4098
4099 /*!\cond */
4100
is_one_pass_rt_params(const AV1_COMP * cpi)4101 static inline int is_one_pass_rt_params(const AV1_COMP *cpi) {
4102 return has_no_stats_stage(cpi) && cpi->oxcf.mode == REALTIME &&
4103 cpi->oxcf.gf_cfg.lag_in_frames == 0;
4104 }
4105
4106 // Use default/internal reference structure for single-layer RTC.
use_rtc_reference_structure_one_layer(const AV1_COMP * cpi)4107 static inline int use_rtc_reference_structure_one_layer(const AV1_COMP *cpi) {
4108 return is_one_pass_rt_params(cpi) && cpi->ppi->number_spatial_layers == 1 &&
4109 cpi->ppi->number_temporal_layers == 1 &&
4110 !cpi->ppi->rtc_ref.set_ref_frame_config;
4111 }
4112
4113 // Check if postencode drop is allowed.
allow_postencode_drop_rtc(const AV1_COMP * cpi)4114 static inline int allow_postencode_drop_rtc(const AV1_COMP *cpi) {
4115 const AV1_COMMON *const cm = &cpi->common;
4116 return is_one_pass_rt_params(cpi) && cpi->oxcf.rc_cfg.mode == AOM_CBR &&
4117 cpi->oxcf.rc_cfg.drop_frames_water_mark > 0 &&
4118 !cpi->rc.rtc_external_ratectrl && !frame_is_intra_only(cm) &&
4119 cpi->svc.spatial_layer_id == 0;
4120 }
4121
4122 // Function return size of frame stats buffer
get_stats_buf_size(int num_lap_buffer,int num_lag_buffer)4123 static inline int get_stats_buf_size(int num_lap_buffer, int num_lag_buffer) {
4124 /* if lookahead is enabled return num_lap_buffers else num_lag_buffers */
4125 return (num_lap_buffer > 0 ? num_lap_buffer + 1 : num_lag_buffer);
4126 }
4127
4128 // TODO(zoeliu): To set up cpi->oxcf.gf_cfg.enable_auto_brf
4129
set_ref_ptrs(const AV1_COMMON * cm,MACROBLOCKD * xd,MV_REFERENCE_FRAME ref0,MV_REFERENCE_FRAME ref1)4130 static inline void set_ref_ptrs(const AV1_COMMON *cm, MACROBLOCKD *xd,
4131 MV_REFERENCE_FRAME ref0,
4132 MV_REFERENCE_FRAME ref1) {
4133 xd->block_ref_scale_factors[0] =
4134 get_ref_scale_factors_const(cm, ref0 >= LAST_FRAME ? ref0 : 1);
4135 xd->block_ref_scale_factors[1] =
4136 get_ref_scale_factors_const(cm, ref1 >= LAST_FRAME ? ref1 : 1);
4137 }
4138
get_chessboard_index(int frame_index)4139 static inline int get_chessboard_index(int frame_index) {
4140 return frame_index & 0x1;
4141 }
4142
cond_cost_list_const(const struct AV1_COMP * cpi,const int * cost_list)4143 static inline const int *cond_cost_list_const(const struct AV1_COMP *cpi,
4144 const int *cost_list) {
4145 const int use_cost_list = cpi->sf.mv_sf.subpel_search_method != SUBPEL_TREE &&
4146 cpi->sf.mv_sf.use_fullpel_costlist;
4147 return use_cost_list ? cost_list : NULL;
4148 }
4149
cond_cost_list(const struct AV1_COMP * cpi,int * cost_list)4150 static inline int *cond_cost_list(const struct AV1_COMP *cpi, int *cost_list) {
4151 const int use_cost_list = cpi->sf.mv_sf.subpel_search_method != SUBPEL_TREE &&
4152 cpi->sf.mv_sf.use_fullpel_costlist;
4153 return use_cost_list ? cost_list : NULL;
4154 }
4155
4156 // Compression ratio of current frame.
