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_ENCODER_H_
13 #define AOM_AV1_ENCODER_ENCODER_H_
14
15 #include <stdio.h>
16
17 #include "config/aom_config.h"
18
19 #include "aom/aomcx.h"
20
21 #include "av1/common/alloccommon.h"
22 #include "av1/common/entropymode.h"
23 #include "av1/common/thread_common.h"
24 #include "av1/common/onyxc_int.h"
25 #include "av1/common/resize.h"
26 #include "av1/common/timing.h"
27 #include "av1/encoder/aq_cyclicrefresh.h"
28 #include "av1/encoder/av1_quantize.h"
29 #include "av1/encoder/context_tree.h"
30 #include "av1/encoder/encodemb.h"
31 #include "av1/encoder/firstpass.h"
32 #include "av1/encoder/lookahead.h"
33 #include "av1/encoder/mbgraph.h"
34 #include "av1/encoder/mcomp.h"
35 #include "av1/encoder/ratectrl.h"
36 #include "av1/encoder/rd.h"
37 #include "av1/encoder/speed_features.h"
38 #include "av1/encoder/tokenize.h"
39
40 #if CONFIG_INTERNAL_STATS
41 #include "aom_dsp/ssim.h"
42 #endif
43 #include "aom_dsp/variance.h"
44 #if CONFIG_DENOISE
45 #include "aom_dsp/noise_model.h"
46 #endif
47 #include "aom/internal/aom_codec_internal.h"
48 #include "aom_util/aom_thread.h"
49
50 #ifdef __cplusplus
51 extern "C" {
52 #endif
53
54 typedef struct {
55 int nmv_vec_cost[MV_JOINTS];
56 int nmv_costs[2][MV_VALS];
57 int nmv_costs_hp[2][MV_VALS];
58
59 FRAME_CONTEXT fc;
60 } CODING_CONTEXT;
61
62 typedef enum {
63 // regular inter frame
64 REGULAR_FRAME = 0,
65 // alternate reference frame
66 ARF_FRAME = 1,
67 // overlay frame
68 OVERLAY_FRAME = 2,
69 // golden frame
70 GLD_FRAME = 3,
71 // backward reference frame
72 BRF_FRAME = 4,
73 // extra alternate reference frame
74 EXT_ARF_FRAME = 5,
75 FRAME_CONTEXT_INDEXES
76 } FRAME_CONTEXT_INDEX;
77
78 typedef enum {
79 NORMAL = 0,
80 FOURFIVE = 1,
81 THREEFIVE = 2,
82 ONETWO = 3
83 } AOM_SCALING;
84
85 typedef enum {
86 // Good Quality Fast Encoding. The encoder balances quality with the amount of
87 // time it takes to encode the output. Speed setting controls how fast.
88 GOOD
89 } MODE;
90
91 typedef enum {
92 FRAMEFLAGS_KEY = 1 << 0,
93 FRAMEFLAGS_GOLDEN = 1 << 1,
94 FRAMEFLAGS_BWDREF = 1 << 2,
95 // TODO(zoeliu): To determine whether a frame flag is needed for ALTREF2_FRAME
96 FRAMEFLAGS_ALTREF = 1 << 3,
97 } FRAMETYPE_FLAGS;
98
99 typedef enum {
100 NO_AQ = 0,
101 VARIANCE_AQ = 1,
102 COMPLEXITY_AQ = 2,
103 CYCLIC_REFRESH_AQ = 3,
104 AQ_MODE_COUNT // This should always be the last member of the enum
105 } AQ_MODE;
106 typedef enum {
107 NO_DELTA_Q = 0,
108 DELTA_Q_ONLY = 1,
109 DELTA_Q_LF = 2,
110 DELTAQ_MODE_COUNT // This should always be the last member of the enum
111 } DELTAQ_MODE;
112
113 typedef enum {
114 RESIZE_NONE = 0, // No frame resizing allowed.
115 RESIZE_FIXED = 1, // All frames are coded at the specified scale.
116 RESIZE_RANDOM = 2, // All frames are coded at a random scale.
117 RESIZE_MODES
118 } RESIZE_MODE;
119
120 typedef enum {
121 SUPERRES_NONE = 0, // No frame superres allowed
122 SUPERRES_FIXED = 1, // All frames are coded at the specified scale,
123 // and super-resolved.
124 SUPERRES_RANDOM = 2, // All frames are coded at a random scale,
125 // and super-resolved.
126 SUPERRES_QTHRESH = 3, // Superres scale for a frame is determined based on
127 // q_index
128 SUPERRES_MODES
129 } SUPERRES_MODE;
130
131 typedef struct AV1EncoderConfig {
132 BITSTREAM_PROFILE profile;
133 aom_bit_depth_t bit_depth; // Codec bit-depth.
134 int width; // width of data passed to the compressor
135 int height; // height of data passed to the compressor
136 int forced_max_frame_width; // forced maximum width of frame (if != 0)
137 int forced_max_frame_height; // forced maximum height of frame (if != 0)
138 unsigned int input_bit_depth; // Input bit depth.
139 double init_framerate; // set to passed in framerate
140 int64_t target_bandwidth; // bandwidth to be used in bits per second
141
142 int noise_sensitivity; // pre processing blur: recommendation 0
143 int sharpness; // sharpening output: recommendation 0:
144 int speed;
145 // maximum allowed bitrate for any intra frame in % of bitrate target.
146 unsigned int rc_max_intra_bitrate_pct;
147 // maximum allowed bitrate for any inter frame in % of bitrate target.
