2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
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.
11 #ifndef VP9_ENCODER_VP9_ENCODER_H_
12 #define VP9_ENCODER_VP9_ENCODER_H_
16 #include "./vpx_config.h"
17 #include "vpx_ports/mem.h"
18 #include "vpx/internal/vpx_codec_internal.h"
19 #include "vpx/vp8cx.h"
21 #include "vp9/common/vp9_ppflags.h"
22 #include "vp9/common/vp9_entropy.h"
23 #include "vp9/common/vp9_entropymode.h"
24 #include "vp9/common/vp9_onyxc_int.h"
26 #include "vp9/encoder/vp9_aq_cyclicrefresh.h"
27 #include "vp9/encoder/vp9_encodemb.h"
28 #include "vp9/encoder/vp9_firstpass.h"
29 #include "vp9/encoder/vp9_lookahead.h"
30 #include "vp9/encoder/vp9_mbgraph.h"
31 #include "vp9/encoder/vp9_mcomp.h"
32 #include "vp9/encoder/vp9_quantize.h"
33 #include "vp9/encoder/vp9_ratectrl.h"
34 #include "vp9/encoder/vp9_speed_features.h"
35 #include "vp9/encoder/vp9_svc_layercontext.h"
36 #include "vp9/encoder/vp9_tokenize.h"
37 #include "vp9/encoder/vp9_variance.h"
43 // #define MODE_TEST_HIT_STATS
45 #define DEFAULT_GF_INTERVAL 10
51 int nmvjointcost[MV_JOINTS];
52 int nmvcosts[2][MV_VALS];
53 int nmvcosts_hp[2][MV_VALS];
55 vp9_prob segment_pred_probs[PREDICTION_PROBS];
57 unsigned char *last_frame_seg_map_copy;
59 // 0 = Intra, Last, GF, ARF
60 signed char last_ref_lf_deltas[MAX_REF_LF_DELTAS];
62 signed char last_mode_lf_deltas[MAX_MODE_LF_DELTAS];
67 // This enumerator type needs to be kept aligned with the mode order in
68 // const MODE_DEFINITION vp9_mode_order[MAX_MODES] used in the rd code.
119 // encode_breakout is disabled.
120 ENCODE_BREAKOUT_DISABLED = 0,
121 // encode_breakout is enabled.
122 ENCODE_BREAKOUT_ENABLED = 1,
123 // encode_breakout is enabled with small max_thresh limit.
124 ENCODE_BREAKOUT_LIMITED = 2
125 } ENCODE_BREAKOUT_TYPE;
137 RC_MODE_CONSTRAINED_QUALITY = 2,
138 RC_MODE_CONSTANT_QUALITY = 3,
142 // Good Quality Fast Encoding. The encoder balances quality with the
143 // amount of time it takes to encode the output. (speed setting
144 // controls how fast)
147 // One Pass - Best Quality. The encoder places priority on the
148 // quality of the output over encoding speed. The output is compressed
149 // at the highest possible quality. This option takes the longest
150 // amount of time to encode. (speed setting ignored)
153 // Two Pass - First Pass. The encoder generates a file of statistics
154 // for use in the second encoding pass. (speed setting controls how fast)
157 // Two Pass - Second Pass. The encoder uses the statistics that were
158 // generated in the first encoding pass to create the compressed
159 // output. (speed setting controls how fast)
160 TWO_PASS_SECOND_GOOD = 4,
162 // Two Pass - Second Pass Best. The encoder uses the statistics that
163 // were generated in the first encoding pass to create the compressed
164 // output using the highest possible quality, and taking a
165 // longer amount of time to encode. (speed setting ignored)
166 TWO_PASS_SECOND_BEST = 5,
168 // Realtime/Live Encoding. This mode is optimized for realtime
169 // encoding (for example, capturing a television signal or feed from
170 // a live camera). (speed setting controls how fast)
175 FRAMEFLAGS_KEY = 1 << 0,
176 FRAMEFLAGS_GOLDEN = 1 << 1,
177 FRAMEFLAGS_ALTREF = 1 << 2,
184 CYCLIC_REFRESH_AQ = 3,
185 AQ_MODE_COUNT // This should always be the last member of the enum
189 typedef struct VP9EncoderConfig {
190 BITSTREAM_PROFILE profile;
192 int width; // width of data passed to the compressor
193 int height; // height of data passed to the compressor
194 double framerate; // set to passed in framerate
195 int64_t target_bandwidth; // bandwidth to be used in kilobits per second
197 int noise_sensitivity; // pre processing blur: recommendation 0
198 int sharpness; // sharpening output: recommendation 0:
200 unsigned int rc_max_intra_bitrate_pct;
204 // Key Framing Operations
205 int auto_key; // autodetect cut scenes and set the keyframes
206 int key_freq; // maximum distance to key frame.