4157 double av1_get_compression_ratio(const AV1_COMMON *const cm,
4158 size_t encoded_frame_size);
4159
4160 void av1_new_framerate(AV1_COMP *cpi, double framerate);
4161
4162 void av1_setup_frame_size(AV1_COMP *cpi);
4163
4164 #define LAYER_IDS_TO_IDX(sl, tl, num_tl) ((sl) * (num_tl) + (tl))
4165
4166 // Returns 1 if a frame is scaled and 0 otherwise.
av1_resize_scaled(const AV1_COMMON * cm)4167 static inline int av1_resize_scaled(const AV1_COMMON *cm) {
4168 return cm->superres_upscaled_width != cm->render_width ||
4169 cm->superres_upscaled_height != cm->render_height;
4170 }
4171
av1_frame_scaled(const AV1_COMMON * cm)4172 static inline int av1_frame_scaled(const AV1_COMMON *cm) {
4173 return av1_superres_scaled(cm) || av1_resize_scaled(cm);
4174 }
4175
4176 // Don't allow a show_existing_frame to coincide with an error resilient
4177 // frame. An exception can be made for a forward keyframe since it has no
4178 // previous dependencies.
encode_show_existing_frame(const AV1_COMMON * cm)4179 static inline int encode_show_existing_frame(const AV1_COMMON *cm) {
4180 return cm->show_existing_frame && (!cm->features.error_resilient_mode ||
4181 cm->current_frame.frame_type == KEY_FRAME);
4182 }
4183
4184 // Get index into the 'cpi->mbmi_ext_info.frame_base' array for the given
4185 // '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)4186 static inline int get_mi_ext_idx(const int mi_row, const int mi_col,
4187 const BLOCK_SIZE mi_alloc_bsize,
4188 const int mbmi_ext_stride) {
4189 const int mi_ext_size_1d = mi_size_wide[mi_alloc_bsize];
4190 const int mi_ext_row = mi_row / mi_ext_size_1d;
4191 const int mi_ext_col = mi_col / mi_ext_size_1d;
4192 return mi_ext_row * mbmi_ext_stride + mi_ext_col;
4193 }
4194
4195 // Lighter version of set_offsets that only sets the mode info
4196 // 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)4197 static inline void set_mode_info_offsets(
4198 const CommonModeInfoParams *const mi_params,
4199 const MBMIExtFrameBufferInfo *const mbmi_ext_info, MACROBLOCK *const x,
4200 MACROBLOCKD *const xd, int mi_row, int mi_col) {
4201 set_mi_offsets(mi_params, xd, mi_row, mi_col);
4202 const int ext_idx = get_mi_ext_idx(mi_row, mi_col, mi_params->mi_alloc_bsize,
4203 mbmi_ext_info->stride);
4204 x->mbmi_ext_frame = mbmi_ext_info->frame_base + ext_idx;
4205 }
4206
4207 // Check to see if the given partition size is allowed for a specified number
4208 // of mi block rows and columns remaining in the image.
4209 // 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)4210 static inline BLOCK_SIZE find_partition_size(BLOCK_SIZE bsize, int rows_left,
4211 int cols_left, int *bh, int *bw) {
4212 int int_size = (int)bsize;
4213 if (rows_left <= 0 || cols_left <= 0) {
4214 return AOMMIN(bsize, BLOCK_8X8);
4215 } else {
4216 for (; int_size > 0; int_size -= 3) {
4217 *bh = mi_size_high[int_size];
4218 *bw = mi_size_wide[int_size];
4219 if ((*bh <= rows_left) && (*bw <= cols_left)) {
4220 break;
4221 }
4222 }
4223 }
4224 return (BLOCK_SIZE)int_size;
4225 }
4226
4227 static const uint8_t av1_ref_frame_flag_list[REF_FRAMES] = { 0,
4228 AOM_LAST_FLAG,
4229 AOM_LAST2_FLAG,
4230 AOM_LAST3_FLAG,
4231 AOM_GOLD_FLAG,
4232 AOM_BWD_FLAG,
4233 AOM_ALT2_FLAG,
4234 AOM_ALT_FLAG };
4235
4236 // When more than 'max_allowed_refs' are available, we reduce the number of
4237 // reference frames one at a time based on this order.
