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 VPX_VP9_ENCODER_VP9_ENCODER_H_
12 #define VPX_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 PSNR_AQ = 5,
123 // AQ based on lookahead temporal
124 // variance (only valid for altref frames)
125 LOOKAHEAD_AQ = 6,
126 AQ_MODE_COUNT // This should always be the last member of the enum
127 } AQ_MODE;
128
129 typedef enum {
130 RESIZE_NONE = 0, // No frame resizing allowed (except for SVC).
131 RESIZE_FIXED = 1, // All frames are coded at the specified dimension.
132 RESIZE_DYNAMIC = 2 // Coded size of each frame is determined by the codec.
133 } RESIZE_TYPE;
134
135 typedef enum {
136 kInvalid = 0,
137 kLowSadLowSumdiff = 1,
138 kLowSadHighSumdiff = 2,
139 kHighSadLowSumdiff = 3,
140 kHighSadHighSumdiff = 4,
141 kLowVarHighSumdiff = 5,
142 kVeryHighSad = 6,
143 } CONTENT_STATE_SB;
144
145 typedef struct VP9EncoderConfig {
146 BITSTREAM_PROFILE profile;
147 vpx_bit_depth_t bit_depth; // Codec bit-depth.
148 int width; // width of data passed to the compressor
149 int height; // height of data passed to the compressor
150 unsigned int input_bit_depth; // Input bit depth.
151 double init_framerate; // set to passed in framerate
152 int64_t target_bandwidth; // bandwidth to be used in bits per second
153
154 int noise_sensitivity; // pre processing blur: recommendation 0
155 int sharpness; // sharpening output: recommendation 0:
156 int speed;
157 // maximum allowed bitrate for any intra frame in % of bitrate target.
158 unsigned int rc_max_intra_bitrate_pct;
159 // maximum allowed bitrate for any inter frame in % of bitrate target.
160 unsigned int rc_max_inter_bitrate_pct;
161 // percent of rate boost for golden frame in CBR mode.
162 unsigned int gf_cbr_boost_pct;
163
164 MODE mode;
165 int pass;
166
167 // Key Framing Operations
168 int auto_key; // autodetect cut scenes and set the keyframes
169 int key_freq; // maximum distance to key frame.
170
171 int lag_in_frames; // how many frames lag before we start encoding
172
173 // ----------------------------------------------------------------
174 // DATARATE CONTROL OPTIONS
175
176 // vbr, cbr, constrained quality or constant quality
177 enum vpx_rc_mode rc_mode;
178
179 // buffer targeting aggressiveness
180 int under_shoot_pct;
181 int over_shoot_pct;
182
183 // buffering parameters
184 int64_t starting_buffer_level_ms;
185 int64_t optimal_buffer_level_ms;
186 int64_t maximum_buffer_size_ms;
187
188 // Frame drop threshold.
189 int drop_frames_water_mark;
190
191 // controlling quality
192 int fixed_q;
193 int worst_allowed_q;
194 int best_allowed_q;
195 int cq_level;
196 AQ_MODE aq_mode; // Adaptive Quantization mode
197
198 // Special handling of Adaptive Quantization for AltRef frames
199 int alt_ref_aq;
200
201 // Internal frame size scaling.
202 RESIZE_TYPE resize_mode;
203 int scaled_frame_width;
204 int scaled_frame_height;
205
206 // Enable feature to reduce the frame quantization every x frames.
207 int frame_periodic_boost;
208
209 // two pass datarate control
210 int two_pass_vbrbias; // two pass datarate control tweaks
211 int two_pass_vbrmin_section;
212 int two_pass_vbrmax_section;
213 int vbr_corpus_complexity; // 0 indicates corpus vbr disabled
214 // END DATARATE CONTROL OPTIONS
215 // ----------------------------------------------------------------
216
217 // Spatial and temporal scalability.
218 int ss_number_layers; // Number of spatial layers.
219 int ts_number_layers; // Number of temporal layers.
220 // Bitrate allocation for spatial layers.
221 int layer_target_bitrate[VPX_MAX_LAYERS];
222 int ss_target_bitrate[VPX_SS_MAX_LAYERS];
223 int ss_enable_auto_arf[VPX_SS_MAX_LAYERS];
224 // Bitrate allocation (CBR mode) and framerate factor, for temporal layers.
225 int ts_rate_decimator[VPX_TS_MAX_LAYERS];
226
227 int enable_auto_arf;
228
229 int encode_breakout; // early breakout : for video conf recommend 800
230
231 /* Bitfield defining the error resiliency features to enable.
232 * Can provide decodable frames after losses in previous
233 * frames and decodable partitions after losses in the same frame.
234 */
235 unsigned int error_resilient_mode;
236
237 /* Bitfield defining the parallel decoding mode where the
238 * decoding in successive frames may be conducted in parallel
239 * just by decoding the frame headers.