208 int lag_in_frames; // how many frames lag before we start encoding
210 // ----------------------------------------------------------------
211 // DATARATE CONTROL OPTIONS
213 RC_MODE rc_mode; // vbr, cbr, constrained quality or constant quality
215 // buffer targeting aggressiveness
219 // buffering parameters
220 int64_t starting_buffer_level; // in seconds
221 int64_t optimal_buffer_level;
222 int64_t maximum_buffer_size;
224 // Frame drop threshold.
225 int drop_frames_water_mark;
227 // controlling quality
233 AQ_MODE aq_mode; // Adaptive Quantization mode
235 // Internal frame size scaling.
236 int allow_spatial_resampling;
237 int scaled_frame_width;
238 int scaled_frame_height;
240 // Enable feature to reduce the frame quantization every x frames.
241 int frame_periodic_boost;
243 // two pass datarate control
244 int two_pass_vbrbias; // two pass datarate control tweaks
245 int two_pass_vbrmin_section;
246 int two_pass_vbrmax_section;
247 // END DATARATE CONTROL OPTIONS
248 // ----------------------------------------------------------------
250 // Spatial and temporal scalability.
251 int ss_number_layers; // Number of spatial layers.
252 int ts_number_layers; // Number of temporal layers.
253 // Bitrate allocation for spatial layers.
254 int ss_target_bitrate[VPX_SS_MAX_LAYERS];
255 // Bitrate allocation (CBR mode) and framerate factor, for temporal layers.
256 int ts_target_bitrate[VPX_TS_MAX_LAYERS];
257 int ts_rate_decimator[VPX_TS_MAX_LAYERS];
259 // these parameters aren't to be used in final build don't use!!!
263 int encode_breakout; // early breakout : for video conf recommend 800
265 /* Bitfield defining the error resiliency features to enable.
266 * Can provide decodable frames after losses in previous
267 * frames and decodable partitions after losses in the same frame.
269 unsigned int error_resilient_mode;
271 /* Bitfield defining the parallel decoding mode where the
272 * decoding in successive frames may be conducted in parallel
273 * just by decoding the frame headers.
275 unsigned int frame_parallel_decoding_mode;
284 struct vpx_fixed_buf two_pass_stats_in;
285 struct vpx_codec_pkt_list *output_pkt_list;
290 static INLINE int is_best_mode(MODE mode) {
291 return mode == ONE_PASS_BEST || mode == TWO_PASS_SECOND_BEST;
294 typedef struct RD_OPT {
295 // Thresh_mult is used to set a threshold for the rd score. A higher value
296 // means that we will accept the best mode so far more often. This number
297 // is used in combination with the current block size, and thresh_freq_fact
298 // to pick a threshold.
299 int thresh_mult[MAX_MODES];
300 int thresh_mult_sub8x8[MAX_REFS];
302 int threshes[MAX_SEGMENTS][BLOCK_SIZES][MAX_MODES];
303 int thresh_freq_fact[BLOCK_SIZES][MAX_MODES];
305 int64_t comp_pred_diff[REFERENCE_MODES];
306 int64_t prediction_type_threshes[MAX_REF_FRAMES][REFERENCE_MODES];
307 int64_t tx_select_diff[TX_MODES];
308 // FIXME(rbultje) can this overflow?