4238 static const MV_REFERENCE_FRAME disable_order[] = {
4239 LAST3_FRAME,
4240 LAST2_FRAME,
4241 ALTREF2_FRAME,
4242 BWDREF_FRAME,
4243 };
4244
4245 static const MV_REFERENCE_FRAME
4246 ref_frame_priority_order[INTER_REFS_PER_FRAME] = {
4247 LAST_FRAME, ALTREF_FRAME, BWDREF_FRAME, GOLDEN_FRAME,
4248 ALTREF2_FRAME, LAST2_FRAME, LAST3_FRAME,
4249 };
4250
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)4251 static inline int get_ref_frame_flags(const SPEED_FEATURES *const sf,
4252 const int use_one_pass_rt_params,
4253 const YV12_BUFFER_CONFIG **ref_frames,
4254 const int ext_ref_frame_flags) {
4255 // cpi->ext_flags.ref_frame_flags allows certain reference types to be
4256 // disabled by the external interface. These are set by
4257 // av1_apply_encoding_flags(). Start with what the external interface allows,
4258 // then suppress any reference types which we have found to be duplicates.
4259 int flags = ext_ref_frame_flags;
4260
4261 for (int i = 1; i < INTER_REFS_PER_FRAME; ++i) {
4262 const YV12_BUFFER_CONFIG *const this_ref = ref_frames[i];
4263 // If this_ref has appeared before, mark the corresponding ref frame as
4264 // invalid. For one_pass_rt mode, only disable GOLDEN_FRAME if it's the
4265 // same as LAST_FRAME or ALTREF_FRAME (if ALTREF is being used in nonrd).
4266 int index =
4267 (use_one_pass_rt_params && ref_frame_priority_order[i] == GOLDEN_FRAME)
4268 ? (1 + sf->rt_sf.use_nonrd_altref_frame)
4269 : i;
4270 for (int j = 0; j < index; ++j) {
4271 // If this_ref has appeared before (same as the reference corresponding
4272 // to lower index j), remove it as a reference only if that reference
4273 // (for index j) is actually used as a reference.
4274 if (this_ref == ref_frames[j] &&
4275 (flags & (1 << (ref_frame_priority_order[j] - 1)))) {
4276 flags &= ~(1 << (ref_frame_priority_order[i] - 1));
4277 break;
4278 }
4279 }
4280 }
4281 return flags;
4282 }
4283
4284 // Returns a Sequence Header OBU stored in an aom_fixed_buf_t, or NULL upon
4285 // failure. When a non-NULL aom_fixed_buf_t pointer is returned by this
4286 // function, the memory must be freed by the caller. Both the buf member of the
4287 // aom_fixed_buf_t, and the aom_fixed_buf_t pointer itself must be freed. Memory
4288 // returned must be freed via call to free().
4289 //
4290 // Note: The OBU returned is in Low Overhead Bitstream Format. Specifically,
4291 // the obu_has_size_field bit is set, and the buffer contains the obu_size
4292 // field.
4293 aom_fixed_buf_t *av1_get_global_headers(AV1_PRIMARY *ppi);
4294
4295 #define MAX_GFUBOOST_FACTOR 10.0
4296 #define MIN_GFUBOOST_FACTOR 4.0
4297
is_frame_tpl_eligible(const GF_GROUP * const gf_group,uint8_t index)4298 static inline int is_frame_tpl_eligible(const GF_GROUP *const gf_group,
4299 uint8_t index) {
4300 const FRAME_UPDATE_TYPE update_type = gf_group->update_type[index];
4301 return update_type == ARF_UPDATE || update_type == GF_UPDATE ||
4302 update_type == KF_UPDATE;
4303 }
4304
is_frame_eligible_for_ref_pruning(const GF_GROUP * gf_group,int selective_ref_frame,int prune_ref_frames,int gf_index)4305 static inline int is_frame_eligible_for_ref_pruning(const GF_GROUP *gf_group,
4306 int selective_ref_frame,
4307 int prune_ref_frames,
4308 int gf_index) {
4309 return (selective_ref_frame > 0) && (prune_ref_frames > 0) &&
4310 !is_frame_tpl_eligible(gf_group, gf_index);
4311 }
4312
4313 // Get update type of the current frame.
