AOMedia AV1 Codec
ratectrl.h
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 #ifndef AOM_AV1_ENCODER_RATECTRL_H_
13 #define AOM_AV1_ENCODER_RATECTRL_H_
14 
15 #include "aom/aom_codec.h"
16 #include "aom/aom_integer.h"
17 
18 #include "aom_ports/mem.h"
19 
20 #include "av1/common/av1_common_int.h"
21 #include "av1/common/blockd.h"
22 
23 #ifdef __cplusplus
24 extern "C" {
25 #endif
26 
29 // Bits Per MB at different Q (Multiplied by 512)
30 #define BPER_MB_NORMBITS 9
31 
32 // Use this macro to turn on/off use of alt-refs in one-pass mode.
33 #define USE_ALTREF_FOR_ONE_PASS 1
34 
35 // Threshold used to define if a KF group is static (e.g. a slide show).
36 // Essentially, this means that no frame in the group has more than 1% of MBs
37 // that are not marked as coded with 0,0 motion in the first pass.
38 #define STATIC_KF_GROUP_THRESH 99
39 #define STATIC_KF_GROUP_FLOAT_THRESH 0.99
40 
41 // The maximum duration of a GF group that is static (e.g. a slide show).
42 #define MAX_STATIC_GF_GROUP_LENGTH 250
43 
44 #define MIN_GF_INTERVAL 4
45 #define MAX_GF_INTERVAL 32
46 #define FIXED_GF_INTERVAL 16
47 #define MAX_GF_LENGTH_LAP 16
48 
49 #define FIXED_GF_INTERVAL_RT 80
50 #define MAX_GF_INTERVAL_RT 160
51 
52 #define MAX_NUM_GF_INTERVALS 15
53 
54 #define MAX_ARF_LAYERS 6
55 // #define STRICT_RC
56 
57 #define DEFAULT_KF_BOOST_RT 2300
58 #define DEFAULT_GF_BOOST_RT 2000
59 
60 // A passive rate control strategy for screen content type in real-time mode.
61 // When it is turned on, the compression performance is improved by
62 // 7.8% (overall_psnr), 5.0% (VMAF) on average. Some clips see gains
63 // over 20% on metric.
64 // The downside is that it does not guarantee frame size.
65 // Since RT mode has a tight restriction on buffer overflow control, we
66 // turn it off by default.
67 #define RT_PASSIVE_STRATEGY 0
68 #define MAX_Q_HISTORY 1000
69 
70 typedef struct {
71  int resize_width;
72  int resize_height;
73  uint8_t superres_denom;
74 } size_params_type;
75 
76 enum {
77  INTER_NORMAL,
78  GF_ARF_LOW,
79  GF_ARF_STD,
80  KF_STD,
81  RATE_FACTOR_LEVELS
82 } UENUM1BYTE(RATE_FACTOR_LEVEL);
83 
84 enum {
85  KF_UPDATE,
86  LF_UPDATE,
87  GF_UPDATE,
88  ARF_UPDATE,
89  OVERLAY_UPDATE,
90  INTNL_OVERLAY_UPDATE, // Internal Overlay Frame
91  INTNL_ARF_UPDATE, // Internal Altref Frame
92  FRAME_UPDATE_TYPES
93 } UENUM1BYTE(FRAME_UPDATE_TYPE);
94 
95 enum {
96  REFBUF_RESET, // Clear reference frame buffer
97  REFBUF_UPDATE, // Refresh reference frame buffer
98  REFBUF_STATES
99 } UENUM1BYTE(REFBUF_STATE);