309 int tx_select_threshes[MAX_REF_FRAMES][TX_MODES];
311 int64_t filter_diff[SWITCHABLE_FILTER_CONTEXTS];
312 int64_t filter_threshes[MAX_REF_FRAMES][SWITCHABLE_FILTER_CONTEXTS];
313 int64_t filter_cache[SWITCHABLE_FILTER_CONTEXTS];
320 typedef struct VP9_COMP {
324 VP9EncoderConfig oxcf;
325 struct lookahead_ctx *lookahead;
326 struct lookahead_entry *source;
327 #if CONFIG_MULTIPLE_ARF
328 struct lookahead_entry *alt_ref_source[REF_FRAMES];
330 struct lookahead_entry *alt_ref_source;
332 struct lookahead_entry *last_source;
334 YV12_BUFFER_CONFIG *Source;
335 YV12_BUFFER_CONFIG *Last_Source; // NULL for first frame and alt_ref frames
336 YV12_BUFFER_CONFIG *un_scaled_source;
337 YV12_BUFFER_CONFIG scaled_source;
338 YV12_BUFFER_CONFIG *unscaled_last_source;
339 YV12_BUFFER_CONFIG scaled_last_source;
341 int gold_is_last; // gold same as last frame ( short circuit gold searches)
342 int alt_is_last; // Alt same as last ( short circuit altref search)
343 int gold_is_alt; // don't do both alt and gold search ( just do gold).
345 int scaled_ref_idx[3];
350 #if CONFIG_MULTIPLE_ARF
351 int alt_ref_fb_idx[REF_FRAMES - 3];
353 int refresh_last_frame;
354 int refresh_golden_frame;
355 int refresh_alt_ref_frame;
357 int ext_refresh_frame_flags_pending;
358 int ext_refresh_last_frame;
359 int ext_refresh_golden_frame;
360 int ext_refresh_alt_ref_frame;
362 int ext_refresh_frame_context_pending;
363 int ext_refresh_frame_context;
365 YV12_BUFFER_CONFIG last_frame_uf;
368 unsigned int tok_count[4][1 << 6];
370 #if CONFIG_MULTIPLE_ARF
371 // Position within a frame coding order (including any additional ARF frames).
372 unsigned int sequence_number;
373 // Next frame in naturally occurring order that has not yet been coded.
374 int next_frame_in_order;
377 // Ambient reconstruction err target for force key frames
382 CODING_CONTEXT coding_context;
385 int zbin_mode_boost_enabled;
386 int active_arnr_frames; // <= cpi->oxcf.arnr_max_frames
387 int active_arnr_strength; // <= cpi->oxcf.arnr_max_strength
389 int64_t last_time_stamp_seen;
390 int64_t last_end_time_stamp_seen;
391 int64_t first_time_stamp_ever;
395 vp9_coeff_count coef_counts[TX_SIZES][PLANE_TYPES];
396 vp9_coeff_probs_model frame_coef_probs[TX_SIZES][PLANE_TYPES];
398 struct vpx_codec_pkt_list *output_pkt_list;
400 MBGRAPH_FRAME_STATS mbgraph_stats[MAX_LAG_BUFFERS];
401 int mbgraph_n_frames; // number of frames filled in the above
402 int static_mb_pct; // % forced skip mbs by segmentation
410 unsigned int max_mv_magnitude;
413 // Default value is 1. From first pass stats, encode_breakout may be disabled.
414 ENCODE_BREAKOUT_TYPE allow_encode_breakout;
416 // Get threshold from external input. In real time mode, it can be
417 // overwritten according to encoding speed.