148 unsigned int rc_max_inter_bitrate_pct;
149 // percent of rate boost for golden frame in CBR mode.
150 unsigned int gf_cbr_boost_pct;
151
152 MODE mode;
153 int pass;
154
155 // Key Framing Operations
156 int auto_key; // autodetect cut scenes and set the keyframes
157 int key_freq; // maximum distance to key frame.
158 int sframe_dist;
159 int sframe_mode;
160 int sframe_enabled;
161 int lag_in_frames; // how many frames lag before we start encoding
162 int fwd_kf_enabled;
163
164 // ----------------------------------------------------------------
165 // DATARATE CONTROL OPTIONS
166
167 // vbr, cbr, constrained quality or constant quality
168 enum aom_rc_mode rc_mode;
169
170 // buffer targeting aggressiveness
171 int under_shoot_pct;
172 int over_shoot_pct;
173
174 // buffering parameters
175 int64_t starting_buffer_level_ms;
176 int64_t optimal_buffer_level_ms;
177 int64_t maximum_buffer_size_ms;
178
179 // Frame drop threshold.
180 int drop_frames_water_mark;
181
182 // controlling quality
183 int fixed_q;
184 int worst_allowed_q;
185 int best_allowed_q;
186 int cq_level;
187 AQ_MODE aq_mode; // Adaptive Quantization mode
188 DELTAQ_MODE deltaq_mode;
189 int enable_cdef;
190 int enable_restoration;
191 int disable_trellis_quant;
192 int using_qm;
193 int qm_y;
194 int qm_u;
195 int qm_v;
196 int qm_minlevel;
197 int qm_maxlevel;
198 #if CONFIG_DIST_8X8
199 int using_dist_8x8;
200 #endif
201 unsigned int num_tile_groups;
202 unsigned int mtu;
203
204 // Internal frame size scaling.
205 RESIZE_MODE resize_mode;
206 uint8_t resize_scale_denominator;
207 uint8_t resize_kf_scale_denominator;
208
209 // Frame Super-Resolution size scaling.
210 SUPERRES_MODE superres_mode;
211 uint8_t superres_scale_denominator;
212 uint8_t superres_kf_scale_denominator;
213 int superres_qthresh;
214 int superres_kf_qthresh;
215
216 // Enable feature to reduce the frame quantization every x frames.
217 int frame_periodic_boost;
218
219 // two pass datarate control
220 int two_pass_vbrbias; // two pass datarate control tweaks
221 int two_pass_vbrmin_section;
222 int two_pass_vbrmax_section;
223 // END DATARATE CONTROL OPTIONS
224 // ----------------------------------------------------------------
225
226 int enable_auto_arf;
227 int enable_auto_brf; // (b)ackward (r)ef (f)rame
228
229 /* Bitfield defining the error resiliency features to enable.
230 * Can provide decodable frames after losses in previous
231 * frames and decodable partitions after losses in the same frame.
232 */
233 unsigned int error_resilient_mode;
234
235 unsigned int s_frame_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 unsigned int limit;
244
245 int arnr_max_frames;
246 int arnr_strength;
247
248 int min_gf_interval;
249 int max_gf_interval;
250
251 int row_mt;
252 int tile_columns;
253 int tile_rows;
254 int tile_width_count;
255 int tile_height_count;
256 int tile_widths[MAX_TILE_COLS];
257 int tile_heights[MAX_TILE_ROWS];
258
259 int max_threads;
260
261 aom_fixed_buf_t two_pass_stats_in;
262 struct aom_codec_pkt_list *output_pkt_list;
263
264 #if CONFIG_FP_MB_STATS
265 aom_fixed_buf_t firstpass_mb_stats_in;
266 #endif
267
268 aom_tune_metric tuning;
269 aom_tune_content content;
270 int use_highbitdepth;
271 aom_color_primaries_t color_primaries;
272 aom_transfer_characteristics_t transfer_characteristics;
273 aom_matrix_coefficients_t matrix_coefficients;
274 aom_chroma_sample_position_t chroma_sample_position;
275 int color_range;
276 int render_width;
277 int render_height;
278 aom_timing_info_type_t timing_info_type;
279 int timing_info_present;
280 aom_timing_info_t timing_info;
281 int decoder_model_info_present_flag;
282 int display_model_info_present_flag;
283 int buffer_removal_time_present;
284 aom_dec_model_info_t buffer_model;
285 aom_dec_model_op_parameters_t op_params[MAX_NUM_OPERATING_POINTS + 1];
