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/internal/vpx_codec_internal.h"
18 #include "vpx/vp8cx.h"
20 #include "vp9/common/vp9_alloccommon.h"
21 #include "vp9/common/vp9_ppflags.h"
22 #include "vp9/common/vp9_entropymode.h"
23 #include "vp9/common/vp9_thread_common.h"
24 #include "vp9/common/vp9_onyxc_int.h"
25 #include "vp9/common/vp9_thread.h"
27 #include "vp9/encoder/vp9_aq_cyclicrefresh.h"
28 #include "vp9/encoder/vp9_context_tree.h"
29 #include "vp9/encoder/vp9_encodemb.h"
30 #include "vp9/encoder/vp9_firstpass.h"
31 #include "vp9/encoder/vp9_lookahead.h"
32 #include "vp9/encoder/vp9_mbgraph.h"
33 #include "vp9/encoder/vp9_mcomp.h"
34 #include "vp9/encoder/vp9_quantize.h"
35 #include "vp9/encoder/vp9_ratectrl.h"
36 #include "vp9/encoder/vp9_rd.h"
37 #if CONFIG_INTERNAL_STATS
38 #include "vp9/encoder/vp9_ssim.h"
40 #include "vp9/encoder/vp9_speed_features.h"
41 #include "vp9/encoder/vp9_svc_layercontext.h"
42 #include "vp9/encoder/vp9_tokenize.h"
43 #include "vp9/encoder/vp9_variance.h"
45 #if CONFIG_VP9_TEMPORAL_DENOISING
46 #include "vp9/encoder/vp9_denoiser.h"
53 #define DEFAULT_GF_INTERVAL 10
56 int nmvjointcost[MV_JOINTS];
57 int nmvcosts[2][MV_VALS];
58 int nmvcosts_hp[2][MV_VALS];
60 vp9_prob segment_pred_probs[PREDICTION_PROBS];
62 unsigned char *last_frame_seg_map_copy;
64 // 0 = Intra, Last, GF, ARF
65 signed char last_ref_lf_deltas[MAX_REF_LF_DELTAS];
67 signed char last_mode_lf_deltas[MAX_MODE_LF_DELTAS];
74 // encode_breakout is disabled.
75 ENCODE_BREAKOUT_DISABLED = 0,
76 // encode_breakout is enabled.
77 ENCODE_BREAKOUT_ENABLED = 1,
78 // encode_breakout is enabled with small max_thresh limit.
79 ENCODE_BREAKOUT_LIMITED = 2
80 } ENCODE_BREAKOUT_TYPE;
90 // Good Quality Fast Encoding. The encoder balances quality with the amount of
91 // time it takes to encode the output. Speed setting controls how fast.
94 // The encoder places priority on the quality of the output over encoding
95 // speed. The output is compressed at the highest possible quality. This
96 // option takes the longest amount of time to encode. Speed setting ignored.
99 // Realtime/Live Encoding. This mode is optimized for realtime encoding (for
100 // example, capturing a television signal or feed from a live camera). Speed
101 // setting controls how fast.
106 FRAMEFLAGS_KEY = 1 << 0,
107 FRAMEFLAGS_GOLDEN = 1 << 1,
108 FRAMEFLAGS_ALTREF = 1 << 2,
115 CYCLIC_REFRESH_AQ = 3,
116 AQ_MODE_COUNT // This should always be the last member of the enum
120 RESIZE_NONE = 0, // No frame resizing allowed (except for SVC).
121 RESIZE_FIXED = 1, // All frames are coded at the specified dimension.
122 RESIZE_DYNAMIC = 2 // Coded size of each frame is determined by the codec.
125 typedef struct VP9EncoderConfig {
126 BITSTREAM_PROFILE profile;
127 vpx_bit_depth_t bit_depth; // Codec bit-depth.
128 int width; // width of data passed to the compressor
129 int height; // height of data passed to the compressor
130 unsigned int input_bit_depth; // Input bit depth.
131 double init_framerate; // set to passed in framerate
132 int64_t target_bandwidth; // bandwidth to be used in kilobits per second
134 int noise_sensitivity; // pre processing blur: recommendation 0
135 int sharpness; // sharpening output: recommendation 0:
137 // maximum allowed bitrate for any intra frame in % of bitrate target.