get_frame_update_type(const GF_GROUP * gf_group,int gf_frame_index)4314 static inline FRAME_UPDATE_TYPE get_frame_update_type(const GF_GROUP *gf_group,
4315 int gf_frame_index) {
4316 return gf_group->update_type[gf_frame_index];
4317 }
4318
av1_pixels_to_mi(int pixels)4319 static inline int av1_pixels_to_mi(int pixels) {
4320 return ALIGN_POWER_OF_TWO(pixels, 3) >> MI_SIZE_LOG2;
4321 }
4322
is_psnr_calc_enabled(const AV1_COMP * cpi)4323 static inline int is_psnr_calc_enabled(const AV1_COMP *cpi) {
4324 const AV1_COMMON *const cm = &cpi->common;
4325
4326 return cpi->ppi->b_calculate_psnr && !is_stat_generation_stage(cpi) &&
4327 cm->show_frame && !cpi->is_dropped_frame;
4328 }
4329
is_frame_resize_pending(const AV1_COMP * const cpi)4330 static inline int is_frame_resize_pending(const AV1_COMP *const cpi) {
4331 const ResizePendingParams *const resize_pending_params =
4332 &cpi->resize_pending_params;
4333 return (resize_pending_params->width && resize_pending_params->height &&
4334 (cpi->common.width != resize_pending_params->width ||
4335 cpi->common.height != resize_pending_params->height));
4336 }
4337
4338 // Check if loop filter is used.
is_loopfilter_used(const AV1_COMMON * const cm)4339 static inline int is_loopfilter_used(const AV1_COMMON *const cm) {
4340 return !cm->features.coded_lossless && !cm->tiles.large_scale;
4341 }
4342
4343 // Check if CDEF is used.
is_cdef_used(const AV1_COMMON * const cm)4344 static inline int is_cdef_used(const AV1_COMMON *const cm) {
4345 return cm->seq_params->enable_cdef && !cm->features.coded_lossless &&
4346 !cm->tiles.large_scale;
4347 }
4348
4349 // Check if loop restoration filter is used.
is_restoration_used(const AV1_COMMON * const cm)4350 static inline int is_restoration_used(const AV1_COMMON *const cm) {
4351 return cm->seq_params->enable_restoration && !cm->features.all_lossless &&
4352 !cm->tiles.large_scale;
4353 }
4354
4355 // Checks if post-processing filters need to be applied.
4356 // NOTE: This function decides if the application of different post-processing
4357 // filters on the reconstructed frame can be skipped at the encoder side.
4358 // However the computation of different filter parameters that are signaled in
4359 // 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)4360 static inline unsigned int derive_skip_apply_postproc_filters(
4361 const AV1_COMP *cpi, int use_loopfilter, int use_cdef, int use_superres,
4362 int use_restoration) {
4363 // Though CDEF parameter selection should be dependent on
4364 // deblocked/loop-filtered pixels for cdef_pick_method <=
4365 // CDEF_FAST_SEARCH_LVL5, CDEF strength values are calculated based on the
4366 // pixel values that are not loop-filtered in svc real-time encoding mode.
4367 // Hence this case is handled separately using the condition below.
