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 #include "vp9/encoder/vp9_speed_features.h"
38 #include "vp9/encoder/vp9_svc_layercontext.h"
39 #include "vp9/encoder/vp9_tokenize.h"
40 #include "vp9/encoder/vp9_variance.h"
42 #if CONFIG_VP9_TEMPORAL_DENOISING
43 #include "vp9/encoder/vp9_denoiser.h"
50 #define DEFAULT_GF_INTERVAL 10
53 int nmvjointcost[MV_JOINTS];
54 int nmvcosts[2][MV_VALS];
55 int nmvcosts_hp[2][MV_VALS];
57 vp9_prob segment_pred_probs[PREDICTION_PROBS];
59 unsigned char *last_frame_seg_map_copy;
61 // 0 = Intra, Last, GF, ARF
62 signed char last_ref_lf_deltas[MAX_REF_LF_DELTAS];
64 signed char last_mode_lf_deltas[MAX_MODE_LF_DELTAS];
71 // encode_breakout is disabled.
72 ENCODE_BREAKOUT_DISABLED = 0,
73 // encode_breakout is enabled.
74 ENCODE_BREAKOUT_ENABLED = 1,
75 // encode_breakout is enabled with small max_thresh limit.
76 ENCODE_BREAKOUT_LIMITED = 2
77 } ENCODE_BREAKOUT_TYPE;
87 // Good Quality Fast Encoding. The encoder balances quality with the amount of
88 // time it takes to encode the output. Speed setting controls how fast.
91 // The encoder places priority on the quality of the output over encoding
92 // speed. The output is compressed at the highest possible quality. This
93 // option takes the longest amount of time to encode. Speed setting ignored.
96 // Realtime/Live Encoding. This mode is optimized for realtime encoding (for
97 // example, capturing a television signal or feed from a live camera). Speed
98 // setting controls how fast.
103 FRAMEFLAGS_KEY = 1 << 0,
104 FRAMEFLAGS_GOLDEN = 1 << 1,
105 FRAMEFLAGS_ALTREF = 1 << 2,
112 CYCLIC_REFRESH_AQ = 3,
113 AQ_MODE_COUNT // This should always be the last member of the enum
117 RESIZE_NONE = 0, // No frame resizing allowed (except for SVC).
118 RESIZE_FIXED = 1, // All frames are coded at the specified dimension.
119 RESIZE_DYNAMIC = 2 // Coded size of each frame is determined by the codec.
122 typedef struct VP9EncoderConfig {
123 BITSTREAM_PROFILE profile;
124 vpx_bit_depth_t bit_depth; // Codec bit-depth.
125 int width; // width of data passed to the compressor
126 int height; // height of data passed to the compressor
127 unsigned int input_bit_depth; // Input bit depth.
128 double init_framerate; // set to passed in framerate
129 int64_t target_bandwidth; // bandwidth to be used in kilobits per second
131 int noise_sensitivity; // pre processing blur: recommendation 0
132 int sharpness; // sharpening output: recommendation 0:
134 // maximum allowed bitrate for any intra frame in % of bitrate target.
135 unsigned int rc_max_intra_bitrate_pct;
136 // maximum allowed bitrate for any inter frame in % of bitrate target.
137 unsigned int rc_max_inter_bitrate_pct;
138 // percent of rate boost for golden frame in CBR mode.
139 unsigned int gf_cbr_boost_pct;
144 // Key Framing Operations
145 int auto_key; // autodetect cut scenes and set the keyframes
146 int key_freq; // maximum distance to key frame.
148 int lag_in_frames; // how many frames lag before we start encoding
150 // ----------------------------------------------------------------
151 // DATARATE CONTROL OPTIONS
153 // vbr, cbr, constrained quality or constant quality
154 enum vpx_rc_mode rc_mode;
156 // buffer targeting aggressiveness
160 // buffering parameters
161 int64_t starting_buffer_level_ms;
162 int64_t optimal_buffer_level_ms;
163 int64_t maximum_buffer_size_ms;
165 // Frame drop threshold.