138 unsigned int rc_max_intra_bitrate_pct;
139 // maximum allowed bitrate for any inter frame in % of bitrate target.
140 unsigned int rc_max_inter_bitrate_pct;
141 // percent of rate boost for golden frame in CBR mode.
142 unsigned int gf_cbr_boost_pct;
147 // Key Framing Operations
148 int auto_key; // autodetect cut scenes and set the keyframes
149 int key_freq; // maximum distance to key frame.
151 int lag_in_frames; // how many frames lag before we start encoding
153 // ----------------------------------------------------------------
154 // DATARATE CONTROL OPTIONS
156 // vbr, cbr, constrained quality or constant quality
157 enum vpx_rc_mode rc_mode;
159 // buffer targeting aggressiveness
163 // buffering parameters
164 int64_t starting_buffer_level_ms;
165 int64_t optimal_buffer_level_ms;
166 int64_t maximum_buffer_size_ms;
168 // Frame drop threshold.
169 int drop_frames_water_mark;
171 // controlling quality
176 AQ_MODE aq_mode; // Adaptive Quantization mode
178 // Internal frame size scaling.
179 RESIZE_TYPE resize_mode;
180 int scaled_frame_width;
181 int scaled_frame_height;
183 // Enable feature to reduce the frame quantization every x frames.
184 int frame_periodic_boost;
186 // two pass datarate control
187 int two_pass_vbrbias; // two pass datarate control tweaks
188 int two_pass_vbrmin_section;
189 int two_pass_vbrmax_section;
190 // END DATARATE CONTROL OPTIONS
191 // ----------------------------------------------------------------
193 // Spatial and temporal scalability.
194 int ss_number_layers; // Number of spatial layers.
195 int ts_number_layers; // Number of temporal layers.
196 // Bitrate allocation for spatial layers.
197 int layer_target_bitrate[VPX_MAX_LAYERS];
198 int ss_target_bitrate[VPX_SS_MAX_LAYERS];
199 int ss_enable_auto_arf[VPX_SS_MAX_LAYERS];
200 // Bitrate allocation (CBR mode) and framerate factor, for temporal layers.
201 int ts_rate_decimator[VPX_TS_MAX_LAYERS];
205 int encode_breakout; // early breakout : for video conf recommend 800
207 /* Bitfield defining the error resiliency features to enable.
208 * Can provide decodable frames after losses in previous
209 * frames and decodable partitions after losses in the same frame.
211 unsigned int error_resilient_mode;
213 /* Bitfield defining the parallel decoding mode where the
214 * decoding in successive frames may be conducted in parallel
215 * just by decoding the frame headers.
217 unsigned int frame_parallel_decoding_mode;
227 vpx_fixed_buf_t two_pass_stats_in;
228 struct vpx_codec_pkt_list *output_pkt_list;
230 #if CONFIG_FP_MB_STATS
231 vpx_fixed_buf_t firstpass_mb_stats_in;
235 vp9e_tune_content content;
236 #if CONFIG_VP9_HIGHBITDEPTH
237 int use_highbitdepth;
239 vpx_color_space_t color_space;
240 VP9E_TEMPORAL_LAYERING_MODE temporal_layering_mode;
243 static INLINE int is_lossless_requested(const VP9EncoderConfig *cfg) {
244 return cfg->best_allowed_q == 0 && cfg->worst_allowed_q == 0;
247 // TODO(jingning) All spatially adaptive variables should go to TileDataEnc.
248 typedef struct TileDataEnc {
250 int thresh_freq_fact[BLOCK_SIZES][MAX_MODES];
251 int mode_map[BLOCK_SIZES][MAX_MODES];
254 typedef struct RD_COUNTS {
255 vp9_coeff_count coef_counts[TX_SIZES][PLANE_TYPES];
256 int64_t comp_pred_diff[REFERENCE_MODES];
257 int64_t tx_select_diff[TX_MODES];
258 int64_t filter_diff[SWITCHABLE_FILTER_CONTEXTS];
261 typedef struct ThreadData {
264 FRAME_COUNTS *counts;
266 PICK_MODE_CONTEXT *leaf_tree;
271 struct EncWorkerData;
273 typedef struct ActiveMap {
286 typedef struct IMAGE_STAT {
291 typedef struct VP9_COMP {
294 DECLARE_ALIGNED(16, int16_t, y_dequant[QINDEX_RANGE][8]);
295 DECLARE_ALIGNED(16, int16_t, uv_dequant[QINDEX_RANGE][8]);
297 VP9EncoderConfig oxcf;
298 struct lookahead_ctx *lookahead;
299 struct lookahead_entry *alt_ref_source;
301 YV12_BUFFER_CONFIG *Source;
302 YV12_BUFFER_CONFIG *Last_Source; // NULL for first frame and alt_ref frames
303 YV12_BUFFER_CONFIG *un_scaled_source;
304 YV12_BUFFER_CONFIG scaled_source;
305 YV12_BUFFER_CONFIG *unscaled_last_source;
306 YV12_BUFFER_CONFIG scaled_last_source;
308 TileDataEnc *tile_data;
309 int allocated_tiles; // Keep track of memory allocated for tiles.