420 unsigned char *segmentation_map;
422 // segment threashold for encode breakout
423 int segment_encode_breakout[MAX_SEGMENTS];
425 unsigned char *complexity_map;
427 unsigned char *active_map;
428 unsigned int active_map_enabled;
430 CYCLIC_REFRESH *cyclic_refresh;
432 fractional_mv_step_fp *find_fractional_mv_step;
433 fractional_mv_step_comp_fp *find_fractional_mv_step_comp;
434 vp9_full_search_fn_t full_search_sad;
435 vp9_refining_search_fn_t refining_search_sad;
436 vp9_diamond_search_fn_t diamond_search_sad;
437 vp9_variance_fn_ptr_t fn_ptr[BLOCK_SIZES];
438 uint64_t time_receive_data;
439 uint64_t time_compress_data;
440 uint64_t time_pick_lpf;
441 uint64_t time_encode_sb_row;
443 struct twopass_rc twopass;
445 YV12_BUFFER_CONFIG alt_ref_buffer;
446 YV12_BUFFER_CONFIG *frames[MAX_LAG_BUFFERS];
448 #if CONFIG_INTERNAL_STATS
449 unsigned int mode_chosen_counts[MAX_MODES];
456 uint64_t total_sq_error;
457 uint64_t total_samples;
463 uint64_t totalp_sq_error;
464 uint64_t totalp_samples;
467 double summed_quality;
468 double summed_weights;
469 double summedp_quality;
470 double summedp_weights;
471 unsigned int tot_recode_hits;
474 double total_ssimg_y;
475 double total_ssimg_u;
476 double total_ssimg_v;
477 double total_ssimg_all;
479 int b_calculate_ssimg;
481 int b_calculate_psnr;
485 int dummy_packing; /* flag to indicate if packing is dummy */
487 unsigned int tx_stepdown_count[TX_SIZES];
496 int use_large_partition_rate;
500 search_site_config ss_cfg;
502 int mbmode_cost[INTRA_MODES];
503 unsigned inter_mode_cost[INTER_MODE_CONTEXTS][INTER_MODES];
504 int intra_uv_mode_cost[FRAME_TYPES][INTRA_MODES];
505 int y_mode_costs[INTRA_MODES][INTRA_MODES][INTRA_MODES];
506 int switchable_interp_costs[SWITCHABLE_FILTER_CONTEXTS][SWITCHABLE_FILTERS];
508 #if CONFIG_MULTIPLE_ARF
509 // ARF tracking variables.
510 int multi_arf_enabled;
511 unsigned int frame_coding_order_period;
512 unsigned int new_frame_coding_order_period;
513 int frame_coding_order[MAX_LAG_BUFFERS * 2];
514 int arf_buffer_idx[MAX_LAG_BUFFERS * 3 / 2];
515 int arf_weight[MAX_LAG_BUFFERS];
517 int this_frame_weight;
521 #ifdef MODE_TEST_HIT_STATS
522 // Debug / test stats
523 int64_t mode_test_hits[BLOCK_SIZES];
527 void vp9_initialize_enc();
529 struct VP9_COMP *vp9_create_compressor(VP9EncoderConfig *oxcf);
530 void vp9_remove_compressor(VP9_COMP *cpi);
532 void vp9_change_config(VP9_COMP *cpi, const VP9EncoderConfig *oxcf);
534 // receive a frames worth of data. caller can assume that a copy of this
535 // frame is made and not just a copy of the pointer..