166 int drop_frames_water_mark;
168 // controlling quality
173 AQ_MODE aq_mode; // Adaptive Quantization mode
175 // Internal frame size scaling.
176 RESIZE_TYPE resize_mode;
177 int scaled_frame_width;
178 int scaled_frame_height;
180 // Enable feature to reduce the frame quantization every x frames.
181 int frame_periodic_boost;
183 // two pass datarate control
184 int two_pass_vbrbias; // two pass datarate control tweaks
185 int two_pass_vbrmin_section;
186 int two_pass_vbrmax_section;
187 // END DATARATE CONTROL OPTIONS
188 // ----------------------------------------------------------------
190 // Spatial and temporal scalability.
191 int ss_number_layers; // Number of spatial layers.
192 int ts_number_layers; // Number of temporal layers.
193 // Bitrate allocation for spatial layers.
194 int ss_target_bitrate[VPX_SS_MAX_LAYERS];
195 int ss_enable_auto_arf[VPX_SS_MAX_LAYERS];
196 // Bitrate allocation (CBR mode) and framerate factor, for temporal layers.
197 int ts_target_bitrate[VPX_TS_MAX_LAYERS];
198 int ts_rate_decimator[VPX_TS_MAX_LAYERS];
202 int encode_breakout; // early breakout : for video conf recommend 800
204 /* Bitfield defining the error resiliency features to enable.
205 * Can provide decodable frames after losses in previous
206 * frames and decodable partitions after losses in the same frame.
208 unsigned int error_resilient_mode;
210 /* Bitfield defining the parallel decoding mode where the
211 * decoding in successive frames may be conducted in parallel
212 * just by decoding the frame headers.
214 unsigned int frame_parallel_decoding_mode;
224 vpx_fixed_buf_t two_pass_stats_in;
225 struct vpx_codec_pkt_list *output_pkt_list;
227 #if CONFIG_FP_MB_STATS
228 vpx_fixed_buf_t firstpass_mb_stats_in;
232 vp9e_tune_content content;
233 #if CONFIG_VP9_HIGHBITDEPTH
234 int use_highbitdepth;
236 vpx_color_space_t color_space;
239 static INLINE int is_lossless_requested(const VP9EncoderConfig *cfg) {
240 return cfg->best_allowed_q == 0 && cfg->worst_allowed_q == 0;
243 // TODO(jingning) All spatially adaptive variables should go to TileDataEnc.
244 typedef struct TileDataEnc {
246 int thresh_freq_fact[BLOCK_SIZES][MAX_MODES];
247 int mode_map[BLOCK_SIZES][MAX_MODES];
250 typedef struct RD_COUNTS {
251 vp9_coeff_count coef_counts[TX_SIZES][PLANE_TYPES];
252 int64_t comp_pred_diff[REFERENCE_MODES];
253 int64_t tx_select_diff[TX_MODES];
254 int64_t filter_diff[SWITCHABLE_FILTER_CONTEXTS];
257 typedef struct ThreadData {
260 FRAME_COUNTS *counts;
262 PICK_MODE_CONTEXT *leaf_tree;
267 struct EncWorkerData;
269 typedef struct ActiveMap {
275 typedef struct VP9_COMP {
278 DECLARE_ALIGNED(16, int16_t, y_dequant[QINDEX_RANGE][8]);
279 DECLARE_ALIGNED(16, int16_t, uv_dequant[QINDEX_RANGE][8]);
281 VP9EncoderConfig oxcf;
282 struct lookahead_ctx *lookahead;
283 struct lookahead_entry *alt_ref_source;
285 YV12_BUFFER_CONFIG *Source;
286 YV12_BUFFER_CONFIG *Last_Source; // NULL for first frame and alt_ref frames
287 YV12_BUFFER_CONFIG *un_scaled_source;
288 YV12_BUFFER_CONFIG scaled_source;
289 YV12_BUFFER_CONFIG *unscaled_last_source;
290 YV12_BUFFER_CONFIG scaled_last_source;
292 TileDataEnc *tile_data;
294 // For a still frame, this flag is set to 1 to skip partition search.