240 */
241 unsigned int frame_parallel_decoding_mode;
242
243 int arnr_max_frames;
244 int arnr_strength;
245
246 int min_gf_interval;
247 int max_gf_interval;
248
249 int tile_columns;
250 int tile_rows;
251
252 int enable_tpl_model;
253
254 int max_threads;
255
256 unsigned int target_level;
257
258 vpx_fixed_buf_t two_pass_stats_in;
259 struct vpx_codec_pkt_list *output_pkt_list;
260
261 #if CONFIG_FP_MB_STATS
262 vpx_fixed_buf_t firstpass_mb_stats_in;
263 #endif
264
265 vp8e_tuning tuning;
266 vp9e_tune_content content;
267 #if CONFIG_VP9_HIGHBITDEPTH
268 int use_highbitdepth;
269 #endif
270 vpx_color_space_t color_space;
271 vpx_color_range_t color_range;
272 int render_width;
273 int render_height;
274 VP9E_TEMPORAL_LAYERING_MODE temporal_layering_mode;
275
276 int row_mt;
277 unsigned int motion_vector_unit_test;
278 } VP9EncoderConfig;
279
is_lossless_requested(const VP9EncoderConfig * cfg)280 static INLINE int is_lossless_requested(const VP9EncoderConfig *cfg) {
281 return cfg->best_allowed_q == 0 && cfg->worst_allowed_q == 0;
282 }
283
284 typedef struct TplDepStats {
285 int64_t intra_cost;
286 int64_t inter_cost;
287 int64_t mc_flow;
288 int64_t mc_dep_cost;
289 int64_t mc_ref_cost;
290
291 int ref_frame_index;
292 int_mv mv;
293
294 #if CONFIG_NON_GREEDY_MV
295 int ready[3];
296 double mv_dist[3];
297 double mv_cost[3];
298 int64_t inter_cost_arr[3];
299 int64_t recon_error_arr[3];
300 int64_t sse_arr[3];
301 double feature_score;
302 #endif
303 } TplDepStats;
304
305 #if CONFIG_NON_GREEDY_MV
306 #define SQUARE_BLOCK_SIZES 4
307 #endif
308
309 typedef struct TplDepFrame {
310 uint8_t is_valid;
311 TplDepStats *tpl_stats_ptr;
312 int stride;
313 int width;
314 int height;
315 int mi_rows;
316 int mi_cols;
317 int base_qindex;
318 #if CONFIG_NON_GREEDY_MV
319 double lambda;
320 double mv_dist_sum[3];
321 double mv_cost_sum[3];
322 int_mv *pyramid_mv_arr[3][SQUARE_BLOCK_SIZES];
323 #endif
324 } TplDepFrame;
325
326 #if CONFIG_NON_GREEDY_MV
get_square_block_idx(BLOCK_SIZE bsize)327 static INLINE int get_square_block_idx(BLOCK_SIZE bsize) {
328 if (bsize == BLOCK_4X4) {
329 return 0;
330 }
331 if (bsize == BLOCK_8X8) {
332 return 1;
333 }
334 if (bsize == BLOCK_16X16) {
335 return 2;
336 }
337 if (bsize == BLOCK_32X32) {
338 return 3;
339 }
340 printf("ERROR: non-square block size\n");
341 assert(0);
342 return -1;
343 }
344
square_block_idx_to_bsize(int square_block_idx)345 static INLINE BLOCK_SIZE square_block_idx_to_bsize(int square_block_idx) {
346 if (square_block_idx == 0) {
347 return BLOCK_4X4;
348 }
349 if (square_block_idx == 1) {
350 return BLOCK_8X8;
351 }
352 if (square_block_idx == 2) {
353 return BLOCK_16X16;
354 }
355 if (square_block_idx == 3) {
356 return BLOCK_32X32;
357 }
358 printf("ERROR: invalid square_block_idx\n");
359 assert(0);
360 return BLOCK_INVALID;
361 }
362
get_pyramid_mv(const TplDepFrame * tpl_frame,int rf_idx,BLOCK_SIZE bsize,int mi_row,int mi_col)363 static INLINE int_mv *get_pyramid_mv(const TplDepFrame *tpl_frame, int rf_idx,
364 BLOCK_SIZE bsize, int mi_row, int mi_col) {
365 return &tpl_frame->pyramid_mv_arr[rf_idx][get_square_block_idx(bsize)]
366 [mi_row * tpl_frame->stride + mi_col];
367 }
368 #endif
369
370 #define TPL_DEP_COST_SCALE_LOG2 4
371
372 // TODO(jingning) All spatially adaptive variables should go to TileDataEnc.