100 
101 typedef enum {
102  NO_RESIZE = 0,
103  DOWN_THREEFOUR = 1, // From orig to 3/4.
104  DOWN_ONEHALF = 2, // From orig or 3/4 to 1/2.
105  UP_THREEFOUR = -1, // From 1/2 to 3/4.
106  UP_ORIG = -2, // From 1/2 or 3/4 to orig.
107 } RESIZE_ACTION;
108 
109 typedef enum { ORIG = 0, THREE_QUARTER = 1, ONE_HALF = 2 } RESIZE_STATE;
110 
111 #define MAX_FIRSTPASS_ANALYSIS_FRAMES 150
112 typedef enum region_types {
113  STABLE_REGION = 0,
114  HIGH_VAR_REGION = 1,
115  SCENECUT_REGION = 2,
116  BLENDING_REGION = 3,
117 } REGION_TYPES;
118 
119 typedef struct regions {
120  int start;
121  int last;
122  double avg_noise_var;
123  double avg_cor_coeff;
124  double avg_sr_fr_ratio;
125  double avg_intra_err;
126  double avg_coded_err;
127  REGION_TYPES type;
128 } REGIONS;
129 
134 typedef struct {
135  // Rate targetting variables
136 
145  int this_frame_target; // Actual frame target after rc adjustment.
146 
151 
156 
161 
166 
171 
176 
178  int min_gf_interval;
179  int max_gf_interval;
180  int static_scene_max_gf_interval;
187  int frames_since_key;
188  int frames_to_fwd_kf;
189  int is_src_frame_alt_ref;
190  int sframe_due;
191 
192  int high_source_sad;
193  int high_motion_content_screen_rtc;
194  uint64_t avg_source_sad;
195  uint64_t prev_avg_source_sad;
196  uint64_t frame_source_sad;
197  uint64_t frame_spatial_variance;
198  int static_since_last_scene_change;
199  int last_encoded_size_keyframe;
200  int last_target_size_keyframe;
201  int frames_since_scene_change;
202  int perc_spatial_flat_blocks;
203 
204  int avg_frame_bandwidth; // Average frame size target for clip
205  int min_frame_bandwidth; // Minimum allocation used for any frame
206  int max_frame_bandwidth; // Maximum burst rate allowed for a frame.
207  int prev_avg_frame_bandwidth;
208 
209  int ni_av_qi;
210  int ni_tot_qi;
211 
212  int decimation_factor;
213  int decimation_count;
214  int prev_frame_is_dropped;
215  int drop_count_consec;
216  int max_consec_drop;
217  int force_max_q;
218  int postencode_drop;
219 
224  unsigned int frame_number_encoded;
225 
235 
238  // rate control history for last frame(1) and the frame before(2).
239  // -1: overshoot
240  // 1: undershoot
241  // 0: not initialized.
242  int rc_1_frame;
243  int rc_2_frame;
244  int q_1_frame;
245  int q_2_frame;
246 
252 
254  // Track amount of low motion in scene
255  int avg_frame_low_motion;
256  int cnt_zeromv;
257 
258  // signals if number of blocks with motion is high
259  int percent_blocks_with_motion;
260 
261  // signals percentage of 16x16 blocks that are inactive, via active_maps
262  int percent_blocks_inactive;
263 
264  // Maximum value of source sad across all blocks of frame.
265  uint64_t max_block_source_sad;
266 
267  // For dynamic resize, 1 pass cbr.
268  RESIZE_STATE resize_state;
269  int resize_avg_qp;
270  int resize_buffer_underflow;
271  int resize_count;
272 
273  // Flag to disable content related qp adjustment.
274  int rtc_external_ratectrl;
275 
276  // Stores fast_extra_bits of the current frame.
277  int frame_level_fast_extra_bits;
278 
279  double frame_level_rate_correction_factors[RATE_FACTOR_LEVELS];
280 
281  int frame_num_last_gf_refresh;
282 
283  int prev_coded_width;
284  int prev_coded_height;
285 
286  // The ratio used for inter frames in bit estimation.
287  // TODO(yunqing): if golden frame is treated differently (e.g. gf_cbr_boost_
288  // pct > THR), consider to add bit_est_ratio_g for golden frames.
289  int bit_est_ratio;
290 
291  // Whether to use a fixed qp for the frame, bypassing internal rate control.
292  // This flag will reset to 0 after every frame.
293  int use_external_qp_one_pass;
295 } RATE_CONTROL;
296 
300 typedef struct {
301  // Sub-gop level Rate targetting variables
302 
306  int64_t gf_group_bits;
307 
311  int kf_boost;
312 
317 
321  int gf_intervals[MAX_NUM_GF_INTERVALS];
322 
327 
329  int num_regions;
330 
331  REGIONS regions[MAX_FIRSTPASS_ANALYSIS_FRAMES];
332  int regions_offset; // offset of regions from the last keyframe
333  int frames_till_regions_update;
334 
335  int baseline_gf_interval;
336 
337  int constrained_gf_group;
338 
339  int this_key_frame_forced;
340 
341  int next_key_frame_forced;
348 
353 
358 
362  int arf_q;
363 
365  float_t arf_boost_factor;
366 
367  int base_layer_qp;
368 
369  // Total number of stats used only for kf_boost calculation.
370  int num_stats_used_for_kf_boost;
371 
372  // Total number of stats used only for gfu_boost calculation.
373  int num_stats_used_for_gfu_boost;
374 
375  // Total number of stats required by gfu_boost calculation.
376  int num_stats_required_for_gfu_boost;
377 
378  int enable_scenecut_detection;
379 
380  int use_arf_in_this_kf_group;
381 
382  int ni_frames;
383 
384  double tot_q;
391 
395  int avg_frame_qindex[FRAME_TYPES];
396 
397 #if CONFIG_FPMT_TEST
402  int temp_active_best_quality[MAX_ARF_LAYERS + 1];
403 
408  int temp_last_boosted_qindex;
409 
414  double temp_avg_q;
415 
420  int temp_last_q[FRAME_TYPES];
421 
426  int temp_projected_frame_size;
427 
432  int64_t temp_total_actual_bits;
433 
438  int64_t temp_buffer_level;
439 
444  int64_t temp_vbr_bits_off_target;
445 
450  int64_t temp_vbr_bits_off_target_fast;
451 
456  double temp_rate_correction_factors[RATE_FACTOR_LEVELS];
457 
462  int temp_rate_error_estimate;
463 
468  int temp_rolling_arf_group_target_bits;
469 
474  int temp_rolling_arf_group_actual_bits;
475 
480  int64_t temp_bits_left;
481 
486  int temp_extend_minq;
487 
492  int temp_extend_maxq;
493 
494 #endif
498  int active_best_quality[MAX_ARF_LAYERS + 1];
499 
504 
508  double avg_q;
509 
513  int last_q[FRAME_TYPES];
514 
519  double rate_correction_factors[RATE_FACTOR_LEVELS];
520 
525 
530 
534  int64_t buffer_level;
535 
540 
545 
550 
556 
561 
567 
572  int q_history[MAX_Q_HISTORY];
574 
577 struct AV1_COMP;
578 struct AV1EncoderConfig;
579 struct GF_GROUP;
580 
581 void av1_primary_rc_init(const struct AV1EncoderConfig *oxcf,
582  PRIMARY_RATE_CONTROL *p_rc);
583 
584 void av1_rc_init(const struct AV1EncoderConfig *oxcf, RATE_CONTROL *rc);
585 
586 int av1_estimate_bits_at_q(const struct AV1_COMP *cpi, int q,
587  double correction_factor);
588 
589 double av1_convert_qindex_to_q(int qindex, aom_bit_depth_t bit_depth);
590 
591 // Converts a Q value to a qindex.
592 int av1_convert_q_to_qindex(double q, aom_bit_depth_t bit_depth);
593 
594 void av1_rc_init_minq_luts(void);
595 
596 int av1_rc_get_default_min_gf_interval(int width, int height, double framerate);
597 
598 // Generally at the high level, the following flow is expected
599 // to be enforced for rate control:
600 // First call per frame, one of:
601 // av1_get_one_pass_rt_params()
602 // av1_get_second_pass_params()
603 // depending on the usage to set the rate control encode parameters desired.
604 //
605 // Then, call encode_frame_to_data_rate() to perform the
606 // actual encode. This function will in turn call encode_frame()
607 // one or more times, followed by:
608 // av1_rc_postencode_update_drop_frame()
609 //
610 // The majority of rate control parameters are only expected
611 // to be set in the av1_get_..._params() functions and
612 // updated during the av1_rc_postencode_update...() functions.
613 // The only exceptions are av1_rc_drop_frame() and
614 // av1_rc_update_rate_correction_factors() functions.
615 
616 // Functions to set parameters for encoding before the actual
617 // encode_frame_to_data_rate() function.
618 struct EncodeFrameInput;
619 
620 // Post encode update of the rate control parameters based
621 // on bytes used
622 void av1_rc_postencode_update(struct AV1_COMP *cpi, uint64_t bytes_used);
623 // Post encode update of the rate control parameters for dropped frames
624 void av1_rc_postencode_update_drop_frame(struct AV1_COMP *cpi);
625 
641  int is_encode_stage, int width,
642  int height);
645 // Decide if we should drop this frame: For 1-pass CBR.
646 // Changes only the decimation count in the rate control structure
647 int av1_rc_drop_frame(struct AV1_COMP *cpi);
648 
649 // Computes frame size bounds.
650 void av1_rc_compute_frame_size_bounds(const struct AV1_COMP *cpi,
651  int this_frame_target,
652  int *frame_under_shoot_limit,
653  int *frame_over_shoot_limit);
654 
669 int av1_rc_pick_q_and_bounds(struct AV1_COMP *cpi, int width, int height,
670  int gf_index, int *bottom_index, int *top_index);
671 
684 int av1_rc_regulate_q(const struct AV1_COMP *cpi, int target_bits_per_frame,
685  int active_best_quality, int active_worst_quality,
686  int width, int height);
687 
689 // Estimates bits per mb for a given qindex and correction factor.
690 int av1_rc_bits_per_mb(const struct AV1_COMP *cpi, FRAME_TYPE frame_type,
691  int qindex, double correction_factor,
692  int accurate_estimate);
693 
694 // Find q_index corresponding to desired_q, within [best_qindex, worst_qindex].
695 // To be precise, 'q_index' is the smallest integer, for which the corresponding
696 // q >= desired_q.
697 // If no such q index is found, returns 'worst_qindex'.
698 int av1_find_qindex(double desired_q, aom_bit_depth_t bit_depth,
699  int best_qindex, int worst_qindex);
700 
701 // Computes a q delta (in "q index" terms) to get from a starting q value
702 // to a target q value
703 int av1_compute_qdelta(const RATE_CONTROL *rc, double qstart, double qtarget,
704  aom_bit_depth_t bit_depth);
705 
706 // Computes a q delta (in "q index" terms) to get from a starting q value
707 // to a value that should equate to the given rate ratio.
708 int av1_compute_qdelta_by_rate(const struct AV1_COMP *cpi,
709  FRAME_TYPE frame_type, int qindex,
710  double rate_target_ratio);
711 
712 void av1_rc_update_framerate(struct AV1_COMP *cpi, int width, int height);
713 
714 void av1_set_target_rate(struct AV1_COMP *cpi, int width, int height);
715 
716 int av1_resize_one_pass_cbr(struct AV1_COMP *cpi);
717 
718 void av1_rc_set_frame_target(struct AV1_COMP *cpi, int target, int width,
719  int height);
720 
721 void av1_adjust_gf_refresh_qp_one_pass_rt(struct AV1_COMP *cpi);
722 
724  int gf_update);
725 
739  const struct AV1_COMP *const cpi, FRAME_UPDATE_TYPE frame_update_type);
740 
752 
765  const struct AV1_COMP *cpi, FRAME_UPDATE_TYPE frame_update_type);
766 
778 
798  FRAME_TYPE *const frame_type,
799  const struct EncodeFrameInput *frame_input,
800  unsigned int frame_flags);
801 
816 int av1_encodedframe_overshoot_cbr(struct AV1_COMP *cpi, int *q);
817 
829 int av1_postencode_drop_cbr(struct AV1_COMP *cpi, size_t *size);
830 
831 #ifdef __cplusplus
832 } // extern "C"
833 #endif
834 
835 #endif // AOM_AV1_ENCODER_RATECTRL_H_
Describes the codec algorithm interface to applications.
enum aom_bit_depth aom_bit_depth_t
Bit depth for codecThis enumeration determines the bit depth of the codec.
int av1_calc_pframe_target_size_one_pass_vbr(const struct AV1_COMP *const cpi, FRAME_UPDATE_TYPE frame_update_type)
Calculates how many bits to use for a P frame in one pass vbr.
void av1_rc_update_rate_correction_factors(struct AV1_COMP *cpi, int is_encode_stage, int width, int height)
Updates the rate correction factor linking Q to output bits.