286 aom_op_timing_info_t op_frame_timing[MAX_NUM_OPERATING_POINTS + 1];
287 int film_grain_test_vector;
288 const char *film_grain_table_filename;
289
290 uint8_t cdf_update_mode;
291 aom_superblock_size_t superblock_size;
292 unsigned int large_scale_tile;
293 unsigned int single_tile_decoding;
294 int monochrome;
295 unsigned int full_still_picture_hdr;
296 int enable_dual_filter;
297 unsigned int motion_vector_unit_test;
298 const cfg_options_t *cfg;
299 int enable_order_hint;
300 int enable_jnt_comp;
301 int enable_ref_frame_mvs;
302 unsigned int allow_ref_frame_mvs;
303 int enable_warped_motion;
304 int allow_warped_motion;
305 int enable_superres;
306 unsigned int save_as_annexb;
307
308 #if CONFIG_DENOISE
309 float noise_level;
310 int noise_block_size;
311 #endif
312
313 unsigned int chroma_subsampling_x;
314 unsigned int chroma_subsampling_y;
315 } AV1EncoderConfig;
316
is_lossless_requested(const AV1EncoderConfig * cfg)317 static INLINE int is_lossless_requested(const AV1EncoderConfig *cfg) {
318 return cfg->best_allowed_q == 0 && cfg->worst_allowed_q == 0;
319 }
320
321 typedef struct FRAME_COUNTS {
322 // Note: This structure should only contain 'unsigned int' fields, or
323 // aggregates built solely from 'unsigned int' fields/elements
324 #if CONFIG_ENTROPY_STATS
325 unsigned int kf_y_mode[KF_MODE_CONTEXTS][KF_MODE_CONTEXTS][INTRA_MODES];
326 unsigned int angle_delta[DIRECTIONAL_MODES][2 * MAX_ANGLE_DELTA + 1];
327 unsigned int y_mode[BLOCK_SIZE_GROUPS][INTRA_MODES];
328 unsigned int uv_mode[CFL_ALLOWED_TYPES][INTRA_MODES][UV_INTRA_MODES];
329 unsigned int cfl_sign[CFL_JOINT_SIGNS];
330 unsigned int cfl_alpha[CFL_ALPHA_CONTEXTS][CFL_ALPHABET_SIZE];
331 unsigned int palette_y_mode[PALATTE_BSIZE_CTXS][PALETTE_Y_MODE_CONTEXTS][2];
332 unsigned int palette_uv_mode[PALETTE_UV_MODE_CONTEXTS][2];
333 unsigned int palette_y_size[PALATTE_BSIZE_CTXS][PALETTE_SIZES];
334 unsigned int palette_uv_size[PALATTE_BSIZE_CTXS][PALETTE_SIZES];
335 unsigned int palette_y_color_index[PALETTE_SIZES]
336 [PALETTE_COLOR_INDEX_CONTEXTS]
337 [PALETTE_COLORS];
338 unsigned int palette_uv_color_index[PALETTE_SIZES]
339 [PALETTE_COLOR_INDEX_CONTEXTS]
340 [PALETTE_COLORS];
341 unsigned int partition[PARTITION_CONTEXTS][EXT_PARTITION_TYPES];
342 unsigned int txb_skip[TOKEN_CDF_Q_CTXS][TX_SIZES][TXB_SKIP_CONTEXTS][2];
343 unsigned int eob_extra[TOKEN_CDF_Q_CTXS][TX_SIZES][PLANE_TYPES]
344 [EOB_COEF_CONTEXTS][2];
345 unsigned int dc_sign[PLANE_TYPES][DC_SIGN_CONTEXTS][2];
346 unsigned int coeff_lps[TX_SIZES][PLANE_TYPES][BR_CDF_SIZE - 1][LEVEL_CONTEXTS]
347 [2];
348 unsigned int eob_flag[TX_SIZES][PLANE_TYPES][EOB_COEF_CONTEXTS][2];
349 unsigned int eob_multi16[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][5];
350 unsigned int eob_multi32[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][6];
351 unsigned int eob_multi64[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][7];
352 unsigned int eob_multi128[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][8];
353 unsigned int eob_multi256[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][9];
354 unsigned int eob_multi512[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][10];
355 unsigned int eob_multi1024[TOKEN_CDF_Q_CTXS][PLANE_TYPES][2][11];
356 unsigned int coeff_lps_multi[TOKEN_CDF_Q_CTXS][TX_SIZES][PLANE_TYPES]
357 [LEVEL_CONTEXTS][BR_CDF_SIZE];
358 unsigned int coeff_base_multi[TOKEN_CDF_Q_CTXS][TX_SIZES][PLANE_TYPES]
359 [SIG_COEF_CONTEXTS][NUM_BASE_LEVELS + 2];
360 unsigned int coeff_base_eob_multi[TOKEN_CDF_Q_CTXS][TX_SIZES][PLANE_TYPES]
361 [SIG_COEF_CONTEXTS_EOB][NUM_BASE_LEVELS + 1];
362 unsigned int newmv_mode[NEWMV_MODE_CONTEXTS][2];
363 unsigned int zeromv_mode[GLOBALMV_MODE_CONTEXTS][2];
364 unsigned int refmv_mode[REFMV_MODE_CONTEXTS][2];