373 typedef struct TileDataEnc {
374 TileInfo tile_info;
375 int thresh_freq_fact[BLOCK_SIZES][MAX_MODES];
376 #if CONFIG_CONSISTENT_RECODE
377 int thresh_freq_fact_prev[BLOCK_SIZES][MAX_MODES];
378 #endif
379 int8_t mode_map[BLOCK_SIZES][MAX_MODES];
380 FIRSTPASS_DATA fp_data;
381 VP9RowMTSync row_mt_sync;
382
383 // Used for adaptive_rd_thresh with row multithreading
384 int *row_base_thresh_freq_fact;
385 } TileDataEnc;
386
387 typedef struct RowMTInfo {
388 JobQueueHandle job_queue_hdl;
389 #if CONFIG_MULTITHREAD
390 pthread_mutex_t job_mutex;
391 #endif
392 } RowMTInfo;
393
394 typedef struct {
395 TOKENEXTRA *start;
396 TOKENEXTRA *stop;
397 unsigned int count;
398 } TOKENLIST;
399
400 typedef struct MultiThreadHandle {
401 int allocated_tile_rows;
402 int allocated_tile_cols;
403 int allocated_vert_unit_rows;
404
405 // Frame level params
406 int num_tile_vert_sbs[MAX_NUM_TILE_ROWS];
407
408 // Job Queue structure and handles
409 JobQueue *job_queue;
410
411 int jobs_per_tile_col;
412
413 RowMTInfo row_mt_info[MAX_NUM_TILE_COLS];
414 int thread_id_to_tile_id[MAX_NUM_THREADS]; // Mapping of threads to tiles
415 } MultiThreadHandle;
416
417 typedef struct RD_COUNTS {
418 vp9_coeff_count coef_counts[TX_SIZES][PLANE_TYPES];
419 int64_t comp_pred_diff[REFERENCE_MODES];
420 int64_t filter_diff[SWITCHABLE_FILTER_CONTEXTS];
421 } RD_COUNTS;
422
423 typedef struct ThreadData {
424 MACROBLOCK mb;
425 RD_COUNTS rd_counts;
426 FRAME_COUNTS *counts;
427
428 PICK_MODE_CONTEXT *leaf_tree;
429 PC_TREE *pc_tree;
430 PC_TREE *pc_root;
431 } ThreadData;
432
433 struct EncWorkerData;
434
435 typedef struct ActiveMap {
436 int enabled;
437 int update;
438 unsigned char *map;
439 } ActiveMap;
440
441 typedef enum { Y, U, V, ALL } STAT_TYPE;
442
443 typedef struct IMAGE_STAT {
444 double stat[ALL + 1];
445 double worst;
446 } ImageStat;
447
448 // Kf noise filtering currently disabled by default in build.
449 // #define ENABLE_KF_DENOISE 1
450
451 #define CPB_WINDOW_SIZE 4
452 #define FRAME_WINDOW_SIZE 128
453 #define SAMPLE_RATE_GRACE_P 0.015
454 #define VP9_LEVELS 14
455
456 typedef enum {
457 LEVEL_UNKNOWN = 0,
458 LEVEL_AUTO = 1,
459 LEVEL_1 = 10,
460 LEVEL_1_1 = 11,
461 LEVEL_2 = 20,
462 LEVEL_2_1 = 21,
463 LEVEL_3 = 30,
464 LEVEL_3_1 = 31,
465 LEVEL_4 = 40,
466 LEVEL_4_1 = 41,
467 LEVEL_5 = 50,
468 LEVEL_5_1 = 51,
469 LEVEL_5_2 = 52,
470 LEVEL_6 = 60,
471 LEVEL_6_1 = 61,
472 LEVEL_6_2 = 62,
473 LEVEL_MAX = 255
474 } VP9_LEVEL;
475
476 typedef struct {
477 VP9_LEVEL level;
478 uint64_t max_luma_sample_rate;
479 uint32_t max_luma_picture_size;
480 uint32_t max_luma_picture_breadth;
481 double average_bitrate; // in kilobits per second
482 double max_cpb_size; // in kilobits
483 double compression_ratio;
484 uint8_t max_col_tiles;
485 uint32_t min_altref_distance;
486 uint8_t max_ref_frame_buffers;
487 } Vp9LevelSpec;
488
489 extern const Vp9LevelSpec vp9_level_defs[VP9_LEVELS];
490
491 typedef struct {
492 int64_t ts; // timestamp
493 uint32_t luma_samples;
494 uint32_t size; // in bytes
495 } FrameRecord;
496
497 typedef struct {
498 FrameRecord buf[FRAME_WINDOW_SIZE];
499 uint8_t start;
500 uint8_t len;
501 } FrameWindowBuffer;
502
503 typedef struct {
504 uint8_t seen_first_altref;
505 uint32_t frames_since_last_altref;
506 uint64_t total_compressed_size;
507 uint64_t total_uncompressed_size;
508 double time_encoded; // in seconds
509 FrameWindowBuffer frame_window_buffer;
510 int ref_refresh_map;
511 } Vp9LevelStats;
512
513 typedef struct {
514 Vp9LevelStats level_stats;
515 Vp9LevelSpec level_spec;
516 } Vp9LevelInfo;
517
518 typedef enum {
519 BITRATE_TOO_LARGE = 0,
520 LUMA_PIC_SIZE_TOO_LARGE,
521 LUMA_PIC_BREADTH_TOO_LARGE,
522 LUMA_SAMPLE_RATE_TOO_LARGE,
523 CPB_TOO_LARGE,
524 COMPRESSION_RATIO_TOO_SMALL,
525 TOO_MANY_COLUMN_TILE,
526 ALTREF_DIST_TOO_SMALL,
527 TOO_MANY_REF_BUFFER,
528 TARGET_LEVEL_FAIL_IDS
529 } TARGET_LEVEL_FAIL_ID;
530
531 typedef struct {
532 int8_t level_index;
533 uint8_t rc_config_updated;
534 uint8_t fail_flag;
535 int max_frame_size; // in bits
536 double max_cpb_size; // in bits
537 } LevelConstraint;
538
539 typedef struct ARNRFilterData {
540 YV12_BUFFER_CONFIG *frames[MAX_LAG_BUFFERS];
541 int strength;
542 int frame_count;
543 int alt_ref_index;
544 struct scale_factors sf;
545 } ARNRFilterData;
546
547 typedef struct EncFrameBuf {
548 int mem_valid;
549 int released;
550 YV12_BUFFER_CONFIG frame;
551 } EncFrameBuf;
552
553 // Maximum operating frame buffer size needed for a GOP using ARF reference.
