vpx/vp9/encoder/vp9_onyx_int.h

652 lines
17 KiB
C
Raw Normal View History

2010-05-18 17:58:33 +02:00
/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
2010-05-18 17:58:33 +02:00
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
2010-05-18 17:58:33 +02:00
*/
#ifndef VP9_ENCODER_VP9_ONYX_INT_H_
#define VP9_ENCODER_VP9_ONYX_INT_H_
2010-05-18 17:58:33 +02:00
#include <stdio.h>
#include "./vpx_config.h"
#include "vp9/common/vp9_onyx.h"
#include "vp9/encoder/vp9_treewriter.h"
#include "vp9/encoder/vp9_tokenize.h"
#include "vp9/common/vp9_onyxc_int.h"
#include "vp9/encoder/vp9_variance.h"
#include "vp9/encoder/vp9_encodemb.h"
#include "vp9/encoder/vp9_quantize.h"
#include "vp9/common/vp9_entropy.h"
#include "vp9/common/vp9_entropymode.h"
2010-05-18 17:58:33 +02:00
#include "vpx_ports/mem.h"
#include "vpx/internal/vpx_codec_internal.h"
#include "vp9/encoder/vp9_mcomp.h"
#include "vp9/common/vp9_findnearmv.h"
#include "vp9/encoder/vp9_lookahead.h"
2010-05-18 17:58:33 +02:00
// Experimental rate control switches
#if CONFIG_ONESHOTQ
#define ONE_SHOT_Q_ESTIMATE 0
#define STRICT_ONE_SHOT_Q 0
#define DISABLE_RC_LONG_TERM_MEM 0
#endif
// #define SPEEDSTATS 1
#if CONFIG_MULTIPLE_ARF
// Set MIN_GF_INTERVAL to 1 for the full decomposition.
#define MIN_GF_INTERVAL 2
#else
2010-05-18 17:58:33 +02:00
#define MIN_GF_INTERVAL 4
#endif
2010-05-18 17:58:33 +02:00
#define DEFAULT_GF_INTERVAL 7
#define KEY_FRAME_CONTEXT 5
#define MAX_MODES 36
2010-05-18 17:58:33 +02:00
#define MIN_THRESHMULT 32
#define MAX_THRESHMULT 512
#define GF_ZEROMV_ZBIN_BOOST 0
#define LF_ZEROMV_ZBIN_BOOST 0
#define MV_ZBIN_BOOST 0
#define SPLIT_MV_ZBIN_BOOST 0
#define INTRA_ZBIN_BOOST 0
2010-05-18 17:58:33 +02:00
typedef struct {
nmv_context nmvc;
int nmvjointcost[MV_JOINTS];
int nmvcosts[2][MV_VALS];
int nmvcosts_hp[2][MV_VALS];
2010-05-18 17:58:33 +02:00
vp9_prob segment_pred_probs[PREDICTION_PROBS];
vp9_prob intra_inter_prob[INTRA_INTER_CONTEXTS];
vp9_prob comp_inter_prob[COMP_INTER_CONTEXTS];
vp9_prob single_ref_prob[REF_CONTEXTS][2];
vp9_prob comp_ref_prob[REF_CONTEXTS];
2010-05-18 17:58:33 +02:00
unsigned char *last_frame_seg_map_copy;
// 0 = Intra, Last, GF, ARF
signed char last_ref_lf_deltas[MAX_REF_LF_DELTAS];
// 0 = ZERO_MV, MV
signed char last_mode_lf_deltas[MAX_MODE_LF_DELTAS];
2010-05-18 17:58:33 +02:00
vp9_coeff_probs_model coef_probs[TX_SIZE_MAX_SB][BLOCK_TYPES];
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
vp9_prob y_mode_prob[4][VP9_INTRA_MODES - 1];
vp9_prob uv_mode_prob[VP9_INTRA_MODES][VP9_INTRA_MODES - 1];
vp9_prob partition_prob[2][NUM_PARTITION_CONTEXTS][PARTITION_TYPES - 1];
vp9_prob switchable_interp_prob[VP9_SWITCHABLE_FILTERS + 1]
[VP9_SWITCHABLE_FILTERS - 1];
int inter_mode_counts[INTER_MODE_CONTEXTS][VP9_INTER_MODES - 1][2];
vp9_prob inter_mode_probs[INTER_MODE_CONTEXTS][VP9_INTER_MODES - 1];
2010-05-18 17:58:33 +02:00
vp9_prob tx_probs_8x8p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 3];
vp9_prob tx_probs_16x16p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 2];
vp9_prob tx_probs_32x32p[TX_SIZE_CONTEXTS][TX_SIZE_MAX_SB - 1];
vp9_prob mbskip_probs[MBSKIP_CONTEXTS];
2010-05-18 17:58:33 +02:00
} CODING_CONTEXT;
typedef struct {
double frame;
double intra_error;
double coded_error;
double sr_coded_error;
double ssim_weighted_pred_err;
double pcnt_inter;
double pcnt_motion;
double pcnt_second_ref;
double pcnt_neutral;
double MVr;
double mvr_abs;
double MVc;
double mvc_abs;
double MVrv;
double MVcv;
double mv_in_out_count;
double new_mv_count;
double duration;
double count;
} FIRSTPASS_STATS;
2010-05-18 17:58:33 +02:00
typedef struct {
int frames_so_far;
double frame_intra_error;
double frame_coded_error;
double frame_pcnt_inter;
double frame_pcnt_motion;
double frame_mvr;
double frame_mvr_abs;
double frame_mvc;
double frame_mvc_abs;
2010-05-18 17:58:33 +02:00
} ONEPASS_FRAMESTATS;
typedef struct {
struct {
int err;
union {
int_mv mv;
MB_PREDICTION_MODE mode;
} m;
} ref[MAX_REF_FRAMES];
2011-10-05 12:26:00 +02:00
} MBGRAPH_MB_STATS;
typedef struct {
MBGRAPH_MB_STATS *mb_stats;
2011-10-05 12:26:00 +02:00
} MBGRAPH_FRAME_STATS;
2010-05-18 17:58:33 +02:00
typedef enum {
THR_ZEROMV,
THR_DC,
THR_NEARESTMV,
THR_NEARMV,
THR_ZEROG,
THR_NEARESTG,
THR_ZEROA,
THR_NEARESTA,
THR_NEARG,
THR_NEARA,
THR_V_PRED,
THR_H_PRED,
THR_D45_PRED,
THR_D135_PRED,
THR_D117_PRED,
THR_D153_PRED,
THR_D27_PRED,
THR_D63_PRED,
THR_TM,
THR_NEWMV,
THR_NEWG,
THR_NEWA,
THR_SPLITMV,
THR_SPLITG,
THR_SPLITA,
THR_B_PRED,
THR_COMP_ZEROLA,
THR_COMP_NEARESTLA,
THR_COMP_NEARLA,
THR_COMP_ZEROGA,
THR_COMP_NEARESTGA,
THR_COMP_NEARGA,
THR_COMP_NEWLA,
THR_COMP_NEWGA,
THR_COMP_SPLITLA,
THR_COMP_SPLITGA,
} THR_MODES;
2010-05-18 17:58:33 +02:00
typedef enum {
DIAMOND = 0,
NSTEP = 1,
HEX = 2
2010-05-18 17:58:33 +02:00
} SEARCH_METHODS;
Tx size selection enhancements (1) Refines the modeling function and uses that to add some speed features. Specifically, intead of using a flag use_largest_txfm as a speed feature, an enum tx_size_search_method is used, of which two of the types are USE_FULL_RD and USE_LARGESTALL. Two other new types are added: USE_LARGESTINTRA (use largest only for intra) USE_LARGESTINTRA_MODELINTER (use largest for intra, and model for inter) (2) Another change is that the framework for deciding transform type is simplified to use a heuristic count based method rather than an rd based method using txfm_cache. In practice the new method is found to work just as well - with derf only -0.01 down. The new method is more compatible with the new framework where certain rd costs are based on full rd and certain others are based on modeled rd or are not computed. In this patch the existing rd based method is still kept for use in the USE_FULL_RD mode. In the other modes, the count based method is used. However the recommendation is to remove it eventually since the benefit is limited, and will remove a lot of complications in the code (3) Finally a bug is fixed with the existing use_largest_txfm speed feature that causes mismatches when the lossless mode and 4x4 WH transform is forced. Results on derf: USE_FULL_RD: +0.03% (due to change in the tables), 0% encode time reduction USE_LARGESTINTRA: -0.21%, 15% encode time reduction (this one is a pretty good compromise) USE_LARGESTINTRA_MODELINTER: -0.98%, 22% encode time reduction (currently the benefit of modeling is limited for txfm size selection, but keeping this enum as a placeholder) . USE_LARGESTALL: -1.05%, 27% encode-time reduction (same as existing use_largest_txfm speed feature). Change-Id: I4d60a5f9ce78fbc90cddf2f97ed91d8bc0d4f936
2013-06-22 01:31:12 +02:00
typedef enum {
USE_FULL_RD = 0,
USE_LARGESTINTRA,
USE_LARGESTINTRA_MODELINTER,
USE_LARGESTALL
} TX_SIZE_SEARCH_METHOD;
typedef struct {
int RD;
SEARCH_METHODS search_method;
int auto_filter;
int recode_loop;
int iterative_sub_pixel;
int half_pixel_search;
int quarter_pixel_search;
int thresh_mult[MAX_MODES];
int max_step_search_steps;
int reduce_first_step_size;
int auto_mv_step_size;
int optimize_coefficients;
int search_best_filter;
int static_segmentation;
int comp_inter_joint_search_thresh;
int adaptive_rd_thresh;
int skip_encode_sb;
int use_lastframe_partitioning;
Tx size selection enhancements (1) Refines the modeling function and uses that to add some speed features. Specifically, intead of using a flag use_largest_txfm as a speed feature, an enum tx_size_search_method is used, of which two of the types are USE_FULL_RD and USE_LARGESTALL. Two other new types are added: USE_LARGESTINTRA (use largest only for intra) USE_LARGESTINTRA_MODELINTER (use largest for intra, and model for inter) (2) Another change is that the framework for deciding transform type is simplified to use a heuristic count based method rather than an rd based method using txfm_cache. In practice the new method is found to work just as well - with derf only -0.01 down. The new method is more compatible with the new framework where certain rd costs are based on full rd and certain others are based on modeled rd or are not computed. In this patch the existing rd based method is still kept for use in the USE_FULL_RD mode. In the other modes, the count based method is used. However the recommendation is to remove it eventually since the benefit is limited, and will remove a lot of complications in the code (3) Finally a bug is fixed with the existing use_largest_txfm speed feature that causes mismatches when the lossless mode and 4x4 WH transform is forced. Results on derf: USE_FULL_RD: +0.03% (due to change in the tables), 0% encode time reduction USE_LARGESTINTRA: -0.21%, 15% encode time reduction (this one is a pretty good compromise) USE_LARGESTINTRA_MODELINTER: -0.98%, 22% encode time reduction (currently the benefit of modeling is limited for txfm size selection, but keeping this enum as a placeholder) . USE_LARGESTALL: -1.05%, 27% encode-time reduction (same as existing use_largest_txfm speed feature). Change-Id: I4d60a5f9ce78fbc90cddf2f97ed91d8bc0d4f936
2013-06-22 01:31:12 +02:00
TX_SIZE_SEARCH_METHOD tx_size_search_method;
int use_8tap_always;
int use_avoid_tested_higherror;
int skip_lots_of_modes;
int adjust_thresholds_by_speed;
int partition_by_variance;
int use_one_partition_size_always;
int less_rectangular_check;
int use_square_partition_only;
BLOCK_SIZE_TYPE always_this_block_size;
int use_partitions_greater_than;
BLOCK_SIZE_TYPE greater_than_block_size;
int use_partitions_less_than;
BLOCK_SIZE_TYPE less_than_block_size;
int adjust_partitioning_from_last_frame;
int last_partitioning_redo_frequency;
int disable_splitmv;
// Search the D27, D63, D117 and D153 modes
// only if the best intra mode so far is one
// of the two directional modes nearest to each.
int conditional_oblique_intramodes;
2010-05-18 17:58:33 +02:00
} SPEED_FEATURES;
enum BlockSize {
BLOCK_4X4,
BLOCK_4X8,
BLOCK_8X4,
BLOCK_8X8,
BLOCK_8X16,
BLOCK_16X8,
BLOCK_16X16,
BLOCK_32X32,
BLOCK_32X16,
BLOCK_16X32,
BLOCK_64X32,
BLOCK_32X64,
BLOCK_64X64,
BLOCK_MAX_SB_SEGMENTS,
};
typedef struct VP9_COMP {
2010-05-18 17:58:33 +02:00
DECLARE_ALIGNED(16, int16_t, y_quant[QINDEX_RANGE][8]);
DECLARE_ALIGNED(16, int16_t, y_quant_shift[QINDEX_RANGE][8]);
DECLARE_ALIGNED(16, int16_t, y_zbin[QINDEX_RANGE][8]);
DECLARE_ALIGNED(16, int16_t, y_round[QINDEX_RANGE][8]);
2010-05-18 17:58:33 +02:00
DECLARE_ALIGNED(16, int16_t, uv_quant[QINDEX_RANGE][8]);
DECLARE_ALIGNED(16, int16_t, uv_quant_shift[QINDEX_RANGE][8]);
DECLARE_ALIGNED(16, int16_t, uv_zbin[QINDEX_RANGE][8]);
DECLARE_ALIGNED(16, int16_t, uv_round[QINDEX_RANGE][8]);
2010-05-18 17:58:33 +02:00
#if CONFIG_ALPHA
DECLARE_ALIGNED(16, int16_t, a_quant[QINDEX_RANGE][8]);
DECLARE_ALIGNED(16, int16_t, a_quant_shift[QINDEX_RANGE][8]);
DECLARE_ALIGNED(16, int16_t, a_zbin[QINDEX_RANGE][8]);
DECLARE_ALIGNED(16, int16_t, a_round[QINDEX_RANGE][8]);
#endif
2010-05-18 17:58:33 +02:00
MACROBLOCK mb;
VP9_COMMON common;
VP9_CONFIG oxcf;
struct lookahead_ctx *lookahead;
struct lookahead_entry *source;
#if CONFIG_MULTIPLE_ARF
struct lookahead_entry *alt_ref_source[NUM_REF_FRAMES];
#else
struct lookahead_entry *alt_ref_source;
#endif
YV12_BUFFER_CONFIG *Source;
YV12_BUFFER_CONFIG *un_scaled_source;
YV12_BUFFER_CONFIG scaled_source;
2010-05-18 17:58:33 +02:00
int source_alt_ref_pending; // frame in src_buffers has been identified to be encoded as an alt ref
int source_alt_ref_active; // an alt ref frame has been encoded and is usable
int is_src_frame_alt_ref; // source of frame to encode is an exact copy of an alt ref frame
2010-05-18 17:58:33 +02:00
int gold_is_last; // golden frame same as last frame ( short circuit gold searches)
int alt_is_last; // Alt reference frame same as last ( short circuit altref search)
int gold_is_alt; // don't do both alt and gold search ( just do gold).
2010-05-18 17:58:33 +02:00
int scaled_ref_idx[3];
int lst_fb_idx;
int gld_fb_idx;
int alt_fb_idx;
#if CONFIG_MULTIPLE_ARF
int alt_ref_fb_idx[NUM_REF_FRAMES - 3];
#endif
int refresh_last_frame;
int refresh_golden_frame;
int refresh_alt_ref_frame;
YV12_BUFFER_CONFIG last_frame_uf;
2010-05-18 17:58:33 +02:00
TOKENEXTRA *tok;
unsigned int tok_count[4][1 << 6];
2010-05-18 17:58:33 +02:00
unsigned int frames_since_key;
unsigned int key_frame_frequency;
unsigned int this_key_frame_forced;
unsigned int next_key_frame_forced;
#if CONFIG_MULTIPLE_ARF
// Position within a frame coding order (including any additional ARF frames).
unsigned int sequence_number;
// