1 /*
2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #ifndef VP9_ENCODER_VP9_ENCODER_H_
12 #define VP9_ENCODER_VP9_ENCODER_H_
13
14 #include <stdio.h>
15
16 #include "./vpx_config.h"
17 #include "vpx/internal/vpx_codec_internal.h"
18 #include "vpx/vp8cx.h"
19 #if CONFIG_INTERNAL_STATS
20 #include "vpx_dsp/ssim.h"
21 #endif
22 #include "vpx_dsp/variance.h"
23 #include "vpx_ports/system_state.h"
24 #include "vpx_util/vpx_thread.h"
25
26 #include "vp9/common/vp9_alloccommon.h"
27 #include "vp9/common/vp9_ppflags.h"
28 #include "vp9/common/vp9_entropymode.h"
29 #include "vp9/common/vp9_thread_common.h"
30 #include "vp9/common/vp9_onyxc_int.h"
31
32 #include "vp9/encoder/vp9_alt_ref_aq.h"
33 #include "vp9/encoder/vp9_aq_cyclicrefresh.h"
34 #include "vp9/encoder/vp9_context_tree.h"
35 #include "vp9/encoder/vp9_encodemb.h"
36 #include "vp9/encoder/vp9_ethread.h"
37 #include "vp9/encoder/vp9_firstpass.h"
38 #include "vp9/encoder/vp9_job_queue.h"
39 #include "vp9/encoder/vp9_lookahead.h"
40 #include "vp9/encoder/vp9_mbgraph.h"
41 #include "vp9/encoder/vp9_mcomp.h"
42 #include "vp9/encoder/vp9_noise_estimate.h"
43 #include "vp9/encoder/vp9_quantize.h"
44 #include "vp9/encoder/vp9_ratectrl.h"
45 #include "vp9/encoder/vp9_rd.h"
46 #include "vp9/encoder/vp9_speed_features.h"
47 #include "vp9/encoder/vp9_svc_layercontext.h"
48 #include "vp9/encoder/vp9_tokenize.h"
49
50 #if CONFIG_VP9_TEMPORAL_DENOISING
51 #include "vp9/encoder/vp9_denoiser.h"
52 #endif
53
54 #ifdef __cplusplus
55 extern "C" {
56 #endif
57
58 // vp9 uses 10,000,000 ticks/second as time stamp
59 #define TICKS_PER_SEC 10000000
60
61 typedef struct {
62 int nmvjointcost[MV_JOINTS];
63 int nmvcosts[2][MV_VALS];
64 int nmvcosts_hp[2][MV_VALS];
65
66 vpx_prob segment_pred_probs[PREDICTION_PROBS];
67
68 unsigned char *last_frame_seg_map_copy;
69
70 // 0 = Intra, Last, GF, ARF
71 signed char last_ref_lf_deltas[MAX_REF_LF_DELTAS];
72 // 0 = ZERO_MV, MV
73 signed char last_mode_lf_deltas[MAX_MODE_LF_DELTAS];
74
75 FRAME_CONTEXT fc;
76 } CODING_CONTEXT;
77
78 typedef enum {
79 // encode_breakout is disabled.
80 ENCODE_BREAKOUT_DISABLED = 0,
81 // encode_breakout is enabled.
82 ENCODE_BREAKOUT_ENABLED = 1,
83 // encode_breakout is enabled with small max_thresh limit.
84 ENCODE_BREAKOUT_LIMITED = 2
85 } ENCODE_BREAKOUT_TYPE;
86
87 typedef enum {
88 NORMAL = 0,
89 FOURFIVE = 1,
90 THREEFIVE = 2,
91 ONETWO = 3
92 } VPX_SCALING;
93
94 typedef enum {
95 // Good Quality Fast Encoding. The encoder balances quality with the amount of
96 // time it takes to encode the output. Speed setting controls how fast.
97 GOOD,
98
99 // The encoder places priority on the quality of the output over encoding
100 // speed. The output is compressed at the highest possible quality. This
101 // option takes the longest amount of time to encode. Speed setting ignored.
102 BEST,
103
104 // Realtime/Live Encoding. This mode is optimized for realtime encoding (for
105 // example, capturing a television signal or feed from a live camera). Speed
106 // setting controls how fast.
107 REALTIME
108 } MODE;
109
110 typedef enum {
111 FRAMEFLAGS_KEY = 1 << 0,
112 FRAMEFLAGS_GOLDEN = 1 << 1,
113 FRAMEFLAGS_ALTREF = 1 << 2,
114 } FRAMETYPE_FLAGS;
115
116 typedef enum {
117 NO_AQ = 0,
118 VARIANCE_AQ = 1,
119 COMPLEXITY_AQ = 2,
120 CYCLIC_REFRESH_AQ = 3,
121 EQUATOR360_AQ = 4,
122 // AQ based on lookahead temporal
123 // variance (only valid for altref frames)
124 LOOKAHEAD_AQ = 5,
125 AQ_MODE_COUNT // This should always be the last member of the enum
126 } AQ_MODE;