554 #define MAX_ARF_GOP_SIZE (2 * MAX_LAG_BUFFERS)
555 #if CONFIG_NON_GREEDY_MV
556 typedef struct FEATURE_SCORE_LOC {
557 int visited;
558 double feature_score;
559 int mi_row;
560 int mi_col;
561 } FEATURE_SCORE_LOC;
562 #endif
563
564 typedef struct VP9_COMP {
565 QUANTS quants;
566 ThreadData td;
567 MB_MODE_INFO_EXT *mbmi_ext_base;
568 DECLARE_ALIGNED(16, int16_t, y_dequant[QINDEX_RANGE][8]);
569 DECLARE_ALIGNED(16, int16_t, uv_dequant[QINDEX_RANGE][8]);
570 VP9_COMMON common;
571 VP9EncoderConfig oxcf;
572 struct lookahead_ctx *lookahead;
573 struct lookahead_entry *alt_ref_source;
574
575 YV12_BUFFER_CONFIG *Source;
576 YV12_BUFFER_CONFIG *Last_Source; // NULL for first frame and alt_ref frames
577 YV12_BUFFER_CONFIG *un_scaled_source;
578 YV12_BUFFER_CONFIG scaled_source;
579 YV12_BUFFER_CONFIG *unscaled_last_source;
580 YV12_BUFFER_CONFIG scaled_last_source;
581 #ifdef ENABLE_KF_DENOISE
582 YV12_BUFFER_CONFIG raw_unscaled_source;
583 YV12_BUFFER_CONFIG raw_scaled_source;
584 #endif
585 YV12_BUFFER_CONFIG *raw_source_frame;
586
587 TplDepFrame tpl_stats[MAX_ARF_GOP_SIZE];
588 YV12_BUFFER_CONFIG *tpl_recon_frames[REF_FRAMES];
589 EncFrameBuf enc_frame_buf[REF_FRAMES];
590 #if CONFIG_NON_GREEDY_MV
591 int feature_score_loc_alloc;
592 FEATURE_SCORE_LOC *feature_score_loc_arr;
593 FEATURE_SCORE_LOC **feature_score_loc_sort;
594 FEATURE_SCORE_LOC **feature_score_loc_heap;
595 #endif
596
597 TileDataEnc *tile_data;
598 int allocated_tiles; // Keep track of memory allocated for tiles.
599
600 // For a still frame, this flag is set to 1 to skip partition search.
601 int partition_search_skippable_frame;
602
603 int scaled_ref_idx[REFS_PER_FRAME];
604 int lst_fb_idx;
605 int gld_fb_idx;
606 int alt_fb_idx;
607
608 int ref_fb_idx[REF_FRAMES];
609
610 int refresh_last_frame;
611 int refresh_golden_frame;
612 int refresh_alt_ref_frame;
613
614 int ext_refresh_frame_flags_pending;
615 int ext_refresh_last_frame;
616 int ext_refresh_golden_frame;
617 int ext_refresh_alt_ref_frame;
618
619 int ext_refresh_frame_context_pending;
620 int ext_refresh_frame_context;
621
622 YV12_BUFFER_CONFIG last_frame_uf;
623
624 TOKENEXTRA *tile_tok[4][1 << 6];
625 TOKENLIST *tplist[4][1 << 6];
626
627 // Ambient reconstruction err target for force key frames
628 int64_t ambient_err;
629
630 RD_OPT rd;
631
632 CODING_CONTEXT coding_context;
633
634 int *nmvcosts[2];
635 int *nmvcosts_hp[2];
636 int *nmvsadcosts[2];
637 int *nmvsadcosts_hp[2];
638
639 int64_t last_time_stamp_seen;
640 int64_t last_end_time_stamp_seen;
641 int64_t first_time_stamp_ever;
642
643 RATE_CONTROL rc;
644 double framerate;
645
646 int interp_filter_selected[REF_FRAMES][SWITCHABLE];
647
648 struct vpx_codec_pkt_list *output_pkt_list;
649
650 MBGRAPH_FRAME_STATS mbgraph_stats[MAX_LAG_BUFFERS];
651 int mbgraph_n_frames; // number of frames filled in the above
652 int static_mb_pct; // % forced skip mbs by segmentation
653 int ref_frame_flags;
654
655 SPEED_FEATURES sf;
656
657 uint32_t max_mv_magnitude;
658 int mv_step_param;
659
660 int allow_comp_inter_inter;
661
662 // Default value is 1. From first pass stats, encode_breakout may be disabled.