295 int partition_search_skippable_frame;
297 int scaled_ref_idx[MAX_REF_FRAMES];
302 int refresh_last_frame;
303 int refresh_golden_frame;
304 int refresh_alt_ref_frame;
306 int ext_refresh_frame_flags_pending;
307 int ext_refresh_last_frame;
308 int ext_refresh_golden_frame;
309 int ext_refresh_alt_ref_frame;
311 int ext_refresh_frame_context_pending;
312 int ext_refresh_frame_context;
314 YV12_BUFFER_CONFIG last_frame_uf;
316 TOKENEXTRA *tile_tok[4][1 << 6];
317 unsigned int tok_count[4][1 << 6];
319 // Ambient reconstruction err target for force key frames
324 CODING_CONTEXT coding_context;
329 int *nmvsadcosts_hp[2];
331 int64_t last_time_stamp_seen;
332 int64_t last_end_time_stamp_seen;
333 int64_t first_time_stamp_ever;
338 int interp_filter_selected[MAX_REF_FRAMES][SWITCHABLE];
340 struct vpx_codec_pkt_list *output_pkt_list;
342 MBGRAPH_FRAME_STATS mbgraph_stats[MAX_LAG_BUFFERS];
343 int mbgraph_n_frames; // number of frames filled in the above
344 int static_mb_pct; // % forced skip mbs by segmentation
349 unsigned int max_mv_magnitude;
352 int allow_comp_inter_inter;
354 // Default value is 1. From first pass stats, encode_breakout may be disabled.
355 ENCODE_BREAKOUT_TYPE allow_encode_breakout;
357 // Get threshold from external input. A suggested threshold is 800 for HD
358 // clips, and 300 for < HD clips.
361 unsigned char *segmentation_map;
363 // segment threashold for encode breakout
364 int segment_encode_breakout[MAX_SEGMENTS];
366 CYCLIC_REFRESH *cyclic_refresh;
367 ActiveMap active_map;
369 fractional_mv_step_fp *find_fractional_mv_step;
370 vp9_full_search_fn_t full_search_sad;
371 vp9_diamond_search_fn_t diamond_search_sad;
372 vp9_variance_fn_ptr_t fn_ptr[BLOCK_SIZES];
373 uint64_t time_receive_data;
374 uint64_t time_compress_data;
375 uint64_t time_pick_lpf;
376 uint64_t time_encode_sb_row;
378 #if CONFIG_FP_MB_STATS
384 YV12_BUFFER_CONFIG alt_ref_buffer;
387 #if CONFIG_INTERNAL_STATS
388 unsigned int mode_chosen_counts[MAX_MODES];
395 uint64_t total_sq_error;
396 uint64_t total_samples;
402 uint64_t totalp_sq_error;
403 uint64_t totalp_samples;
406 double summed_quality;
407 double summed_weights;
408 double summedp_quality;
409 double summedp_weights;
410 unsigned int tot_recode_hits;
413 double total_ssimg_y;
414 double total_ssimg_u;
415 double total_ssimg_v;
416 double total_ssimg_all;
418 int b_calculate_ssimg;
420 int b_calculate_psnr;
426 int initial_mbs; // Number of MBs in the full-size frame; to be used to
427 // normalize the firstpass stats. This will differ from the
428 // number of MBs in the current frame when the frame is
435 // Store frame variance info in SOURCE_VAR_BASED_PARTITION search type.
436 diff *source_diff_var;
437 // The threshold used in SOURCE_VAR_BASED_PARTITION search type.