127
128 typedef enum {
129 RESIZE_NONE = 0, // No frame resizing allowed (except for SVC).
130 RESIZE_FIXED = 1, // All frames are coded at the specified dimension.
131 RESIZE_DYNAMIC = 2 // Coded size of each frame is determined by the codec.
132 } RESIZE_TYPE;
133
134 typedef enum {
135 kInvalid = 0,
136 kLowSadLowSumdiff = 1,
137 kLowSadHighSumdiff = 2,
138 kHighSadLowSumdiff = 3,
139 kHighSadHighSumdiff = 4,
140 kLowVarHighSumdiff = 5,
141 kVeryHighSad = 6,
142 } CONTENT_STATE_SB;
143
144 typedef struct VP9EncoderConfig {
145 BITSTREAM_PROFILE profile;
146 vpx_bit_depth_t bit_depth; // Codec bit-depth.
147 int width; // width of data passed to the compressor
148 int height; // height of data passed to the compressor
149 unsigned int input_bit_depth; // Input bit depth.
150 double init_framerate; // set to passed in framerate
151 int64_t target_bandwidth; // bandwidth to be used in bits per second
152
153 int noise_sensitivity; // pre processing blur: recommendation 0
154 int sharpness; // sharpening output: recommendation 0:
155 int speed;
156 // maximum allowed bitrate for any intra frame in % of bitrate target.
157 unsigned int rc_max_intra_bitrate_pct;
158 // maximum allowed bitrate for any inter frame in % of bitrate target.
159 unsigned int rc_max_inter_bitrate_pct;
160 // percent of rate boost for golden frame in CBR mode.
161 unsigned int gf_cbr_boost_pct;
162
163 MODE mode;
164 int pass;
165
166 // Key Framing Operations
167 int auto_key; // autodetect cut scenes and set the keyframes
168 int key_freq; // maximum distance to key frame.
169
170 int lag_in_frames; // how many frames lag before we start encoding
171
172 // ----------------------------------------------------------------
173 // DATARATE CONTROL OPTIONS
174
175 // vbr, cbr, constrained quality or constant quality
176 enum vpx_rc_mode rc_mode;
177
178 // buffer targeting aggressiveness
179 int under_shoot_pct;
180 int over_shoot_pct;
181
182 // buffering parameters
183 int64_t starting_buffer_level_ms;
184 int64_t optimal_buffer_level_ms;
185 int64_t maximum_buffer_size_ms;
186
187 // Frame drop threshold.
188 int drop_frames_water_mark;
189
190 // controlling quality
191 int fixed_q;
192 int worst_allowed_q;
193 int best_allowed_q;
194 int cq_level;
195 AQ_MODE aq_mode; // Adaptive Quantization mode
196
197 // Special handling of Adaptive Quantization for AltRef frames
198 int alt_ref_aq;
199
200 // Internal frame size scaling.
201 RESIZE_TYPE resize_mode;
202 int scaled_frame_width;
203 int scaled_frame_height;
204
205 // Enable feature to reduce the frame quantization every x frames.
206 int frame_periodic_boost;
207
208 // two pass datarate control
209 int two_pass_vbrbias; // two pass datarate control tweaks
210 int two_pass_vbrmin_section;
211 int two_pass_vbrmax_section;
212 int vbr_corpus_complexity; // 0 indicates corpus vbr disabled
213 // END DATARATE CONTROL OPTIONS
214 // ----------------------------------------------------------------
215
216 // Spatial and temporal scalability.
217 int ss_number_layers; // Number of spatial layers.
218 int ts_number_layers; // Number of temporal layers.
219 // Bitrate allocation for spatial layers.
220 int layer_target_bitrate[VPX_MAX_LAYERS];
221 int ss_target_bitrate[VPX_SS_MAX_LAYERS];
222 int ss_enable_auto_arf[VPX_SS_MAX_LAYERS];
223 // Bitrate allocation (CBR mode) and framerate factor, for temporal layers.
224 int ts_rate_decimator[VPX_TS_MAX_LAYERS];
225
226 int enable_auto_arf;
227
228 int encode_breakout; // early breakout : for video conf recommend 800
229
230 /* Bitfield defining the error resiliency features to enable.
231 * Can provide decodable frames after losses in previous
232 * frames and decodable partitions after losses in the same frame.
233 */
234 unsigned int error_resilient_mode;
235
236 /* Bitfield defining the parallel decoding mode where the
237 * decoding in successive frames may be conducted in parallel
238 * just by decoding the frame headers.
239 */
240 unsigned int frame_parallel_decoding_mode;
241
242 int arnr_max_frames;
243 int arnr_strength;
244
245 int min_gf_interval;
246 int max_gf_interval;
247
248 int tile_columns;
249 int tile_rows;
250
251 int max_threads;
252
253 unsigned int target_level;
254
255 vpx_fixed_buf_t two_pass_stats_in;
256 struct vpx_codec_pkt_list *output_pkt_list;
257
258 #if CONFIG_FP_MB_STATS
259 vpx_fixed_buf_t firstpass_mb_stats_in;
260 #endif
261
262 vp8e_tuning tuning;
263 vp9e_tune_content content;
264 #if CONFIG_VP9_HIGHBITDEPTH
265 int use_highbitdepth;
266 #endif
267 vpx_color_space_t color_space;
268 vpx_color_range_t color_range;
269 int render_width;
270 int render_height;
271 VP9E_TEMPORAL_LAYERING_MODE temporal_layering_mode;
272
273 int row_mt;
274 unsigned int motion_vector_unit_test;
275 } VP9EncoderConfig;
276
is_lossless_requested(const VP9EncoderConfig * cfg)277 static INLINE int is_lossless_requested(const VP9EncoderConfig *cfg) {
278 return cfg->best_allowed_q == 0 && cfg->worst_allowed_q == 0;
279 }
280
281 // TODO(jingning) All spatially adaptive variables should go to TileDataEnc.
282 typedef struct TileDataEnc {
283 TileInfo tile_info;
284 int thresh_freq_fact[BLOCK_SIZES][MAX_MODES];
285 int mode_map[BLOCK_SIZES][MAX_MODES];
286 FIRSTPASS_DATA fp_data;
287 VP9RowMTSync row_mt_sync;
288
289 // Used for adaptive_rd_thresh with row multithreading
290 int *row_base_thresh_freq_fact;
291 } TileDataEnc;
292
293 typedef struct RowMTInfo {
294 JobQueueHandle job_queue_hdl;
295 #if CONFIG_MULTITHREAD
296 pthread_mutex_t job_mutex;
297 #endif
298 } RowMTInfo;
299
300 typedef struct {
301 TOKENEXTRA *start;
302 TOKENEXTRA *stop;
303 unsigned int count;
304 } TOKENLIST;
305
306 typedef struct MultiThreadHandle {
307 int allocated_tile_rows;
308 int allocated_tile_cols;
309 int allocated_vert_unit_rows;
310
311 // Frame level params
312 int num_tile_vert_sbs[MAX_NUM_TILE_ROWS];
313
314 // Job Queue structure and handles
315 JobQueue *job_queue;
316
317 int jobs_per_tile_col;
318
319 RowMTInfo row_mt_info[MAX_NUM_TILE_COLS];
320 int thread_id_to_tile_id[MAX_NUM_THREADS]; // Mapping of threads to tiles
321 } MultiThreadHandle;
322
323 typedef struct RD_COUNTS {
324 vp9_coeff_count coef_counts[TX_SIZES][PLANE_TYPES];
325 int64_t comp_pred_diff[REFERENCE_MODES];
326 int64_t filter_diff[SWITCHABLE_FILTER_CONTEXTS];
327 } RD_COUNTS;
328
329 typedef struct ThreadData {
330 MACROBLOCK mb;
331 RD_COUNTS rd_counts;
332 FRAME_COUNTS *counts;
333
334 PICK_MODE_CONTEXT *leaf_tree;
335 PC_TREE *pc_tree;
336 PC_TREE *pc_root;
337 } ThreadData;
338
339 struct EncWorkerData;
340
341 typedef struct ActiveMap {
342 int enabled;
343 int update;
344 unsigned char *map;
345 } ActiveMap;
346
347 typedef enum { Y, U, V, ALL } STAT_TYPE;
348
349 typedef struct IMAGE_STAT {
350 double stat[ALL + 1];
351 double worst;
352 } ImageStat;
353
354 // Kf noise filtering currently disabled by default in build.