663 ENCODE_BREAKOUT_TYPE allow_encode_breakout;
664
665 // Get threshold from external input. A suggested threshold is 800 for HD
666 // clips, and 300 for < HD clips.
667 int encode_breakout;
668
669 uint8_t *segmentation_map;
670
671 uint8_t *skin_map;
672
673 // segment threashold for encode breakout
674 int segment_encode_breakout[MAX_SEGMENTS];
675
676 CYCLIC_REFRESH *cyclic_refresh;
677 ActiveMap active_map;
678
679 fractional_mv_step_fp *find_fractional_mv_step;
680 struct scale_factors me_sf;
681 vp9_diamond_search_fn_t diamond_search_sad;
682 vp9_variance_fn_ptr_t fn_ptr[BLOCK_SIZES];
683 uint64_t time_receive_data;
684 uint64_t time_compress_data;
685 uint64_t time_pick_lpf;
686 uint64_t time_encode_sb_row;
687
688 #if CONFIG_FP_MB_STATS
689 int use_fp_mb_stats;
690 #endif
691
692 TWO_PASS twopass;
693
694 // Force recalculation of segment_ids for each mode info
695 uint8_t force_update_segmentation;
696
697 YV12_BUFFER_CONFIG alt_ref_buffer;
698
699 // class responsible for adaptive
700 // quantization of altref frames
701 struct ALT_REF_AQ *alt_ref_aq;
702
703 #if CONFIG_INTERNAL_STATS
704 unsigned int mode_chosen_counts[MAX_MODES];
705
706 int count;
707 uint64_t total_sq_error;
708 uint64_t total_samples;
709 ImageStat psnr;
710
711 uint64_t totalp_sq_error;
712 uint64_t totalp_samples;
713 ImageStat psnrp;
714
715 double total_blockiness;
716 double worst_blockiness;
717
718 int bytes;
719 double summed_quality;
720 double summed_weights;
721 double summedp_quality;
722 double summedp_weights;
723 unsigned int tot_recode_hits;
724 double worst_ssim;
725
726 ImageStat ssimg;
727 ImageStat fastssim;
728 ImageStat psnrhvs;
729
730 int b_calculate_ssimg;
731 int b_calculate_blockiness;
732
733 int b_calculate_consistency;
734
735 double total_inconsistency;
736 double worst_consistency;
737 Ssimv *ssim_vars;
738 Metrics metrics;
739 #endif
740 int b_calculate_psnr;
741
742 int droppable;
743
744 int initial_width;
745 int initial_height;
746 int initial_mbs; // Number of MBs in the full-size frame; to be used to
747 // normalize the firstpass stats. This will differ from the
748 // number of MBs in the current frame when the frame is
749 // scaled.
750
751 int use_svc;
752
753 SVC svc;
754
755 // Store frame variance info in SOURCE_VAR_BASED_PARTITION search type.
756 diff *source_diff_var;
757 // The threshold used in SOURCE_VAR_BASED_PARTITION search type.
758 unsigned int source_var_thresh;
759 int frames_till_next_var_check;
760
761 int frame_flags;
762
763 search_site_config ss_cfg;
764
765 int mbmode_cost[INTRA_MODES];
766 unsigned int inter_mode_cost[INTER_MODE_CONTEXTS][INTER_MODES];
767 int intra_uv_mode_cost[FRAME_TYPES][INTRA_MODES][INTRA_MODES];
768 int y_mode_costs[INTRA_MODES][INTRA_MODES][INTRA_MODES];
769 int switchable_interp_costs[SWITCHABLE_FILTER_CONTEXTS][SWITCHABLE_FILTERS];
770 int partition_cost[PARTITION_CONTEXTS][PARTITION_TYPES];
771 // Indices are: max_tx_size-1, tx_size_ctx, tx_size
772 int tx_size_cost[TX_SIZES - 1][TX_SIZE_CONTEXTS][TX_SIZES];
773
774 #if CONFIG_VP9_TEMPORAL_DENOISING
775 VP9_DENOISER denoiser;
776 #endif
777
778 int resize_pending;
779 RESIZE_STATE resize_state;
780 int external_resize;
781 int resize_scale_num;
782 int resize_scale_den;
783 int resize_avg_qp;
784 int resize_buffer_underflow;
785 int resize_count;
786
787 int use_skin_detection;
788
789 int target_level;
790
791 NOISE_ESTIMATE noise_estimate;
792
793 // Count on how many consecutive times a block uses small/zeromv for encoding.