311 // For a still frame, this flag is set to 1 to skip partition search.
312 int partition_search_skippable_frame;
314 int scaled_ref_idx[MAX_REF_FRAMES];
319 int refresh_last_frame;
320 int refresh_golden_frame;
321 int refresh_alt_ref_frame;
323 int ext_refresh_frame_flags_pending;
324 int ext_refresh_last_frame;
325 int ext_refresh_golden_frame;
326 int ext_refresh_alt_ref_frame;
328 int ext_refresh_frame_context_pending;
329 int ext_refresh_frame_context;
331 YV12_BUFFER_CONFIG last_frame_uf;
333 TOKENEXTRA *tile_tok[4][1 << 6];
334 unsigned int tok_count[4][1 << 6];
336 // Ambient reconstruction err target for force key frames
341 CODING_CONTEXT coding_context;
346 int *nmvsadcosts_hp[2];
348 int64_t last_time_stamp_seen;
349 int64_t last_end_time_stamp_seen;
350 int64_t first_time_stamp_ever;
355 int interp_filter_selected[MAX_REF_FRAMES][SWITCHABLE];
357 struct vpx_codec_pkt_list *output_pkt_list;
359 MBGRAPH_FRAME_STATS mbgraph_stats[MAX_LAG_BUFFERS];
360 int mbgraph_n_frames; // number of frames filled in the above
361 int static_mb_pct; // % forced skip mbs by segmentation
366 unsigned int max_mv_magnitude;
369 int allow_comp_inter_inter;
371 // Default value is 1. From first pass stats, encode_breakout may be disabled.
372 ENCODE_BREAKOUT_TYPE allow_encode_breakout;
374 // Get threshold from external input. A suggested threshold is 800 for HD
375 // clips, and 300 for < HD clips.
378 unsigned char *segmentation_map;
380 // segment threashold for encode breakout
381 int segment_encode_breakout[MAX_SEGMENTS];
383 CYCLIC_REFRESH *cyclic_refresh;
384 ActiveMap active_map;
386 fractional_mv_step_fp *find_fractional_mv_step;
387 vp9_full_search_fn_t full_search_sad;
388 vp9_diamond_search_fn_t diamond_search_sad;
389 vp9_variance_fn_ptr_t fn_ptr[BLOCK_SIZES];
390 uint64_t time_receive_data;
391 uint64_t time_compress_data;
392 uint64_t time_pick_lpf;
393 uint64_t time_encode_sb_row;
395 #if CONFIG_FP_MB_STATS
401 YV12_BUFFER_CONFIG alt_ref_buffer;
404 #if CONFIG_INTERNAL_STATS
405 unsigned int mode_chosen_counts[MAX_MODES];
408 uint64_t total_sq_error;
409 uint64_t total_samples;
412 uint64_t totalp_sq_error;
413 uint64_t totalp_samples;
416 double total_blockiness;
417 double worst_blockiness;
420 double summed_quality;
421 double summed_weights;
422 double summedp_quality;
423 double summedp_weights;
424 unsigned int tot_recode_hits;
431 int b_calculate_ssimg;
432 int b_calculate_blockiness;
434 int b_calculate_consistency;
436 double total_inconsistency;
437 double worst_consistency;
441 int b_calculate_psnr;
447 int initial_mbs; // Number of MBs in the full-size frame; to be used to
448 // normalize the firstpass stats. This will differ from the
449 // number of MBs in the current frame when the frame is
456 // Store frame variance info in SOURCE_VAR_BASED_PARTITION search type.