355 // #define ENABLE_KF_DENOISE 1
356
357 #define CPB_WINDOW_SIZE 4
358 #define FRAME_WINDOW_SIZE 128
359 #define SAMPLE_RATE_GRACE_P 0.015
360 #define VP9_LEVELS 14
361
362 typedef enum {
363 LEVEL_UNKNOWN = 0,
364 LEVEL_AUTO = 1,
365 LEVEL_1 = 10,
366 LEVEL_1_1 = 11,
367 LEVEL_2 = 20,
368 LEVEL_2_1 = 21,
369 LEVEL_3 = 30,
370 LEVEL_3_1 = 31,
371 LEVEL_4 = 40,
372 LEVEL_4_1 = 41,
373 LEVEL_5 = 50,
374 LEVEL_5_1 = 51,
375 LEVEL_5_2 = 52,
376 LEVEL_6 = 60,
377 LEVEL_6_1 = 61,
378 LEVEL_6_2 = 62,
379 LEVEL_MAX = 255
380 } VP9_LEVEL;
381
382 typedef struct {
383 VP9_LEVEL level;
384 uint64_t max_luma_sample_rate;
385 uint32_t max_luma_picture_size;
386 uint32_t max_luma_picture_breadth;
387 double average_bitrate; // in kilobits per second
388 double max_cpb_size; // in kilobits
389 double compression_ratio;
390 uint8_t max_col_tiles;
391 uint32_t min_altref_distance;
392 uint8_t max_ref_frame_buffers;
393 } Vp9LevelSpec;
394
395 extern const Vp9LevelSpec vp9_level_defs[VP9_LEVELS];
396
397 typedef struct {
398 int64_t ts; // timestamp
399 uint32_t luma_samples;
400 uint32_t size; // in bytes
401 } FrameRecord;
402
403 typedef struct {
404 FrameRecord buf[FRAME_WINDOW_SIZE];
405 uint8_t start;
406 uint8_t len;
407 } FrameWindowBuffer;
408
409 typedef struct {
410 uint8_t seen_first_altref;
411 uint32_t frames_since_last_altref;
412 uint64_t total_compressed_size;
413 uint64_t total_uncompressed_size;
414 double time_encoded; // in seconds
415 FrameWindowBuffer frame_window_buffer;
416 int ref_refresh_map;
417 } Vp9LevelStats;
418
419 typedef struct {
420 Vp9LevelStats level_stats;
421 Vp9LevelSpec level_spec;
422 } Vp9LevelInfo;
423
424 typedef enum {
425 BITRATE_TOO_LARGE = 0,
426 LUMA_PIC_SIZE_TOO_LARGE,
427 LUMA_PIC_BREADTH_TOO_LARGE,
428 LUMA_SAMPLE_RATE_TOO_LARGE,
429 CPB_TOO_LARGE,
430 COMPRESSION_RATIO_TOO_SMALL,
431 TOO_MANY_COLUMN_TILE,
432 ALTREF_DIST_TOO_SMALL,
433 TOO_MANY_REF_BUFFER,
434 TARGET_LEVEL_FAIL_IDS
435 } TARGET_LEVEL_FAIL_ID;
436
437 typedef struct {
438 int8_t level_index;
439 uint8_t rc_config_updated;
440 uint8_t fail_flag;
441 int max_frame_size; // in bits
442 double max_cpb_size; // in bits
443 } LevelConstraint;
444
445 typedef struct ARNRFilterData {
446 YV12_BUFFER_CONFIG *frames[MAX_LAG_BUFFERS];
447 int strength;
448 int frame_count;
449 int alt_ref_index;
450 struct scale_factors sf;
451 } ARNRFilterData;
452
453 typedef struct VP9_COMP {
454 QUANTS quants;
455 ThreadData td;
456 MB_MODE_INFO_EXT *mbmi_ext_base;
457 DECLARE_ALIGNED(16, int16_t, y_dequant[QINDEX_RANGE][8]);
458 DECLARE_ALIGNED(16, int16_t, uv_dequant[QINDEX_RANGE][8]);
459 VP9_COMMON common;
460 VP9EncoderConfig oxcf;
461 struct lookahead_ctx *lookahead;
462 struct lookahead_entry *alt_ref_source;
463
464 YV12_BUFFER_CONFIG *Source;
465 YV12_BUFFER_CONFIG *Last_Source; // NULL for first frame and alt_ref frames
466 YV12_BUFFER_CONFIG *un_scaled_source;
467 YV12_BUFFER_CONFIG scaled_source;
468 YV12_BUFFER_CONFIG *unscaled_last_source;
469 YV12_BUFFER_CONFIG scaled_last_source;
470 #ifdef ENABLE_KF_DENOISE
471 YV12_BUFFER_CONFIG raw_unscaled_source;
472 YV12_BUFFER_CONFIG raw_scaled_source;
473 #endif
474 YV12_BUFFER_CONFIG *raw_source_frame;
475
476 TileDataEnc *tile_data;
477 int allocated_tiles; // Keep track of memory allocated for tiles.
478
479 // For a still frame, this flag is set to 1 to skip partition search.
480 int partition_search_skippable_frame;
481
482 int scaled_ref_idx[MAX_REF_FRAMES];
483 int lst_fb_idx;
484 int gld_fb_idx;
485 int alt_fb_idx;
486
487 int refresh_last_frame;
488 int refresh_golden_frame;
489 int refresh_alt_ref_frame;
490
491 int ext_refresh_frame_flags_pending;
492 int ext_refresh_last_frame;
493 int ext_refresh_golden_frame;
494 int ext_refresh_alt_ref_frame;
495
496 int ext_refresh_frame_context_pending;
497 int ext_refresh_frame_context;
498
499 YV12_BUFFER_CONFIG last_frame_uf;
500
501 TOKENEXTRA *tile_tok[4][1 << 6];
502 uint32_t tok_count[4][1 << 6];
503 TOKENLIST *tplist[4][1 << 6];
504
505 // Ambient reconstruction err target for force key frames
506 int64_t ambient_err;
507
508 RD_OPT rd;
509
510 CODING_CONTEXT coding_context;
511
512 int *nmvcosts[2];
513 int *nmvcosts_hp[2];
514 int *nmvsadcosts[2];
515 int *nmvsadcosts_hp[2];
516
517 int64_t last_time_stamp_seen;
518 int64_t last_end_time_stamp_seen;
519 int64_t first_time_stamp_ever;
520
521 RATE_CONTROL rc;
522 double framerate;
523
524 int interp_filter_selected[MAX_REF_FRAMES][SWITCHABLE];
525
526 struct vpx_codec_pkt_list *output_pkt_list;
527
528 MBGRAPH_FRAME_STATS mbgraph_stats[MAX_LAG_BUFFERS];
529 int mbgraph_n_frames; // number of frames filled in the above
530 int static_mb_pct; // % forced skip mbs by segmentation
531 int ref_frame_flags;
532
533 SPEED_FEATURES sf;
534
535 uint32_t max_mv_magnitude;
536 int mv_step_param;
537
538 int allow_comp_inter_inter;
539
540 // Default value is 1. From first pass stats, encode_breakout may be disabled.