365 unsigned int drl_mode[DRL_MODE_CONTEXTS][2];
366 unsigned int inter_compound_mode[INTER_MODE_CONTEXTS][INTER_COMPOUND_MODES];
367 unsigned int wedge_idx[BLOCK_SIZES_ALL][16];
368 unsigned int interintra[BLOCK_SIZE_GROUPS][2];
369 unsigned int interintra_mode[BLOCK_SIZE_GROUPS][INTERINTRA_MODES];
370 unsigned int wedge_interintra[BLOCK_SIZES_ALL][2];
371 unsigned int compound_type[BLOCK_SIZES_ALL][COMPOUND_TYPES - 1];
372 unsigned int motion_mode[BLOCK_SIZES_ALL][MOTION_MODES];
373 unsigned int obmc[BLOCK_SIZES_ALL][2];
374 unsigned int intra_inter[INTRA_INTER_CONTEXTS][2];
375 unsigned int comp_inter[COMP_INTER_CONTEXTS][2];
376 unsigned int comp_ref_type[COMP_REF_TYPE_CONTEXTS][2];
377 unsigned int uni_comp_ref[UNI_COMP_REF_CONTEXTS][UNIDIR_COMP_REFS - 1][2];
378 unsigned int single_ref[REF_CONTEXTS][SINGLE_REFS - 1][2];
379 unsigned int comp_ref[REF_CONTEXTS][FWD_REFS - 1][2];
380 unsigned int comp_bwdref[REF_CONTEXTS][BWD_REFS - 1][2];
381 unsigned int intrabc[2];
382
383 unsigned int txfm_partition[TXFM_PARTITION_CONTEXTS][2];
384 unsigned int intra_tx_size[MAX_TX_CATS][TX_SIZE_CONTEXTS][MAX_TX_DEPTH + 1];
385 unsigned int skip_mode[SKIP_MODE_CONTEXTS][2];
386 unsigned int skip[SKIP_CONTEXTS][2];
387 unsigned int compound_index[COMP_INDEX_CONTEXTS][2];
388 unsigned int comp_group_idx[COMP_GROUP_IDX_CONTEXTS][2];
389 unsigned int delta_q[DELTA_Q_PROBS][2];
390 unsigned int delta_lf_multi[FRAME_LF_COUNT][DELTA_LF_PROBS][2];
391 unsigned int delta_lf[DELTA_LF_PROBS][2];
392
393 unsigned int inter_ext_tx[EXT_TX_SETS_INTER][EXT_TX_SIZES][TX_TYPES];
394 unsigned int intra_ext_tx[EXT_TX_SETS_INTRA][EXT_TX_SIZES][INTRA_MODES]
395 [TX_TYPES];
396 unsigned int filter_intra_mode[FILTER_INTRA_MODES];
397 unsigned int filter_intra[BLOCK_SIZES_ALL][2];
398 unsigned int switchable_restore[RESTORE_SWITCHABLE_TYPES];
399 unsigned int wiener_restore[2];
400 unsigned int sgrproj_restore[2];
401 #endif // CONFIG_ENTROPY_STATS
402
403 unsigned int switchable_interp[SWITCHABLE_FILTER_CONTEXTS]
404 [SWITCHABLE_FILTERS];
405 } FRAME_COUNTS;
406
407 #if CONFIG_COLLECT_INTER_MODE_RD_STATS
408 #define INTER_MODE_RD_DATA_OVERALL_SIZE 6400
409
410 typedef struct {
411 int ready;
412 double a;
413 double b;
414 double dist_mean;
415 double ld_mean;
416 double sse_mean;
417 double sse_sse_mean;
418 double sse_ld_mean;
419 int num;
420 double dist_sum;
421 double ld_sum;
422 double sse_sum;
423 double sse_sse_sum;
424 double sse_ld_sum;
425 } InterModeRdModel;
426
427 typedef struct {
428 int idx;
429 int64_t rd;
430 } RdIdxPair;
431 // TODO(angiebird): This is an estimated size. We still need to figure what is
432 // the maximum number of modes.
433 #define MAX_INTER_MODES 1024
434 typedef struct inter_modes_info {
435 int num;
436 MB_MODE_INFO mbmi_arr[MAX_INTER_MODES];
437 int mode_rate_arr[MAX_INTER_MODES];
438 int64_t sse_arr[MAX_INTER_MODES];
439 int64_t est_rd_arr[MAX_INTER_MODES];
440 RdIdxPair rd_idx_pair_arr[MAX_INTER_MODES];
441 } InterModesInfo;
442 #endif
443
444 // TODO(jingning) All spatially adaptive variables should go to TileDataEnc.
445 typedef struct TileDataEnc {
446 TileInfo tile_info;
447 int thresh_freq_fact[BLOCK_SIZES_ALL][MAX_MODES];
448 int mode_map[BLOCK_SIZES_ALL][MAX_MODES];
449 int m_search_count;
450 int ex_search_count;
451 CFL_CTX cfl;
452 DECLARE_ALIGNED(16, FRAME_CONTEXT, tctx);
453 uint8_t allow_update_cdf;
454 #if CONFIG_COLLECT_INTER_MODE_RD_STATS
455 InterModeRdModel inter_mode_rd_models[BLOCK_SIZES_ALL];
456 InterModesInfo inter_modes_info;
457 #endif
458 } TileDataEnc;
459
460 typedef struct {
461 TOKENEXTRA *start;
462 TOKENEXTRA *stop;
463 unsigned int count;
464 } TOKENLIST;
465
466 typedef struct RD_COUNTS {
467 int64_t comp_pred_diff[REFERENCE_MODES];
468 // Stores number of 4x4 blocks using global motion per reference frame.