457 diff *source_diff_var;
458 // The threshold used in SOURCE_VAR_BASED_PARTITION search type.
459 unsigned int source_var_thresh;
460 int frames_till_next_var_check;
464 search_site_config ss_cfg;
466 int mbmode_cost[INTRA_MODES];
467 unsigned int inter_mode_cost[INTER_MODE_CONTEXTS][INTER_MODES];
468 int intra_uv_mode_cost[FRAME_TYPES][INTRA_MODES];
469 int y_mode_costs[INTRA_MODES][INTRA_MODES][INTRA_MODES];
470 int switchable_interp_costs[SWITCHABLE_FILTER_CONTEXTS][SWITCHABLE_FILTERS];
471 int partition_cost[PARTITION_CONTEXTS][PARTITION_TYPES];
473 int multi_arf_allowed;
474 int multi_arf_enabled;
475 int multi_arf_last_grp_enabled;
477 #if CONFIG_VP9_TEMPORAL_DENOISING
478 VP9_DENOISER denoiser;
483 // VAR_BASED_PARTITION thresholds
484 // 0 - threshold_64x64; 1 - threshold_32x32;
485 // 2 - threshold_16x16; 3 - vbp_threshold_8x8;
486 int64_t vbp_thresholds[4];
487 int64_t vbp_threshold_minmax;
488 int64_t vbp_threshold_sad;
489 BLOCK_SIZE vbp_bsize_min;
494 struct EncWorkerData *tile_thr_data;
495 VP9LfSync lf_row_sync;
498 void vp9_initialize_enc(void);
500 struct VP9_COMP *vp9_create_compressor(VP9EncoderConfig *oxcf,
501 BufferPool *const pool);
502 void vp9_remove_compressor(VP9_COMP *cpi);
504 void vp9_change_config(VP9_COMP *cpi, const VP9EncoderConfig *oxcf);
506 // receive a frames worth of data. caller can assume that a copy of this
507 // frame is made and not just a copy of the pointer..
508 int vp9_receive_raw_frame(VP9_COMP *cpi, unsigned int frame_flags,
509 YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
510 int64_t end_time_stamp);
512 int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
513 size_t *size, uint8_t *dest,
514 int64_t *time_stamp, int64_t *time_end, int flush);
516 int vp9_get_preview_raw_frame(VP9_COMP *cpi, YV12_BUFFER_CONFIG *dest,
517 vp9_ppflags_t *flags);
519 int vp9_use_as_reference(VP9_COMP *cpi, int ref_frame_flags);
521 void vp9_update_reference(VP9_COMP *cpi, int ref_frame_flags);
523 int vp9_copy_reference_enc(VP9_COMP *cpi, VP9_REFFRAME ref_frame_flag,
524 YV12_BUFFER_CONFIG *sd);
526 int vp9_set_reference_enc(VP9_COMP *cpi, VP9_REFFRAME ref_frame_flag,
527 YV12_BUFFER_CONFIG *sd);
529 int vp9_update_entropy(VP9_COMP *cpi, int update);
531 int vp9_set_active_map(VP9_COMP *cpi, unsigned char *map, int rows, int cols);
533 int vp9_get_active_map(VP9_COMP *cpi, unsigned char *map, int rows, int cols);
535 int vp9_set_internal_size(VP9_COMP *cpi,
536 VPX_SCALING horiz_mode, VPX_SCALING vert_mode);
538 int vp9_set_size_literal(VP9_COMP *cpi, unsigned int width,
539 unsigned int height);
541 void vp9_set_svc(VP9_COMP *cpi, int use_svc);
543 int vp9_get_quantizer(struct VP9_COMP *cpi);
545 static INLINE int frame_is_kf_gf_arf(const VP9_COMP *cpi) {
546 return frame_is_intra_only(&cpi->common) ||
547 cpi->refresh_alt_ref_frame ||
548 (cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref);
551 static INLINE int get_ref_frame_map_idx(const VP9_COMP *cpi,
552 MV_REFERENCE_FRAME ref_frame) {
553 if (ref_frame == LAST_FRAME) {
554 return cpi->lst_fb_idx;
555 } else if (ref_frame == GOLDEN_FRAME) {
556 return cpi->gld_fb_idx;
558 return cpi->alt_fb_idx;
562 static INLINE int get_ref_frame_buf_idx(const VP9_COMP *const cpi,
564 const VP9_COMMON *const cm = &cpi->common;
565 const int map_idx = get_ref_frame_map_idx(cpi, ref_frame);
566 return (map_idx != INVALID_IDX) ? cm->ref_frame_map[map_idx] : INVALID_IDX;
569 static INLINE YV12_BUFFER_CONFIG *get_ref_frame_buffer(
570 VP9_COMP *cpi, MV_REFERENCE_FRAME ref_frame) {
571 VP9_COMMON *const cm = &cpi->common;
572 const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
574 buf_idx != INVALID_IDX ? &cm->buffer_pool->frame_bufs[buf_idx].buf : NULL;
577 static INLINE int get_token_alloc(int mb_rows, int mb_cols) {
578 // TODO(JBB): double check we can't exceed this token count if we have a
579 // 32x32 transform crossing a boundary at a multiple of 16.