541 ENCODE_BREAKOUT_TYPE allow_encode_breakout;
542
543 // Get threshold from external input. A suggested threshold is 800 for HD
544 // clips, and 300 for < HD clips.
545 int encode_breakout;
546
547 uint8_t *segmentation_map;
548
549 uint8_t *skin_map;
550
551 // segment threashold for encode breakout
552 int segment_encode_breakout[MAX_SEGMENTS];
553
554 CYCLIC_REFRESH *cyclic_refresh;
555 ActiveMap active_map;
556
557 fractional_mv_step_fp *find_fractional_mv_step;
558 vp9_diamond_search_fn_t diamond_search_sad;
559 vp9_variance_fn_ptr_t fn_ptr[BLOCK_SIZES];
560 uint64_t time_receive_data;
561 uint64_t time_compress_data;
562 uint64_t time_pick_lpf;
563 uint64_t time_encode_sb_row;
564
565 #if CONFIG_FP_MB_STATS
566 int use_fp_mb_stats;
567 #endif
568
569 TWO_PASS twopass;
570
571 // Force recalculation of segment_ids for each mode info
572 uint8_t force_update_segmentation;
573
574 YV12_BUFFER_CONFIG alt_ref_buffer;
575
576 // class responsible for adaptive
577 // quantization of altref frames
578 struct ALT_REF_AQ *alt_ref_aq;
579
580 #if CONFIG_INTERNAL_STATS
581 unsigned int mode_chosen_counts[MAX_MODES];
582
583 int count;
584 uint64_t total_sq_error;
585 uint64_t total_samples;
586 ImageStat psnr;
587
588 uint64_t totalp_sq_error;
589 uint64_t totalp_samples;
590 ImageStat psnrp;
591
592 double total_blockiness;
593 double worst_blockiness;
594
595 int bytes;
596 double summed_quality;
597 double summed_weights;
598 double summedp_quality;
599 double summedp_weights;
600 unsigned int tot_recode_hits;
601 double worst_ssim;
602
603 ImageStat ssimg;
604 ImageStat fastssim;
605 ImageStat psnrhvs;
606
607 int b_calculate_ssimg;
608 int b_calculate_blockiness;
609
610 int b_calculate_consistency;
611
612 double total_inconsistency;
613 double worst_consistency;
614 Ssimv *ssim_vars;
615 Metrics metrics;
616 #endif
617 int b_calculate_psnr;
618
619 int droppable;
620
621 int initial_width;
622 int initial_height;
623 int initial_mbs; // Number of MBs in the full-size frame; to be used to
624 // normalize the firstpass stats. This will differ from the
625 // number of MBs in the current frame when the frame is
626 // scaled.
627
628 int use_svc;
629
630 SVC svc;
631
632 // Store frame variance info in SOURCE_VAR_BASED_PARTITION search type.
633 diff *source_diff_var;
634 // The threshold used in SOURCE_VAR_BASED_PARTITION search type.
635 unsigned int source_var_thresh;
636 int frames_till_next_var_check;
637
638 int frame_flags;
639
640 search_site_config ss_cfg;
641
642 int mbmode_cost[INTRA_MODES];
643 unsigned int inter_mode_cost[INTER_MODE_CONTEXTS][INTER_MODES];
644 int intra_uv_mode_cost[FRAME_TYPES][INTRA_MODES][INTRA_MODES];
645 int y_mode_costs[INTRA_MODES][INTRA_MODES][INTRA_MODES];
646 int switchable_interp_costs[SWITCHABLE_FILTER_CONTEXTS][SWITCHABLE_FILTERS];
647 int partition_cost[PARTITION_CONTEXTS][PARTITION_TYPES];
648
649 int multi_arf_allowed;
650 int multi_arf_enabled;
651 int multi_arf_last_grp_enabled;
652
653 #if CONFIG_VP9_TEMPORAL_DENOISING
654 VP9_DENOISER denoiser;
655 #endif
656
657 int resize_pending;
658 RESIZE_STATE resize_state;
659 int external_resize;
660 int resize_scale_num;
661 int resize_scale_den;
662 int resize_avg_qp;
663 int resize_buffer_underflow;
664 int resize_count;
665
666 int use_skin_detection;
667
668 int target_level;
669
670 NOISE_ESTIMATE noise_estimate;
671
672 // Count on how many consecutive times a block uses small/zeromv for encoding.