469 int global_motion_used[REF_FRAMES];
470 int compound_ref_used_flag;
471 int skip_mode_used_flag;
472 } RD_COUNTS;
473
474 typedef struct ThreadData {
475 MACROBLOCK mb;
476 RD_COUNTS rd_counts;
477 FRAME_COUNTS *counts;
478 PC_TREE *pc_tree;
479 PC_TREE *pc_root[MAX_MIB_SIZE_LOG2 - MIN_MIB_SIZE_LOG2 + 1];
480 uint32_t *hash_value_buffer[2][2];
481 int32_t *wsrc_buf;
482 int32_t *mask_buf;
483 uint8_t *above_pred_buf;
484 uint8_t *left_pred_buf;
485 PALETTE_BUFFER *palette_buffer;
486 CONV_BUF_TYPE *tmp_conv_dst;
487 uint8_t *tmp_obmc_bufs[2];
488 int intrabc_used_this_tile;
489 } ThreadData;
490
491 struct EncWorkerData;
492
493 typedef struct ActiveMap {
494 int enabled;
495 int update;
496 unsigned char *map;
497 } ActiveMap;
498
499 #if CONFIG_INTERNAL_STATS
500 // types of stats
501 typedef enum {
502 STAT_Y,
503 STAT_U,
504 STAT_V,
505 STAT_ALL,
506 NUM_STAT_TYPES // This should always be the last member of the enum
507 } StatType;
508
509 typedef struct IMAGE_STAT {
510 double stat[NUM_STAT_TYPES];
511 double worst;
512 } ImageStat;
513 #endif // CONFIG_INTERNAL_STATS
514
515 typedef struct {
516 int ref_count;
517 YV12_BUFFER_CONFIG buf;
518 } EncRefCntBuffer;
519
520 typedef struct TileBufferEnc {
521 uint8_t *data;
522 size_t size;
523 } TileBufferEnc;
524
525 typedef struct AV1_COMP {
526 QUANTS quants;
527 ThreadData td;
528 FRAME_COUNTS counts;
529 MB_MODE_INFO_EXT *mbmi_ext_base;
530 CB_COEFF_BUFFER *coeff_buffer_base;
531 Dequants dequants;
532 AV1_COMMON common;
533 AV1EncoderConfig oxcf;
534 struct lookahead_ctx *lookahead;
535 struct lookahead_entry *alt_ref_source;
536 int no_show_kf;
537
538 int optimize_speed_feature;
539 int optimize_seg_arr[MAX_SEGMENTS];
540
541 YV12_BUFFER_CONFIG *source;
542 YV12_BUFFER_CONFIG *last_source; // NULL for first frame and alt_ref frames
543 YV12_BUFFER_CONFIG *unscaled_source;
544 YV12_BUFFER_CONFIG scaled_source;
545 YV12_BUFFER_CONFIG *unscaled_last_source;
546 YV12_BUFFER_CONFIG scaled_last_source;
547
548 // For a still frame, this flag is set to 1 to skip partition search.
549 int partition_search_skippable_frame;
550 double csm_rate_array[32];
551 double m_rate_array[32];
552 int rate_size;
553 int rate_index;
554 hash_table *previous_hash_table;
555 int previous_index;
556 int cur_poc; // DebugInfo
557
558 unsigned int row_mt;
559 int scaled_ref_idx[REF_FRAMES];
560 int ref_fb_idx[REF_FRAMES];
561 int refresh_fb_idx; // ref frame buffer index to refresh
562
563 int last_show_frame_buf_idx; // last show frame buffer index
564
565 int refresh_last_frame;
566 int refresh_golden_frame;
567 int refresh_bwd_ref_frame;
568 int refresh_alt2_ref_frame;
569 int refresh_alt_ref_frame;
570 #if USE_SYMM_MULTI_LAYER
571 int new_bwdref_update_rule;
572 #endif
573
574 int ext_refresh_frame_flags_pending;
575 int ext_refresh_last_frame;
576 int ext_refresh_golden_frame;
577 int ext_refresh_bwd_ref_frame;
578 int ext_refresh_alt2_ref_frame;
579 int ext_refresh_alt_ref_frame;
580
581 int ext_refresh_frame_context_pending;
582 int ext_refresh_frame_context;
583 int ext_use_ref_frame_mvs;
584 int ext_use_error_resilient;
585 int ext_use_s_frame;
586 int ext_use_primary_ref_none;
587
588 YV12_BUFFER_CONFIG last_frame_uf;
589 YV12_BUFFER_CONFIG trial_frame_rst;
590
591 // Ambient reconstruction err target for force key frames
592 int64_t ambient_err;
593
594 RD_OPT rd;
595
596 CODING_CONTEXT coding_context;
597
598 int gmtype_cost[TRANS_TYPES];
599 int gmparams_cost[REF_FRAMES];
600
601 int nmv_costs[2][MV_VALS];
602 int nmv_costs_hp[2][MV_VALS];
603
604 int64_t last_time_stamp_seen;
605 int64_t last_end_time_stamp_seen;
606 int64_t first_time_stamp_ever;
607
608 RATE_CONTROL rc;
609 double framerate;
610
611 // NOTE(zoeliu): Any inter frame allows maximum of REF_FRAMES inter
612 // references; Plus the currently coded frame itself, it is needed to allocate
613 // sufficient space to the size of the maximum possible number of frames.
614 int interp_filter_selected[REF_FRAMES + 1][SWITCHABLE];
615
616 struct aom_codec_pkt_list *output_pkt_list;
617
618 MBGRAPH_FRAME_STATS mbgraph_stats[MAX_LAG_BUFFERS];
619 int mbgraph_n_frames; // number of frames filled in the above
620 int static_mb_pct; // % forced skip mbs by segmentation
621 int ref_frame_flags;
622 int ext_ref_frame_flags;
623 RATE_FACTOR_LEVEL frame_rf_level[FRAME_BUFFERS];
624
625 SPEED_FEATURES sf;
626
627 unsigned int max_mv_magnitude;
628 int mv_step_param;
629
630 int allow_comp_inter_inter;
631 int all_one_sided_refs;
632
633 uint8_t *segmentation_map;
634
635 CYCLIC_REFRESH *cyclic_refresh;
636 ActiveMap active_map;
637
638 fractional_mv_step_fp *find_fractional_mv_step;
639 av1_diamond_search_fn_t diamond_search_sad;
640 aom_variance_fn_ptr_t fn_ptr[BLOCK_SIZES_ALL];
641 uint64_t time_receive_data;
642 uint64_t time_compress_data;
643 uint64_t time_pick_lpf;
644 uint64_t time_encode_sb_row;
645
646 #if CONFIG_FP_MB_STATS
647 int use_fp_mb_stats;
648 #endif
649
650 TWO_PASS twopass;
651
652 YV12_BUFFER_CONFIG alt_ref_buffer;
653
654 #if CONFIG_INTERNAL_STATS
655 unsigned int mode_chosen_counts[MAX_MODES];
656
657 int count;
658 uint64_t total_sq_error;
659 uint64_t total_samples;
660 ImageStat psnr;
661
662 double total_blockiness;
663 double worst_blockiness;
664
665 int bytes;
666 double summed_quality;
667 double summed_weights;
668 unsigned int tot_recode_hits;
669 double worst_ssim;
670
671 ImageStat fastssim;
672 ImageStat psnrhvs;
673
674 int b_calculate_blockiness;
675 int b_calculate_consistency;
676
677 double total_inconsistency;
678 double worst_consistency;
679 Ssimv *ssim_vars;
680 Metrics metrics;
681 #endif
682 int b_calculate_psnr;
683
684 int droppable;
685
686 int initial_width;
687 int initial_height;
688 int initial_mbs; // Number of MBs in the full-size frame; to be used to
689 // normalize the firstpass stats. This will differ from the
690 // number of MBs in the current frame when the frame is
691 // scaled.