4368 if (cpi->ppi->rtc_ref.non_reference_frame)
4369 return (SKIP_APPLY_LOOPFILTER | SKIP_APPLY_CDEF);
4370
4371 if (!cpi->oxcf.algo_cfg.skip_postproc_filtering || cpi->ppi->b_calculate_psnr)
4372 return 0;
4373 assert(cpi->oxcf.mode == ALLINTRA);
4374
4375 // The post-processing filters are applied one after the other in the
4376 // following order: deblocking->cdef->superres->restoration. In case of
4377 // ALLINTRA encoding, the reconstructed frame is not used as a reference
4378 // frame. Hence, the application of these filters can be skipped when
4379 // 1. filter parameters of the subsequent stages are not dependent on the
4380 // filtered output of the current stage or
4381 // 2. subsequent filtering stages are disabled
4382 if (use_restoration) return SKIP_APPLY_RESTORATION;
4383 if (use_superres) return SKIP_APPLY_SUPERRES;
4384 if (use_cdef) {
4385 // CDEF parameter selection is not dependent on the deblocked frame if
4386 // cdef_pick_method is CDEF_PICK_FROM_Q. Hence the application of deblocking
4387 // filters and cdef filters can be skipped in this case.
4388 return (cpi->sf.lpf_sf.cdef_pick_method == CDEF_PICK_FROM_Q &&
4389 use_loopfilter)
4390 ? (SKIP_APPLY_LOOPFILTER | SKIP_APPLY_CDEF)
4391 : SKIP_APPLY_CDEF;
4392 }
4393 if (use_loopfilter) return SKIP_APPLY_LOOPFILTER;
4394
4395 // If we reach here, all post-processing stages are disabled, so none need to
4396 // be skipped.
4397 return 0;
4398 }
4399
set_postproc_filter_default_params(AV1_COMMON * cm)4400 static inline void set_postproc_filter_default_params(AV1_COMMON *cm) {
4401 struct loopfilter *const lf = &cm->lf;
4402 CdefInfo *const cdef_info = &cm->cdef_info;
4403 RestorationInfo *const rst_info = cm->rst_info;
4404
4405 lf->filter_level[0] = 0;
4406 lf->filter_level[1] = 0;
4407 cdef_info->cdef_bits = 0;
4408 cdef_info->cdef_strengths[0] = 0;
4409 cdef_info->nb_cdef_strengths = 1;
4410 cdef_info->cdef_uv_strengths[0] = 0;
4411 rst_info[0].frame_restoration_type = RESTORE_NONE;
4412 rst_info[1].frame_restoration_type = RESTORE_NONE;
4413 rst_info[2].frame_restoration_type = RESTORE_NONE;
4414 }
4415
is_inter_tx_size_search_level_one(const TX_SPEED_FEATURES * tx_sf)4416 static inline int is_inter_tx_size_search_level_one(
4417 const TX_SPEED_FEATURES *tx_sf) {
4418 return (tx_sf->inter_tx_size_search_init_depth_rect >= 1 &&
4419 tx_sf->inter_tx_size_search_init_depth_sqr >= 1);
4420 }
4421
get_lpf_opt_level(const SPEED_FEATURES * sf)4422 static inline int get_lpf_opt_level(const SPEED_FEATURES *sf) {
4423 int lpf_opt_level = 0;
4424 if (is_inter_tx_size_search_level_one(&sf->tx_sf))
4425 lpf_opt_level = (sf->lpf_sf.lpf_pick == LPF_PICK_FROM_Q) ? 2 : 1;
4426 return lpf_opt_level;
4427 }
4428
4429 // Enable switchable motion mode only if warp and OBMC tools are allowed
is_switchable_motion_mode_allowed(bool allow_warped_motion,bool enable_obmc)4430 static inline bool is_switchable_motion_mode_allowed(bool allow_warped_motion,
4431 bool enable_obmc) {
4432 return (allow_warped_motion || enable_obmc);
4433 }
4434
4435 #if CONFIG_AV1_TEMPORAL_DENOISING
denoise_svc(const struct AV1_COMP * const cpi)4436 static inline int denoise_svc(const struct AV1_COMP *const cpi) {
4437 return (!cpi->ppi->use_svc ||