438 unsigned int source_var_thresh;
439 int frames_till_next_var_check;
443 search_site_config ss_cfg;
445 int mbmode_cost[INTRA_MODES];
446 unsigned int inter_mode_cost[INTER_MODE_CONTEXTS][INTER_MODES];
447 int intra_uv_mode_cost[FRAME_TYPES][INTRA_MODES];
448 int y_mode_costs[INTRA_MODES][INTRA_MODES][INTRA_MODES];
449 int switchable_interp_costs[SWITCHABLE_FILTER_CONTEXTS][SWITCHABLE_FILTERS];
450 int partition_cost[PARTITION_CONTEXTS][PARTITION_TYPES];
452 int multi_arf_allowed;
453 int multi_arf_enabled;
454 int multi_arf_last_grp_enabled;
456 #if CONFIG_VP9_TEMPORAL_DENOISING
457 VP9_DENOISER denoiser;
462 // VAR_BASED_PARTITION thresholds
463 int64_t vbp_threshold;
464 int64_t vbp_threshold_bsize_min;
465 int64_t vbp_threshold_bsize_max;
466 int64_t vbp_threshold_16x16;
467 BLOCK_SIZE vbp_bsize_min;
472 struct EncWorkerData *tile_thr_data;
473 VP9LfSync lf_row_sync;
476 void vp9_initialize_enc(void);
478 struct VP9_COMP *vp9_create_compressor(VP9EncoderConfig *oxcf,
479 BufferPool *const pool);
480 void vp9_remove_compressor(VP9_COMP *cpi);
482 void vp9_change_config(VP9_COMP *cpi, const VP9EncoderConfig *oxcf);
484 // receive a frames worth of data. caller can assume that a copy of this
485 // frame is made and not just a copy of the pointer..
486 int vp9_receive_raw_frame(VP9_COMP *cpi, unsigned int frame_flags,
487 YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
488 int64_t end_time_stamp);
490 int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
491 size_t *size, uint8_t *dest,
492 int64_t *time_stamp, int64_t *time_end, int flush);
494 int vp9_get_preview_raw_frame(VP9_COMP *cpi, YV12_BUFFER_CONFIG *dest,
495 vp9_ppflags_t *flags);
497 int vp9_use_as_reference(VP9_COMP *cpi, int ref_frame_flags);
499 void vp9_update_reference(VP9_COMP *cpi, int ref_frame_flags);
501 int vp9_copy_reference_enc(VP9_COMP *cpi, VP9_REFFRAME ref_frame_flag,
502 YV12_BUFFER_CONFIG *sd);
504 int vp9_set_reference_enc(VP9_COMP *cpi, VP9_REFFRAME ref_frame_flag,
505 YV12_BUFFER_CONFIG *sd);
507 int vp9_update_entropy(VP9_COMP *cpi, int update);
509 int vp9_set_active_map(VP9_COMP *cpi, unsigned char *map, int rows, int cols);
511 int vp9_get_active_map(VP9_COMP *cpi, unsigned char *map, int rows, int cols);
513 int vp9_set_internal_size(VP9_COMP *cpi,
514 VPX_SCALING horiz_mode, VPX_SCALING vert_mode);
516 int vp9_set_size_literal(VP9_COMP *cpi, unsigned int width,
517 unsigned int height);
519 void vp9_set_svc(VP9_COMP *cpi, int use_svc);
521 int vp9_get_quantizer(struct VP9_COMP *cpi);
523 static INLINE int frame_is_kf_gf_arf(const VP9_COMP *cpi) {
524 return frame_is_intra_only(&cpi->common) ||
525 cpi->refresh_alt_ref_frame ||
526 (cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref);
529 static INLINE int get_ref_frame_map_idx(const VP9_COMP *cpi,
530 MV_REFERENCE_FRAME ref_frame) {
531 if (ref_frame == LAST_FRAME) {
532 return cpi->lst_fb_idx;
533 } else if (ref_frame == GOLDEN_FRAME) {
534 return cpi->gld_fb_idx;
536 return cpi->alt_fb_idx;
540 static INLINE int get_ref_frame_buf_idx(const VP9_COMP *const cpi,
542 const VP9_COMMON *const cm = &cpi->common;
543 const int map_idx = get_ref_frame_map_idx(cpi, ref_frame);
544 return (map_idx != INVALID_IDX) ? cm->ref_frame_map[map_idx] : INVALID_IDX;
547 static INLINE YV12_BUFFER_CONFIG *get_ref_frame_buffer(
548 VP9_COMP *cpi, MV_REFERENCE_FRAME ref_frame) {
549 VP9_COMMON *const cm = &cpi->common;
550 const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
552 buf_idx != INVALID_IDX ? &cm->buffer_pool->frame_bufs[buf_idx].buf : NULL;
555 static INLINE int get_token_alloc(int mb_rows, int mb_cols) {
556 // TODO(JBB): double check we can't exceed this token count if we have a
557 // 32x32 transform crossing a boundary at a multiple of 16.