692
693 // When resize is triggered through external control, the desired width/height
694 // are stored here until use in the next frame coded. They are effective only
695 // for
696 // one frame and are reset after use.
697 int resize_pending_width;
698 int resize_pending_height;
699
700 int frame_flags;
701
702 search_site_config ss_cfg;
703
704 TileDataEnc *tile_data;
705 int allocated_tiles; // Keep track of memory allocated for tiles.
706
707 TOKENEXTRA *tile_tok[MAX_TILE_ROWS][MAX_TILE_COLS];
708 unsigned int tok_count[MAX_TILE_ROWS][MAX_TILE_COLS];
709 TOKENLIST *tplist[MAX_TILE_ROWS][MAX_TILE_COLS];
710
711 TileBufferEnc tile_buffers[MAX_TILE_ROWS][MAX_TILE_COLS];
712
713 int resize_state;
714 int resize_avg_qp;
715 int resize_buffer_underflow;
716 int resize_count;
717
718 // Sequence parameters have been transmitted already and locked
719 // or not. Once locked av1_change_config cannot change the seq
720 // parameters.
721 int seq_params_locked;
722
723 // VARIANCE_AQ segment map refresh
724 int vaq_refresh;
725
726 // Multi-threading
727 int num_workers;
728 AVxWorker *workers;
729 struct EncWorkerData *tile_thr_data;
730 int refresh_frame_mask;
731 int existing_fb_idx_to_show;
732 int is_arf_filter_off[MAX_EXT_ARFS + 1];
733 int num_extra_arfs;
734 int arf_pos_in_gf[MAX_EXT_ARFS + 1];
735 int arf_pos_for_ovrly[MAX_EXT_ARFS + 1];
736 int global_motion_search_done;
737 tran_low_t *tcoeff_buf[MAX_MB_PLANE];
738 int extra_arf_allowed;
739 // A flag to indicate if intrabc is ever used in current frame.
740 int intrabc_used;
741 int dv_cost[2][MV_VALS];
742 // TODO(huisu@google.com): we can update dv_joint_cost per SB.
743 int dv_joint_cost[MV_JOINTS];
744 int has_lossless_segment;
745
746 // For frame refs short signaling:
747 // A mapping of each reference frame from its encoder side value to the
748 // decoder side value obtained following the short signaling procedure.
749 int ref_conv[REF_FRAMES];
750
751 AV1LfSync lf_row_sync;
752 AV1LrSync lr_row_sync;
753 AV1LrStruct lr_ctxt;
754
755 aom_film_grain_table_t *film_grain_table;
756 #if CONFIG_DENOISE
757 struct aom_denoise_and_model_t *denoise_and_model;
758 #endif
759 // Stores the default value of skip flag depending on chroma format
760 // Set as 1 for monochrome and 3 for other color formats
761 int default_interp_skip_flags;
762 int preserve_arf_as_gld;
763 } AV1_COMP;
764
765 // Must not be called more than once.
766 void av1_initialize_enc(void);
767
768 struct AV1_COMP *av1_create_compressor(AV1EncoderConfig *oxcf,
769 BufferPool *const pool);
770 void av1_remove_compressor(AV1_COMP *cpi);
771
772 void av1_change_config(AV1_COMP *cpi, const AV1EncoderConfig *oxcf);
773
774 // receive a frames worth of data. caller can assume that a copy of this
775 // frame is made and not just a copy of the pointer..