536 int vp9_receive_raw_frame(VP9_COMP *cpi, unsigned int frame_flags,
537 YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
538 int64_t end_time_stamp);
540 int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
541 size_t *size, uint8_t *dest,
542 int64_t *time_stamp, int64_t *time_end, int flush);
544 int vp9_get_preview_raw_frame(VP9_COMP *cpi, YV12_BUFFER_CONFIG *dest,
545 vp9_ppflags_t *flags);
547 int vp9_use_as_reference(VP9_COMP *cpi, int ref_frame_flags);
549 void vp9_update_reference(VP9_COMP *cpi, int ref_frame_flags);
551 int vp9_copy_reference_enc(VP9_COMP *cpi, VP9_REFFRAME ref_frame_flag,
552 YV12_BUFFER_CONFIG *sd);
554 int vp9_get_reference_enc(VP9_COMP *cpi, int index,
555 YV12_BUFFER_CONFIG **fb);
557 int vp9_set_reference_enc(VP9_COMP *cpi, VP9_REFFRAME ref_frame_flag,
558 YV12_BUFFER_CONFIG *sd);
560 int vp9_update_entropy(VP9_COMP *cpi, int update);
562 int vp9_set_roimap(VP9_COMP *cpi, unsigned char *map,
563 unsigned int rows, unsigned int cols,
564 int delta_q[MAX_SEGMENTS],
565 int delta_lf[MAX_SEGMENTS],
566 unsigned int threshold[MAX_SEGMENTS]);
568 int vp9_set_active_map(VP9_COMP *cpi, unsigned char *map,
569 unsigned int rows, unsigned int cols);
571 int vp9_set_internal_size(VP9_COMP *cpi,
572 VPX_SCALING horiz_mode, VPX_SCALING vert_mode);
574 int vp9_set_size_literal(VP9_COMP *cpi, unsigned int width,
575 unsigned int height);
577 void vp9_set_svc(VP9_COMP *cpi, int use_svc);
579 int vp9_get_quantizer(struct VP9_COMP *cpi);
581 static INLINE int get_ref_frame_idx(const VP9_COMP *cpi,
582 MV_REFERENCE_FRAME ref_frame) {
583 if (ref_frame == LAST_FRAME) {
584 return cpi->lst_fb_idx;
585 } else if (ref_frame == GOLDEN_FRAME) {
586 return cpi->gld_fb_idx;
588 return cpi->alt_fb_idx;
592 static INLINE YV12_BUFFER_CONFIG *get_ref_frame_buffer(
593 VP9_COMP *cpi, MV_REFERENCE_FRAME ref_frame) {
594 VP9_COMMON * const cm = &cpi->common;
595 return &cm->frame_bufs[cm->ref_frame_map[get_ref_frame_idx(cpi, ref_frame)]]
599 // Intra only frames, golden frames (except alt ref overlays) and
600 // alt ref frames tend to be coded at a higher than ambient quality
601 static INLINE int frame_is_boosted(const VP9_COMP *cpi) {
602 return frame_is_intra_only(&cpi->common) || cpi->refresh_alt_ref_frame ||
603 (cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref) ||
604 vp9_is_upper_layer_key_frame(cpi);
607 static INLINE int get_token_alloc(int mb_rows, int mb_cols) {
608 // TODO(JBB): make this work for alpha channel and double check we can't
609 // exceed this token count if we have a 32x32 transform crossing a boundary
610 // at a multiple of 16.
611 // mb_rows, cols are in units of 16 pixels. We assume 3 planes all at full
612 // resolution. We assume up to 1 token per pixel, and then allow
614 return mb_rows * mb_cols * (16 * 16 * 3 + 4);
617 int vp9_get_y_sse(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b);
619 void vp9_alloc_compressor_data(VP9_COMP *cpi);
621 void vp9_scale_references(VP9_COMP *cpi);
623 void vp9_update_reference_frames(VP9_COMP *cpi);
625 int64_t vp9_rescale(int64_t val, int64_t num, int denom);
627 YV12_BUFFER_CONFIG *vp9_scale_if_required(VP9_COMMON *cm,
628 YV12_BUFFER_CONFIG *unscaled,
629 YV12_BUFFER_CONFIG *scaled);
631 static INLINE void set_ref_ptrs(VP9_COMMON *cm, MACROBLOCKD *xd,
632 MV_REFERENCE_FRAME ref0,
633 MV_REFERENCE_FRAME ref1) {
634 xd->block_refs[0] = &cm->frame_refs[ref0 >= LAST_FRAME ? ref0 - LAST_FRAME
636 xd->block_refs[1] = &cm->frame_refs[ref1 >= LAST_FRAME ? ref1 - LAST_FRAME
644 #endif // VP9_ENCODER_VP9_ENCODER_H_