558 // mb_rows, cols are in units of 16 pixels. We assume 3 planes all at full
559 // resolution. We assume up to 1 token per pixel, and then allow
561 return mb_rows * mb_cols * (16 * 16 * 3 + 4);
564 // Get the allocated token size for a tile. It does the same calculation as in
565 // the frame token allocation.
566 static INLINE int allocated_tokens(TileInfo tile) {
567 int tile_mb_rows = (tile.mi_row_end - tile.mi_row_start + 1) >> 1;
568 int tile_mb_cols = (tile.mi_col_end - tile.mi_col_start + 1) >> 1;
570 return get_token_alloc(tile_mb_rows, tile_mb_cols);
573 int64_t vp9_get_y_sse(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b);
574 #if CONFIG_VP9_HIGHBITDEPTH
575 int64_t vp9_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
576 const YV12_BUFFER_CONFIG *b);
577 #endif // CONFIG_VP9_HIGHBITDEPTH
579 void vp9_alloc_compressor_data(VP9_COMP *cpi);
581 void vp9_scale_references(VP9_COMP *cpi);
583 void vp9_update_reference_frames(VP9_COMP *cpi);
585 void vp9_set_high_precision_mv(VP9_COMP *cpi, int allow_high_precision_mv);
587 YV12_BUFFER_CONFIG *vp9_scale_if_required(VP9_COMMON *cm,
588 YV12_BUFFER_CONFIG *unscaled,
589 YV12_BUFFER_CONFIG *scaled);
591 void vp9_apply_encoding_flags(VP9_COMP *cpi, vpx_enc_frame_flags_t flags);
593 static INLINE int is_two_pass_svc(const struct VP9_COMP *const cpi) {
594 return cpi->use_svc &&
595 ((cpi->svc.number_spatial_layers > 1) ||
596 (cpi->svc.number_temporal_layers > 1 && cpi->oxcf.pass != 0));
599 static INLINE int is_altref_enabled(const VP9_COMP *const cpi) {
600 return cpi->oxcf.mode != REALTIME && cpi->oxcf.lag_in_frames > 0 &&
601 (cpi->oxcf.enable_auto_arf &&
602 (!is_two_pass_svc(cpi) ||
603 cpi->oxcf.ss_enable_auto_arf[cpi->svc.spatial_layer_id]));
606 static INLINE void set_ref_ptrs(VP9_COMMON *cm, MACROBLOCKD *xd,
607 MV_REFERENCE_FRAME ref0,
608 MV_REFERENCE_FRAME ref1) {
609 xd->block_refs[0] = &cm->frame_refs[ref0 >= LAST_FRAME ? ref0 - LAST_FRAME
611 xd->block_refs[1] = &cm->frame_refs[ref1 >= LAST_FRAME ? ref1 - LAST_FRAME
615 static INLINE int get_chessboard_index(const int frame_index) {
616 return frame_index & 0x1;
619 static INLINE int *cond_cost_list(const struct VP9_COMP *cpi, int *cost_list) {
620 return cpi->sf.mv.subpel_search_method != SUBPEL_TREE ? cost_list : NULL;
623 void vp9_new_framerate(VP9_COMP *cpi, double framerate);
629 #endif // VP9_ENCODER_VP9_ENCODER_H_