776 int av1_receive_raw_frame(AV1_COMP *cpi, aom_enc_frame_flags_t frame_flags,
777 YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
778 int64_t end_time_stamp);
779
780 int av1_get_compressed_data(AV1_COMP *cpi, unsigned int *frame_flags,
781 size_t *size, uint8_t *dest, int64_t *time_stamp,
782 int64_t *time_end, int flush,
783 const aom_rational_t *timebase);
784
785 int av1_get_preview_raw_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *dest);
786
787 int av1_get_last_show_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *frame);
788
789 aom_codec_err_t av1_copy_new_frame_enc(AV1_COMMON *cm,
790 YV12_BUFFER_CONFIG *new_frame,
791 YV12_BUFFER_CONFIG *sd);
792
793 int av1_use_as_reference(AV1_COMP *cpi, int ref_frame_flags);
794
795 void av1_update_reference(AV1_COMP *cpi, int ref_frame_flags);
796
797 int av1_copy_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd);
798
799 int av1_set_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd);
800
801 int av1_update_entropy(AV1_COMP *cpi, int update);
802
803 int av1_set_active_map(AV1_COMP *cpi, unsigned char *map, int rows, int cols);
804
805 int av1_get_active_map(AV1_COMP *cpi, unsigned char *map, int rows, int cols);
806
807 int av1_set_internal_size(AV1_COMP *cpi, AOM_SCALING horiz_mode,
808 AOM_SCALING vert_mode);
809
810 int av1_get_quantizer(struct AV1_COMP *cpi);
811
812 int av1_convert_sect5obus_to_annexb(uint8_t *buffer, size_t *input_size);
813
814 int64_t timebase_units_to_ticks(const aom_rational_t *timebase, int64_t n);
815 int64_t ticks_to_timebase_units(const aom_rational_t *timebase, int64_t n);
816
frame_is_kf_gf_arf(const AV1_COMP * cpi)817 static INLINE int frame_is_kf_gf_arf(const AV1_COMP *cpi) {
818 return frame_is_intra_only(&cpi->common) || cpi->refresh_alt_ref_frame ||
819 (cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref);
820 }
821
get_ref_frame_map_idx(const AV1_COMP * cpi,MV_REFERENCE_FRAME ref_frame)822 static INLINE int get_ref_frame_map_idx(const AV1_COMP *cpi,
823 MV_REFERENCE_FRAME ref_frame) {
824 return (ref_frame >= 1) ? cpi->ref_fb_idx[ref_frame - 1] : INVALID_IDX;
825 }
826
get_ref_frame_buf_idx(const AV1_COMP * cpi,MV_REFERENCE_FRAME ref_frame)827 static INLINE int get_ref_frame_buf_idx(const AV1_COMP *cpi,
828 MV_REFERENCE_FRAME ref_frame) {
829 const AV1_COMMON *const cm = &cpi->common;
830 const int map_idx = get_ref_frame_map_idx(cpi, ref_frame);
831 return (map_idx != INVALID_IDX) ? cm->ref_frame_map[map_idx] : INVALID_IDX;
832 }
833
834 // TODO(huisu@google.com, youzhou@microsoft.com): enable hash-me for HBD.
av1_use_hash_me(const AV1_COMMON * const cm)835 static INLINE int av1_use_hash_me(const AV1_COMMON *const cm) {
836 return cm->allow_screen_content_tools;
837 }
838
av1_get_ref_frame_hash_map(const AV1_COMP * cpi,MV_REFERENCE_FRAME ref_frame)839 static INLINE hash_table *av1_get_ref_frame_hash_map(
840 const AV1_COMP *cpi, MV_REFERENCE_FRAME ref_frame) {
841 const AV1_COMMON *const cm = &cpi->common;
842 const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
843 return buf_idx != INVALID_IDX
844 ? &cm->buffer_pool->frame_bufs[buf_idx].hash_table
845 : NULL;
846 }
847
get_ref_frame_buffer(const AV1_COMP * cpi,MV_REFERENCE_FRAME ref_frame)848 static INLINE YV12_BUFFER_CONFIG *get_ref_frame_buffer(
849 const AV1_COMP *cpi, MV_REFERENCE_FRAME ref_frame) {
850 const AV1_COMMON *const cm = &cpi->common;
851 const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
852 return buf_idx != INVALID_IDX ? &cm->buffer_pool->frame_bufs[buf_idx].buf
853 : NULL;
854 }
855
enc_is_ref_frame_buf(AV1_COMP * cpi,RefCntBuffer * frame_buf)856 static INLINE int enc_is_ref_frame_buf(AV1_COMP *cpi, RefCntBuffer *frame_buf) {
857 MV_REFERENCE_FRAME ref_frame;
858 AV1_COMMON *const cm = &cpi->common;
859 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
860 const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
861 if (buf_idx == INVALID_IDX) continue;
862 if (frame_buf == &cm->buffer_pool->frame_bufs[buf_idx]) break;
863 }
864 return (ref_frame <= ALTREF_FRAME);
865 }
866
867 // Token buffer is only used for palette tokens.
get_token_alloc(int mb_rows,int mb_cols,int sb_size_log2,const int num_planes)868 static INLINE unsigned int get_token_alloc(int mb_rows, int mb_cols,
869 int sb_size_log2,
870 const int num_planes) {
871 // Calculate the maximum number of max superblocks in the image.
872 const int shift = sb_size_log2 - 4;
873 const int sb_size = 1 << sb_size_log2;
874 const int sb_size_square = sb_size * sb_size;
875 const int sb_rows = ALIGN_POWER_OF_TWO(mb_rows, shift) >> shift;
876 const int sb_cols = ALIGN_POWER_OF_TWO(mb_cols, shift) >> shift;
877
878 // One palette token for each pixel. There can be palettes on two planes.