Definition: ratectrl.c:894
int av1_encodedframe_overshoot_cbr(struct AV1_COMP *cpi, int *q)
Increase q on expected encoder overshoot, for CBR mode.
Definition: ratectrl.c:3865
void av1_get_one_pass_rt_params(struct AV1_COMP *cpi, FRAME_TYPE *const frame_type, const struct EncodeFrameInput *frame_input, unsigned int frame_flags)
Setup the rate control parameters for 1 pass real-time mode.
int av1_calc_pframe_target_size_one_pass_cbr(const struct AV1_COMP *cpi, FRAME_UPDATE_TYPE frame_update_type)
Calculates how many bits to use for a P frame in one pass cbr.
int av1_postencode_drop_cbr(struct AV1_COMP *cpi, size_t *size)
Check if frame should be dropped, for RTC mode.
Definition: ratectrl.c:3960
int av1_rc_pick_q_and_bounds(struct AV1_COMP *cpi, int width, int height, int gf_index, int *bottom_index, int *top_index)
Picks q and q bounds given the rate control parameters in cpi->rc.
Definition: ratectrl.c:2238
int av1_calc_iframe_target_size_one_pass_cbr(const struct AV1_COMP *cpi)
Calculates how many bits to use for an i frame in one pass cbr.
int av1_calc_iframe_target_size_one_pass_vbr(const struct AV1_COMP *const cpi)
Calculates how many bits to use for an i frame in one pass vbr.
int av1_rc_regulate_q(const struct AV1_COMP *cpi, int target_bits_per_frame, int active_best_quality, int active_worst_quality, int width, int height)
Estimates q to achieve a target bits per frame.
void av1_set_rtc_reference_structure_one_layer(AV1_COMP *cpi, int gf_update)
Setup the reference prediction structure for 1 pass real-time.
Definition: ratectrl.c:2965
Main encoder configuration data structure.
Definition: encoder.h:933
Top level encoder structure.
Definition: encoder.h:2878
Input frames and last input frame.
Definition: encoder.h:3682
Data related to the current GF/ARF group and the individual frames within the group.
Definition: firstpass.h:339
Primary Rate Control parameters and status.
Definition: ratectrl.h:300
int64_t bits_off_target
Definition: ratectrl.h:555
int rate_error_estimate
Definition: ratectrl.h:539
double avg_q
Definition: ratectrl.h:508
int64_t maximum_buffer_size
Definition: ratectrl.h:357
int kf_boost
Definition: ratectrl.h:311
int64_t starting_buffer_level
Definition: ratectrl.h:347
int64_t total_actual_bits
Definition: ratectrl.h:524
int64_t vbr_bits_off_target_fast
Definition: ratectrl.h:549
int rolling_target_bits
Definition: ratectrl.h:560
int64_t buffer_level
Definition: ratectrl.h:534
int gfu_boost
Definition: ratectrl.h:316
int64_t optimal_buffer_level
Definition: ratectrl.h:352
int arf_q
Definition: ratectrl.h:362
int rolling_actual_bits
Definition: ratectrl.h:566
int last_boosted_qindex
Definition: ratectrl.h:503
int cur_gf_index
Definition: ratectrl.h:326
int64_t total_target_bits
Definition: ratectrl.h:529
int last_kf_qindex
Definition: ratectrl.h:390
int64_t gf_group_bits
Definition: ratectrl.h:306
int64_t vbr_bits_off_target
Definition: ratectrl.h:544
Rate Control parameters and status.
Definition: ratectrl.h:134
int best_quality
Definition: ratectrl.h:234
int intervals_till_gf_calculate_due
Definition: ratectrl.h:175
int this_frame_target
Definition: ratectrl.h:145
int frames_to_key
Definition: ratectrl.h:185
int base_frame_target
Definition: ratectrl.h:141
int projected_frame_size
Definition: ratectrl.h:150
int worst_quality
Definition: ratectrl.h:230
int sb64_target_rate
Definition: ratectrl.h:160
int coefficient_size
Definition: ratectrl.h:155
int active_worst_quality
Definition: ratectrl.h:251
int frames_since_golden
Definition: ratectrl.h:165
int frames_till_gf_update_due
Definition: ratectrl.h:170