673 uint8_t *consec_zero_mv;
674
675 // VAR_BASED_PARTITION thresholds
676 // 0 - threshold_64x64; 1 - threshold_32x32;
677 // 2 - threshold_16x16; 3 - vbp_threshold_8x8;
678 int64_t vbp_thresholds[4];
679 int64_t vbp_threshold_minmax;
680 int64_t vbp_threshold_sad;
681 // Threshold used for partition copy
682 int64_t vbp_threshold_copy;
683 BLOCK_SIZE vbp_bsize_min;
684
685 // Multi-threading
686 int num_workers;
687 VPxWorker *workers;
688 struct EncWorkerData *tile_thr_data;
689 VP9LfSync lf_row_sync;
690 struct VP9BitstreamWorkerData *vp9_bitstream_worker_data;
691
692 int keep_level_stats;
693 Vp9LevelInfo level_info;
694 MultiThreadHandle multi_thread_ctxt;
695 void (*row_mt_sync_read_ptr)(VP9RowMTSync *const, int, int);
696 void (*row_mt_sync_write_ptr)(VP9RowMTSync *const, int, int, const int);
697 ARNRFilterData arnr_filter_data;
698
699 int row_mt;
700 unsigned int row_mt_bit_exact;
701
702 // Previous Partition Info
703 BLOCK_SIZE *prev_partition;
704 int8_t *prev_segment_id;
705 // Used to save the status of whether a block has a low variance in
706 // choose_partitioning. 0 for 64x64, 1~2 for 64x32, 3~4 for 32x64, 5~8 for
707 // 32x32, 9~24 for 16x16.
708 // This is for the last frame and is copied to the current frame
709 // when partition copy happens.
710 uint8_t *prev_variance_low;
711 uint8_t *copied_frame_cnt;
712 uint8_t max_copied_frame;
713 // If the last frame is dropped, we don't copy partition.
714 uint8_t last_frame_dropped;
715
716 // For each superblock: keeps track of the last time (in frame distance) the
717 // the superblock did not have low source sad.
718 uint8_t *content_state_sb_fd;
719
720 int compute_source_sad_onepass;
721
722 LevelConstraint level_constraint;
723
724 uint8_t *count_arf_frame_usage;
725 uint8_t *count_lastgolden_frame_usage;
726 } VP9_COMP;
727
728 void vp9_initialize_enc(void);
729
730 struct VP9_COMP *vp9_create_compressor(VP9EncoderConfig *oxcf,
731 BufferPool *const pool);
732 void vp9_remove_compressor(VP9_COMP *cpi);
733
734 void vp9_change_config(VP9_COMP *cpi, const VP9EncoderConfig *oxcf);
735
736 // receive a frames worth of data. caller can assume that a copy of this
737 // frame is made and not just a copy of the pointer..
738 int vp9_receive_raw_frame(VP9_COMP *cpi, vpx_enc_frame_flags_t frame_flags,
739 YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
740 int64_t end_time_stamp);
741
742 int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
743 size_t *size, uint8_t *dest, int64_t *time_stamp,
744 int64_t *time_end, int flush);
745
746 int vp9_get_preview_raw_frame(VP9_COMP *cpi, YV12_BUFFER_CONFIG *dest,
747 vp9_ppflags_t *flags);
748
749 int vp9_use_as_reference(VP9_COMP *cpi, int ref_frame_flags);
750
751 void vp9_update_reference(VP9_COMP *cpi, int ref_frame_flags);
752
753 int vp9_copy_reference_enc(VP9_COMP *cpi, VP9_REFFRAME ref_frame_flag,
754 YV12_BUFFER_CONFIG *sd);
755
756 int vp9_set_reference_enc(VP9_COMP *cpi, VP9_REFFRAME ref_frame_flag,
757 YV12_BUFFER_CONFIG *sd);
758
759 int vp9_update_entropy(VP9_COMP *cpi, int update);
760
761 int vp9_set_active_map(VP9_COMP *cpi, unsigned char *map, int rows, int cols);
762
763 int vp9_get_active_map(VP9_COMP *cpi, unsigned char *map, int rows, int cols);
764
765 int vp9_set_internal_size(VP9_COMP *cpi, VPX_SCALING horiz_mode,
766 VPX_SCALING vert_mode);
767
768 int vp9_set_size_literal(VP9_COMP *cpi, unsigned int width,
769 unsigned int height);
770
771 void vp9_set_svc(VP9_COMP *cpi, int use_svc);
772
773 int vp9_get_quantizer(struct VP9_COMP *cpi);
774
frame_is_kf_gf_arf(const VP9_COMP * cpi)775 static INLINE int frame_is_kf_gf_arf(const VP9_COMP *cpi) {