580 // mb_rows, cols are in units of 16 pixels. We assume 3 planes all at full
581 // resolution. We assume up to 1 token per pixel, and then allow
583 return mb_rows * mb_cols * (16 * 16 * 3 + 4);
586 // Get the allocated token size for a tile. It does the same calculation as in
587 // the frame token allocation.
588 static INLINE int allocated_tokens(TileInfo tile) {
589 int tile_mb_rows = (tile.mi_row_end - tile.mi_row_start + 1) >> 1;
590 int tile_mb_cols = (tile.mi_col_end - tile.mi_col_start + 1) >> 1;
592 return get_token_alloc(tile_mb_rows, tile_mb_cols);
595 int64_t vp9_get_y_sse(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b);
596 #if CONFIG_VP9_HIGHBITDEPTH
597 int64_t vp9_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
598 const YV12_BUFFER_CONFIG *b);
599 #endif // CONFIG_VP9_HIGHBITDEPTH
601 void vp9_alloc_compressor_data(VP9_COMP *cpi);
603 void vp9_scale_references(VP9_COMP *cpi);
605 void vp9_update_reference_frames(VP9_COMP *cpi);
607 void vp9_set_high_precision_mv(VP9_COMP *cpi, int allow_high_precision_mv);
609 YV12_BUFFER_CONFIG *vp9_scale_if_required(VP9_COMMON *cm,
610 YV12_BUFFER_CONFIG *unscaled,
611 YV12_BUFFER_CONFIG *scaled);
613 void vp9_apply_encoding_flags(VP9_COMP *cpi, vpx_enc_frame_flags_t flags);
615 static INLINE int is_two_pass_svc(const struct VP9_COMP *const cpi) {
616 return cpi->use_svc && cpi->oxcf.pass != 0;
619 static INLINE int is_one_pass_cbr_svc(const struct VP9_COMP *const cpi) {
620 return (cpi->use_svc && cpi->oxcf.pass == 0);
623 static INLINE int is_altref_enabled(const VP9_COMP *const cpi) {
624 return cpi->oxcf.mode != REALTIME && cpi->oxcf.lag_in_frames > 0 &&
625 (cpi->oxcf.enable_auto_arf &&
626 (!is_two_pass_svc(cpi) ||
627 cpi->oxcf.ss_enable_auto_arf[cpi->svc.spatial_layer_id]));
630 static INLINE void set_ref_ptrs(VP9_COMMON *cm, MACROBLOCKD *xd,
631 MV_REFERENCE_FRAME ref0,
632 MV_REFERENCE_FRAME ref1) {
633 xd->block_refs[0] = &cm->frame_refs[ref0 >= LAST_FRAME ? ref0 - LAST_FRAME
635 xd->block_refs[1] = &cm->frame_refs[ref1 >= LAST_FRAME ? ref1 - LAST_FRAME
639 static INLINE int get_chessboard_index(const int frame_index) {
640 return frame_index & 0x1;
643 static INLINE int *cond_cost_list(const struct VP9_COMP *cpi, int *cost_list) {
644 return cpi->sf.mv.subpel_search_method != SUBPEL_TREE ? cost_list : NULL;
647 void vp9_new_framerate(VP9_COMP *cpi, double framerate);
649 #define LAYER_IDS_TO_IDX(sl, tl, num_tl) ((sl) * (num_tl) + (tl))
655 #endif // VP9_ENCODER_VP9_ENCODER_H_