879 const int sb_palette_toks = AOMMIN(2, num_planes) * sb_size_square;
880
881 return sb_rows * sb_cols * sb_palette_toks;
882 }
883
884 // Get the allocated token size for a tile. It does the same calculation as in
885 // the frame token allocation.
allocated_tokens(TileInfo tile,int sb_size_log2,int num_planes)886 static INLINE unsigned int allocated_tokens(TileInfo tile, int sb_size_log2,
887 int num_planes) {
888 int tile_mb_rows = (tile.mi_row_end - tile.mi_row_start + 2) >> 2;
889 int tile_mb_cols = (tile.mi_col_end - tile.mi_col_start + 2) >> 2;
890
891 return get_token_alloc(tile_mb_rows, tile_mb_cols, sb_size_log2, num_planes);
892 }
893
get_start_tok(AV1_COMP * cpi,int tile_row,int tile_col,int mi_row,TOKENEXTRA ** tok,int sb_size_log2,int num_planes)894 static INLINE void get_start_tok(AV1_COMP *cpi, int tile_row, int tile_col,
895 int mi_row, TOKENEXTRA **tok, int sb_size_log2,
896 int num_planes) {
897 AV1_COMMON *const cm = &cpi->common;
898 const int tile_cols = cm->tile_cols;
899 TileDataEnc *this_tile = &cpi->tile_data[tile_row * tile_cols + tile_col];
900 const TileInfo *const tile_info = &this_tile->tile_info;
901
902 const int tile_mb_cols =
903 (tile_info->mi_col_end - tile_info->mi_col_start + 2) >> 2;
904 const int tile_mb_row = (mi_row - tile_info->mi_row_start + 2) >> 2;
905
906 *tok = cpi->tile_tok[tile_row][tile_col] +
907 get_token_alloc(tile_mb_row, tile_mb_cols, sb_size_log2, num_planes);
908 }
909
910 void av1_apply_encoding_flags(AV1_COMP *cpi, aom_enc_frame_flags_t flags);
911
912 #define ALT_MIN_LAG 3
is_altref_enabled(const AV1_COMP * const cpi)913 static INLINE int is_altref_enabled(const AV1_COMP *const cpi) {
914 return cpi->oxcf.lag_in_frames >= ALT_MIN_LAG && cpi->oxcf.enable_auto_arf;
915 }
916
917 // TODO(zoeliu): To set up cpi->oxcf.enable_auto_brf
918
set_ref_ptrs(const AV1_COMMON * cm,MACROBLOCKD * xd,MV_REFERENCE_FRAME ref0,MV_REFERENCE_FRAME ref1)919 static INLINE void set_ref_ptrs(const AV1_COMMON *cm, MACROBLOCKD *xd,
920 MV_REFERENCE_FRAME ref0,
921 MV_REFERENCE_FRAME ref1) {
922 xd->block_refs[0] =
923 &cm->frame_refs[ref0 >= LAST_FRAME ? ref0 - LAST_FRAME : 0];
924 xd->block_refs[1] =
925 &cm->frame_refs[ref1 >= LAST_FRAME ? ref1 - LAST_FRAME : 0];
926 }
927
get_chessboard_index(int frame_index)928 static INLINE int get_chessboard_index(int frame_index) {
929 return frame_index & 0x1;
930 }
931
cond_cost_list(const struct AV1_COMP * cpi,int * cost_list)932 static INLINE int *cond_cost_list(const struct AV1_COMP *cpi, int *cost_list) {
933 return cpi->sf.mv.subpel_search_method != SUBPEL_TREE ? cost_list : NULL;
934 }
935
936 void av1_new_framerate(AV1_COMP *cpi, double framerate);
937
938 #define LAYER_IDS_TO_IDX(sl, tl, num_tl) ((sl) * (num_tl) + (tl))
939
940 // Update up-sampled reference frame index.
uref_cnt_fb(EncRefCntBuffer * ubufs,int * uidx,int new_uidx)941 static INLINE void uref_cnt_fb(EncRefCntBuffer *ubufs, int *uidx,
942 int new_uidx) {
943 const int ref_index = *uidx;
944
945 if (ref_index >= 0 && ubufs[ref_index].ref_count > 0)
946 ubufs[ref_index].ref_count--;
947
948 *uidx = new_uidx;
949 ubufs[new_uidx].ref_count++;
950 }
951
952 // Returns 1 if a frame is scaled and 0 otherwise.
av1_resize_scaled(const AV1_COMMON * cm)953 static INLINE int av1_resize_scaled(const AV1_COMMON *cm) {
954 return !(cm->superres_upscaled_width == cm->render_width &&
955 cm->superres_upscaled_height == cm->render_height);
956 }
957
av1_frame_scaled(const AV1_COMMON * cm)958 static INLINE int av1_frame_scaled(const AV1_COMMON *cm) {
959 return !av1_superres_scaled(cm) && av1_resize_scaled(cm);
960 }
961
962 // Don't allow a show_existing_frame to coincide with an error resilient
963 // frame. An exception can be made for a forward keyframe since it has no
964 // previous dependencies.
encode_show_existing_frame(const AV1_COMMON * cm)965 static INLINE int encode_show_existing_frame(const AV1_COMMON *cm) {
966 return cm->show_existing_frame &&
967 (!cm->error_resilient_mode || cm->frame_type == KEY_FRAME);
968 }
969
970 // Returns a Sequence Header OBU stored in an aom_fixed_buf_t, or NULL upon
971 // failure. When a non-NULL aom_fixed_buf_t pointer is returned by this
972 // function, the memory must be freed by the caller. Both the buf member of the
973 // aom_fixed_buf_t, and the aom_fixed_buf_t pointer itself must be freed. Memory
974 // returned must be freed via call to free().
975 //
976 // Note: The OBU returned is in Low Overhead Bitstream Format. Specifically,
977 // the obu_has_size_field bit is set, and the buffer contains the obu_size
978 // field.
979 aom_fixed_buf_t *av1_get_global_headers(AV1_COMP *cpi);
980
981 #ifdef __cplusplus
982 } // extern "C"
983 #endif
984
985 #endif // AOM_AV1_ENCODER_ENCODER_H_
986