776 return frame_is_intra_only(&cpi->common) || cpi->refresh_alt_ref_frame ||
777 (cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref);
778 }
779
get_ref_frame_map_idx(const VP9_COMP * cpi,MV_REFERENCE_FRAME ref_frame)780 static INLINE int get_ref_frame_map_idx(const VP9_COMP *cpi,
781 MV_REFERENCE_FRAME ref_frame) {
782 if (ref_frame == LAST_FRAME) {
783 return cpi->lst_fb_idx;
784 } else if (ref_frame == GOLDEN_FRAME) {
785 return cpi->gld_fb_idx;
786 } else {
787 return cpi->alt_fb_idx;
788 }
789 }
790
get_ref_frame_buf_idx(const VP9_COMP * const cpi,int ref_frame)791 static INLINE int get_ref_frame_buf_idx(const VP9_COMP *const cpi,
792 int ref_frame) {
793 const VP9_COMMON *const cm = &cpi->common;
794 const int map_idx = get_ref_frame_map_idx(cpi, ref_frame);
795 return (map_idx != INVALID_IDX) ? cm->ref_frame_map[map_idx] : INVALID_IDX;
796 }
797
get_ref_frame_buffer(VP9_COMP * cpi,MV_REFERENCE_FRAME ref_frame)798 static INLINE YV12_BUFFER_CONFIG *get_ref_frame_buffer(
799 VP9_COMP *cpi, MV_REFERENCE_FRAME ref_frame) {
800 VP9_COMMON *const cm = &cpi->common;
801 const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
802 return buf_idx != INVALID_IDX ? &cm->buffer_pool->frame_bufs[buf_idx].buf
803 : NULL;
804 }
805
get_token_alloc(int mb_rows,int mb_cols)806 static INLINE int get_token_alloc(int mb_rows, int mb_cols) {
807 // TODO(JBB): double check we can't exceed this token count if we have a
808 // 32x32 transform crossing a boundary at a multiple of 16.
809 // mb_rows, cols are in units of 16 pixels. We assume 3 planes all at full
810 // resolution. We assume up to 1 token per pixel, and then allow
811 // a head room of 4.
812 return mb_rows * mb_cols * (16 * 16 * 3 + 4);
813 }
814
815 // Get the allocated token size for a tile. It does the same calculation as in
816 // the frame token allocation.
allocated_tokens(TileInfo tile)817 static INLINE int allocated_tokens(TileInfo tile) {
818 int tile_mb_rows = (tile.mi_row_end - tile.mi_row_start + 1) >> 1;
819 int tile_mb_cols = (tile.mi_col_end - tile.mi_col_start + 1) >> 1;
820
821 return get_token_alloc(tile_mb_rows, tile_mb_cols);
822 }
823
get_start_tok(VP9_COMP * cpi,int tile_row,int tile_col,int mi_row,TOKENEXTRA ** tok)824 static INLINE void get_start_tok(VP9_COMP *cpi, int tile_row, int tile_col,
825 int mi_row, TOKENEXTRA **tok) {
826 VP9_COMMON *const cm = &cpi->common;
827 const int tile_cols = 1 << cm->log2_tile_cols;
828 TileDataEnc *this_tile = &cpi->tile_data[tile_row * tile_cols + tile_col];
829 const TileInfo *const tile_info = &this_tile->tile_info;
830
831 int tile_mb_cols = (tile_info->mi_col_end - tile_info->mi_col_start + 1) >> 1;
832 const int mb_row = (mi_row - tile_info->mi_row_start) >> 1;
833
834 *tok =
835 cpi->tile_tok[tile_row][tile_col] + get_token_alloc(mb_row, tile_mb_cols);
836 }
837
838 int64_t vp9_get_y_sse(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b);
839 #if CONFIG_VP9_HIGHBITDEPTH
840 int64_t vp9_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
841 const YV12_BUFFER_CONFIG *b);
842 #endif // CONFIG_VP9_HIGHBITDEPTH
843
844 void vp9_scale_references(VP9_COMP *cpi);
845
846 void vp9_update_reference_frames(VP9_COMP *cpi);
847
848 void vp9_set_high_precision_mv(VP9_COMP *cpi, int allow_high_precision_mv);
849
850 YV12_BUFFER_CONFIG *vp9_svc_twostage_scale(
851 VP9_COMMON *cm, YV12_BUFFER_CONFIG *unscaled, YV12_BUFFER_CONFIG *scaled,
852 YV12_BUFFER_CONFIG *scaled_temp, INTERP_FILTER filter_type,
853 int phase_scaler, INTERP_FILTER filter_type2, int phase_scaler2);
854
855 YV12_BUFFER_CONFIG *vp9_scale_if_required(
856 VP9_COMMON *cm, YV12_BUFFER_CONFIG *unscaled, YV12_BUFFER_CONFIG *scaled,
857 int use_normative_scaler, INTERP_FILTER filter_type, int phase_scaler);