4438 (cpi->ppi->use_svc &&
4439 cpi->svc.spatial_layer_id >= cpi->svc.first_layer_denoise));
4440 }
4441 #endif
4442
4443 #if CONFIG_COLLECT_PARTITION_STATS == 2
av1_print_fr_partition_timing_stats(const FramePartitionTimingStats * part_stats,const char * filename)4444 static inline void av1_print_fr_partition_timing_stats(
4445 const FramePartitionTimingStats *part_stats, const char *filename) {
4446 FILE *f = fopen(filename, "w");
4447 if (!f) {
4448 return;
4449 }
4450
4451 fprintf(f, "bsize,redo,");
4452 for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
4453 fprintf(f, "decision_%d,", part);
4454 }
4455 for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
4456 fprintf(f, "attempt_%d,", part);
4457 }
4458 for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
4459 fprintf(f, "time_%d,", part);
4460 }
4461 fprintf(f, "\n");
4462
4463 static const int bsizes[6] = { 128, 64, 32, 16, 8, 4 };
4464
4465 for (int bsize_idx = 0; bsize_idx < 6; bsize_idx++) {
4466 fprintf(f, "%d,%d,", bsizes[bsize_idx], part_stats->partition_redo);
4467 for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
4468 fprintf(f, "%d,", part_stats->partition_decisions[bsize_idx][part]);
4469 }
4470 for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
4471 fprintf(f, "%d,", part_stats->partition_attempts[bsize_idx][part]);
4472 }
4473 for (int part = 0; part < EXT_PARTITION_TYPES; part++) {
4474 fprintf(f, "%ld,", part_stats->partition_times[bsize_idx][part]);
4475 }
4476 fprintf(f, "\n");
4477 }
4478 fclose(f);
4479 }
4480 #endif // CONFIG_COLLECT_PARTITION_STATS == 2
4481
4482 #if CONFIG_COLLECT_PARTITION_STATS
av1_get_bsize_idx_for_part_stats(BLOCK_SIZE bsize)4483 static inline int av1_get_bsize_idx_for_part_stats(BLOCK_SIZE bsize) {
4484 assert(bsize == BLOCK_128X128 || bsize == BLOCK_64X64 ||
4485 bsize == BLOCK_32X32 || bsize == BLOCK_16X16 || bsize == BLOCK_8X8 ||
4486 bsize == BLOCK_4X4);
4487 switch (bsize) {
4488 case BLOCK_128X128: return 0;
4489 case BLOCK_64X64: return 1;
4490 case BLOCK_32X32: return 2;
4491 case BLOCK_16X16: return 3;
4492 case BLOCK_8X8: return 4;
4493 case BLOCK_4X4: return 5;
4494 default: assert(0 && "Invalid bsize for partition_stats."); return -1;
4495 }
4496 }
4497 #endif // CONFIG_COLLECT_PARTITION_STATS
4498
4499 #if CONFIG_COLLECT_COMPONENT_TIMING
start_timing(AV1_COMP * cpi,int component)4500 static inline void start_timing(AV1_COMP *cpi, int component) {
4501 aom_usec_timer_start(&cpi->component_timer[component]);
4502 }
end_timing(AV1_COMP * cpi,int component)4503 static inline void end_timing(AV1_COMP *cpi, int component) {
4504 aom_usec_timer_mark(&cpi->component_timer[component]);
4505 cpi->frame_component_time[component] +=
4506 aom_usec_timer_elapsed(&cpi->component_timer[component]);
4507 }
get_frame_type_enum(int type)4508 static inline char const *get_frame_type_enum(int type) {
4509 switch (type) {
4510 case 0: return "KEY_FRAME";
4511 case 1: return "INTER_FRAME";
4512 case 2: return "INTRA_ONLY_FRAME";
4513 case 3: return "S_FRAME";
4514 default: assert(0);
4515 }
4516 return "error";
4517 }
4518 #endif
4519
4520 /*!\endcond */
4521
4522 #ifdef __cplusplus
4523 } // extern "C"
4524 #endif
4525
4526 #endif // AOM_AV1_ENCODER_ENCODER_H_
4527