794 uint8_t *consec_zero_mv;
795
796 // VAR_BASED_PARTITION thresholds
797 // 0 - threshold_64x64; 1 - threshold_32x32;
798 // 2 - threshold_16x16; 3 - vbp_threshold_8x8;
799 int64_t vbp_thresholds[4];
800 int64_t vbp_threshold_minmax;
801 int64_t vbp_threshold_sad;
802 // Threshold used for partition copy
803 int64_t vbp_threshold_copy;
804 BLOCK_SIZE vbp_bsize_min;
805
806 // Multi-threading
807 int num_workers;
808 VPxWorker *workers;
809 struct EncWorkerData *tile_thr_data;
810 VP9LfSync lf_row_sync;
811 struct VP9BitstreamWorkerData *vp9_bitstream_worker_data;
812
813 int keep_level_stats;
814 Vp9LevelInfo level_info;
815 MultiThreadHandle multi_thread_ctxt;
816 void (*row_mt_sync_read_ptr)(VP9RowMTSync *const, int, int);
817 void (*row_mt_sync_write_ptr)(VP9RowMTSync *const, int, int, const int);
818 ARNRFilterData arnr_filter_data;
819
820 int row_mt;
821 unsigned int row_mt_bit_exact;
822
823 // Previous Partition Info
824 BLOCK_SIZE *prev_partition;
825 int8_t *prev_segment_id;
826 // Used to save the status of whether a block has a low variance in
827 // choose_partitioning. 0 for 64x64, 1~2 for 64x32, 3~4 for 32x64, 5~8 for
828 // 32x32, 9~24 for 16x16.
829 // This is for the last frame and is copied to the current frame
830 // when partition copy happens.
831 uint8_t *prev_variance_low;
832 uint8_t *copied_frame_cnt;
833 uint8_t max_copied_frame;
834 // If the last frame is dropped, we don't copy partition.
835 uint8_t last_frame_dropped;
836
837 // For each superblock: keeps track of the last time (in frame distance) the
838 // the superblock did not have low source sad.
839 uint8_t *content_state_sb_fd;
840
841 int compute_source_sad_onepass;
842
843 LevelConstraint level_constraint;
844
845 uint8_t *count_arf_frame_usage;
846 uint8_t *count_lastgolden_frame_usage;
847
848 int multi_layer_arf;
849 vpx_roi_map_t roi;
850 } VP9_COMP;
851
852 void vp9_initialize_enc(void);
853
854 struct VP9_COMP *vp9_create_compressor(VP9EncoderConfig *oxcf,
855 BufferPool *const pool);
856 void vp9_remove_compressor(VP9_COMP *cpi);
857
858 void vp9_change_config(VP9_COMP *cpi, const VP9EncoderConfig *oxcf);
859
860 // receive a frames worth of data. caller can assume that a copy of this
861 // frame is made and not just a copy of the pointer..
862 int vp9_receive_raw_frame(VP9_COMP *cpi, vpx_enc_frame_flags_t frame_flags,
863 YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
864 int64_t end_time);
865
866 int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
867 size_t *size, uint8_t *dest, int64_t *time_stamp,
868 int64_t *time_end, int flush);
869
870 int vp9_get_preview_raw_frame(VP9_COMP *cpi, YV12_BUFFER_CONFIG *dest,
871 vp9_ppflags_t *flags);
872
873 int vp9_use_as_reference(VP9_COMP *cpi, int ref_frame_flags);
874
875 void vp9_update_reference(VP9_COMP *cpi, int ref_frame_flags);
876
877 int vp9_copy_reference_enc(VP9_COMP *cpi, VP9_REFFRAME ref_frame_flag,
878 YV12_BUFFER_CONFIG *sd);
879
880 int vp9_set_reference_enc(VP9_COMP *cpi, VP9_REFFRAME ref_frame_flag,
881 YV12_BUFFER_CONFIG *sd);
882
883 int vp9_update_entropy(VP9_COMP *cpi, int update);
884
885 int vp9_set_active_map(VP9_COMP *cpi, unsigned char *new_map_16x16, int rows,
886 int cols);
887
888 int vp9_get_active_map(VP9_COMP *cpi, unsigned char *new_map_16x16, int rows,
889 int cols);
890
891 int vp9_set_internal_size(VP9_COMP *cpi, VPX_SCALING horiz_mode,
892 VPX_SCALING vert_mode);
893
894 int vp9_set_size_literal(VP9_COMP *cpi, unsigned int width,
895 unsigned int height);
896
897 void vp9_set_svc(VP9_COMP *cpi, int use_svc);
898
stack_pop(int * stack,int stack_size)899 static INLINE int stack_pop(int *stack, int stack_size) {
900 int idx;
901 const int r = stack[0];
902 for (idx = 1; idx < stack_size; ++idx) stack[idx - 1] = stack[idx];
903
904 return r;
905 }
906
stack_top(const int * stack)907 static INLINE int stack_top(const int *stack) { return stack[0]; }
908
stack_push(int * stack,int new_item,int stack_size)909 static INLINE void stack_push(int *stack, int new_item, int stack_size) {