858
859 void vp9_apply_encoding_flags(VP9_COMP *cpi, vpx_enc_frame_flags_t flags);
860
is_two_pass_svc(const struct VP9_COMP * const cpi)861 static INLINE int is_two_pass_svc(const struct VP9_COMP *const cpi) {
862 return cpi->use_svc && cpi->oxcf.pass != 0;
863 }
864
is_one_pass_cbr_svc(const struct VP9_COMP * const cpi)865 static INLINE int is_one_pass_cbr_svc(const struct VP9_COMP *const cpi) {
866 return (cpi->use_svc && cpi->oxcf.pass == 0);
867 }
868
869 #if CONFIG_VP9_TEMPORAL_DENOISING
denoise_svc(const struct VP9_COMP * const cpi)870 static INLINE int denoise_svc(const struct VP9_COMP *const cpi) {
871 return (!cpi->use_svc ||
872 (cpi->use_svc &&
873 cpi->svc.spatial_layer_id >= cpi->svc.first_layer_denoise));
874 }
875 #endif
876
877 #define MIN_LOOKAHEAD_FOR_ARFS 4
is_altref_enabled(const VP9_COMP * const cpi)878 static INLINE int is_altref_enabled(const VP9_COMP *const cpi) {
879 return !(cpi->oxcf.mode == REALTIME && cpi->oxcf.rc_mode == VPX_CBR) &&
880 cpi->oxcf.lag_in_frames >= MIN_LOOKAHEAD_FOR_ARFS &&
881 (cpi->oxcf.enable_auto_arf &&
882 (!is_two_pass_svc(cpi) ||
883 cpi->oxcf.ss_enable_auto_arf[cpi->svc.spatial_layer_id]));
884 }
885
set_ref_ptrs(VP9_COMMON * cm,MACROBLOCKD * xd,MV_REFERENCE_FRAME ref0,MV_REFERENCE_FRAME ref1)886 static INLINE void set_ref_ptrs(VP9_COMMON *cm, MACROBLOCKD *xd,
887 MV_REFERENCE_FRAME ref0,
888 MV_REFERENCE_FRAME ref1) {
889 xd->block_refs[0] =
890 &cm->frame_refs[ref0 >= LAST_FRAME ? ref0 - LAST_FRAME : 0];
891 xd->block_refs[1] =
892 &cm->frame_refs[ref1 >= LAST_FRAME ? ref1 - LAST_FRAME : 0];
893 }
894
get_chessboard_index(const int frame_index)895 static INLINE int get_chessboard_index(const int frame_index) {
896 return frame_index & 0x1;
897 }
898
cond_cost_list(const struct VP9_COMP * cpi,int * cost_list)899 static INLINE int *cond_cost_list(const struct VP9_COMP *cpi, int *cost_list) {
900 return cpi->sf.mv.subpel_search_method != SUBPEL_TREE ? cost_list : NULL;
901 }
902
get_num_vert_units(TileInfo tile,int shift)903 static INLINE int get_num_vert_units(TileInfo tile, int shift) {
904 int num_vert_units =
905 (tile.mi_row_end - tile.mi_row_start + (1 << shift) - 1) >> shift;
906 return num_vert_units;
907 }
908
get_num_cols(TileInfo tile,int shift)909 static INLINE int get_num_cols(TileInfo tile, int shift) {
910 int num_cols =
911 (tile.mi_col_end - tile.mi_col_start + (1 << shift) - 1) >> shift;
912 return num_cols;
913 }
914
get_level_index(VP9_LEVEL level)915 static INLINE int get_level_index(VP9_LEVEL level) {
916 int i;
917 for (i = 0; i < VP9_LEVELS; ++i) {
918 if (level == vp9_level_defs[i].level) return i;
919 }
920 return -1;
921 }
922
923 // Return the log2 value of max column tiles corresponding to the level that
924 // the picture size fits into.
log_tile_cols_from_picsize_level(uint32_t width,uint32_t height)925 static INLINE int log_tile_cols_from_picsize_level(uint32_t width,
926 uint32_t height) {
927 int i;
928 const uint32_t pic_size = width * height;
929 const uint32_t pic_breadth = VPXMAX(width, height);
930 for (i = LEVEL_1; i < LEVEL_MAX; ++i) {
931 if (vp9_level_defs[i].max_luma_picture_size >= pic_size &&
932 vp9_level_defs[i].max_luma_picture_breadth >= pic_breadth) {
933 return get_msb(vp9_level_defs[i].max_col_tiles);
934 }
935 }
936 return INT_MAX;
937 }
938
939 VP9_LEVEL vp9_get_level(const Vp9LevelSpec *const level_spec);
940
941 void vp9_new_framerate(VP9_COMP *cpi, double framerate);
942
943 void vp9_set_row_mt(VP9_COMP *cpi);
944
945 #define LAYER_IDS_TO_IDX(sl, tl, num_tl) ((sl) * (num_tl) + (tl))
946
947 #ifdef __cplusplus
948 } // extern "C"
949 #endif
950
951 #endif // VP9_ENCODER_VP9_ENCODER_H_
952