910 int idx;
911 for (idx = stack_size; idx > 0; --idx) stack[idx] = stack[idx - 1];
912 stack[0] = new_item;
913 }
914
stack_init(int * stack,int length)915 static INLINE void stack_init(int *stack, int length) {
916 int idx;
917 for (idx = 0; idx < length; ++idx) stack[idx] = -1;
918 }
919
920 int vp9_get_quantizer(struct VP9_COMP *cpi);
921
frame_is_kf_gf_arf(const VP9_COMP * cpi)922 static INLINE int frame_is_kf_gf_arf(const VP9_COMP *cpi) {
923 return frame_is_intra_only(&cpi->common) || cpi->refresh_alt_ref_frame ||
924 (cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref);
925 }
926
get_ref_frame_map_idx(const VP9_COMP * cpi,MV_REFERENCE_FRAME ref_frame)927 static INLINE int get_ref_frame_map_idx(const VP9_COMP *cpi,
928 MV_REFERENCE_FRAME ref_frame) {
929 if (ref_frame == LAST_FRAME) {
930 return cpi->lst_fb_idx;
931 } else if (ref_frame == GOLDEN_FRAME) {
932 return cpi->gld_fb_idx;
933 } else {
934 return cpi->alt_fb_idx;
935 }
936 }
937
get_ref_frame_buf_idx(const VP9_COMP * const cpi,int ref_frame)938 static INLINE int get_ref_frame_buf_idx(const VP9_COMP *const cpi,
939 int ref_frame) {
940 const VP9_COMMON *const cm = &cpi->common;
941 const int map_idx = get_ref_frame_map_idx(cpi, ref_frame);
942 return (map_idx != INVALID_IDX) ? cm->ref_frame_map[map_idx] : INVALID_IDX;
943 }
944
get_ref_cnt_buffer(VP9_COMMON * cm,int fb_idx)945 static INLINE RefCntBuffer *get_ref_cnt_buffer(VP9_COMMON *cm, int fb_idx) {
946 return fb_idx != INVALID_IDX ? &cm->buffer_pool->frame_bufs[fb_idx] : NULL;
947 }
948
get_ref_frame_buffer(VP9_COMP * cpi,MV_REFERENCE_FRAME ref_frame)949 static INLINE YV12_BUFFER_CONFIG *get_ref_frame_buffer(
950 VP9_COMP *cpi, MV_REFERENCE_FRAME ref_frame) {
951 VP9_COMMON *const cm = &cpi->common;
952 const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
953 return buf_idx != INVALID_IDX ? &cm->buffer_pool->frame_bufs[buf_idx].buf
954 : NULL;
955 }
956
get_token_alloc(int mb_rows,int mb_cols)957 static INLINE int get_token_alloc(int mb_rows, int mb_cols) {
958 // TODO(JBB): double check we can't exceed this token count if we have a
959 // 32x32 transform crossing a boundary at a multiple of 16.
960 // mb_rows, cols are in units of 16 pixels. We assume 3 planes all at full
961 // resolution. We assume up to 1 token per pixel, and then allow
962 // a head room of 4.
963 return mb_rows * mb_cols * (16 * 16 * 3 + 4);
964 }
965
966 // Get the allocated token size for a tile. It does the same calculation as in
967 // the frame token allocation.
allocated_tokens(TileInfo tile)968 static INLINE int allocated_tokens(TileInfo tile) {
969 int tile_mb_rows = (tile.mi_row_end - tile.mi_row_start + 1) >> 1;
970 int tile_mb_cols = (tile.mi_col_end - tile.mi_col_start + 1) >> 1;
971
972 return get_token_alloc(tile_mb_rows, tile_mb_cols);
973 }
974
get_start_tok(VP9_COMP * cpi,int tile_row,int tile_col,int mi_row,TOKENEXTRA ** tok)975 static INLINE void get_start_tok(VP9_COMP *cpi, int tile_row, int tile_col,
976 int mi_row, TOKENEXTRA **tok) {
977 VP9_COMMON *const cm = &cpi->common;
978 const int tile_cols = 1 << cm->log2_tile_cols;
979 TileDataEnc *this_tile = &cpi->tile_data[tile_row * tile_cols + tile_col];
980 const TileInfo *const tile_info = &this_tile->tile_info;
981
982 int tile_mb_cols = (tile_info->mi_col_end - tile_info->mi_col_start + 1) >> 1;
983 const int mb_row = (mi_row - tile_info->mi_row_start) >> 1;
984
985 *tok =
986 cpi->tile_tok[tile_row][tile_col] + get_token_alloc(mb_row, tile_mb_cols);
987 }
988
989 int64_t vp9_get_y_sse(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b);
990 #if CONFIG_VP9_HIGHBITDEPTH
991 int64_t vp9_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
992 const YV12_BUFFER_CONFIG *b);
993 #endif // CONFIG_VP9_HIGHBITDEPTH
994
995 void vp9_scale_references(VP9_COMP *cpi);
996
997 void vp9_update_reference_frames(VP9_COMP *cpi);
998
999 void vp9_set_high_precision_mv(VP9_COMP *cpi, int allow_high_precision_mv);
1000
1001 YV12_BUFFER_CONFIG *vp9_svc_twostage_scale(
1002 VP9_COMMON *cm, YV12_BUFFER_CONFIG *unscaled, YV12_BUFFER_CONFIG *scaled,
1003 YV12_BUFFER_CONFIG *scaled_temp, INTERP_FILTER filter_type,
1004 int phase_scaler, INTERP_FILTER filter_type2, int phase_scaler2);
1005
1006 YV12_BUFFER_CONFIG *vp9_scale_if_required(
1007 VP9_COMMON *cm, YV12_BUFFER_CONFIG *unscaled, YV12_BUFFER_CONFIG *scaled,
1008 int use_normative_scaler, INTERP_FILTER filter_type, int phase_scaler);
1009
1010 void vp9_apply_encoding_flags(VP9_COMP *cpi, vpx_enc_frame_flags_t flags);
1011
is_one_pass_cbr_svc(const struct VP9_COMP * const cpi)1012 static INLINE int is_one_pass_cbr_svc(const struct VP9_COMP *const cpi) {
1013 return (cpi->use_svc && cpi->oxcf.pass == 0);
1014 }
1015
1016 #if CONFIG_VP9_TEMPORAL_DENOISING
denoise_svc(const struct VP9_COMP * const cpi)1017 static INLINE int denoise_svc(const struct VP9_COMP *const cpi) {
1018 return (!cpi->use_svc || (cpi->use_svc && cpi->svc.spatial_layer_id >=
1019 cpi->svc.first_layer_denoise));
1020 }
1021 #endif
1022
1023 #define MIN_LOOKAHEAD_FOR_ARFS 4
is_altref_enabled(const VP9_COMP * const cpi)1024 static INLINE int is_altref_enabled(const VP9_COMP *const cpi) {
1025 return !(cpi->oxcf.mode == REALTIME && cpi->oxcf.rc_mode == VPX_CBR) &&
1026 cpi->oxcf.lag_in_frames >= MIN_LOOKAHEAD_FOR_ARFS &&
1027 cpi->oxcf.enable_auto_arf;
1028 }
1029
set_ref_ptrs(VP9_COMMON * cm,MACROBLOCKD * xd,MV_REFERENCE_FRAME ref0,MV_REFERENCE_FRAME ref1)1030 static INLINE void set_ref_ptrs(VP9_COMMON *cm, MACROBLOCKD *xd,
1031 MV_REFERENCE_FRAME ref0,
1032 MV_REFERENCE_FRAME ref1) {
1033 xd->block_refs[0] =
1034 &cm->frame_refs[ref0 >= LAST_FRAME ? ref0 - LAST_FRAME : 0];
1035 xd->block_refs[1] =
1036 &cm->frame_refs[ref1 >= LAST_FRAME ? ref1 - LAST_FRAME : 0];
1037 }
1038
get_chessboard_index(const int frame_index)1039 static INLINE int get_chessboard_index(const int frame_index) {
1040 return frame_index & 0x1;
1041 }
1042
cond_cost_list(const struct VP9_COMP * cpi,int * cost_list)1043 static INLINE int *cond_cost_list(const struct VP9_COMP *cpi, int *cost_list) {
1044 return cpi->sf.mv.subpel_search_method != SUBPEL_TREE ? cost_list : NULL;
1045 }
1046
get_num_vert_units(TileInfo tile,int shift)1047 static INLINE int get_num_vert_units(TileInfo tile, int shift) {
1048 int num_vert_units =
1049 (tile.mi_row_end - tile.mi_row_start + (1 << shift) - 1) >> shift;
1050 return num_vert_units;
1051 }
1052
get_num_cols(TileInfo tile,int shift)1053 static INLINE int get_num_cols(TileInfo tile, int shift) {
1054 int num_cols =
1055 (tile.mi_col_end - tile.mi_col_start + (1 << shift) - 1) >> shift;
1056 return num_cols;
1057 }
1058
get_level_index(VP9_LEVEL level)1059 static INLINE int get_level_index(VP9_LEVEL level) {
1060 int i;
1061 for (i = 0; i < VP9_LEVELS; ++i) {
1062 if (level == vp9_level_defs[i].level) return i;
1063 }
1064 return -1;
1065 }
1066
1067 // Return the log2 value of max column tiles corresponding to the level that
1068 // the picture size fits into.
log_tile_cols_from_picsize_level(uint32_t width,uint32_t height)1069 static INLINE int log_tile_cols_from_picsize_level(uint32_t width,
1070 uint32_t height) {
1071 int i;
1072 const uint32_t pic_size = width * height;
1073 const uint32_t pic_breadth = VPXMAX(width, height);
1074 for (i = LEVEL_1; i < LEVEL_MAX; ++i) {
1075 if (vp9_level_defs[i].max_luma_picture_size >= pic_size &&
1076 vp9_level_defs[i].max_luma_picture_breadth >= pic_breadth) {
1077 return get_msb(vp9_level_defs[i].max_col_tiles);
1078 }
1079 }
1080 return INT_MAX;
1081 }
1082
1083 VP9_LEVEL vp9_get_level(const Vp9LevelSpec *const level_spec);
1084
1085 int vp9_set_roi_map(VP9_COMP *cpi, unsigned char *map, unsigned int rows,
1086 unsigned int cols, int delta_q[8], int delta_lf[8],
1087 int skip[8], int ref_frame[8]);
1088
1089 void vp9_new_framerate(VP9_COMP *cpi, double framerate);
1090
1091 void vp9_set_row_mt(VP9_COMP *cpi);
1092
1093 #define LAYER_IDS_TO_IDX(sl, tl, num_tl) ((sl) * (num_tl) + (tl))
1094
1095 #ifdef __cplusplus
1096 } // extern "C"
1097 #endif
1098
1099 #endif // VPX_VP9_ENCODER_VP9_ENCODER_H_
1100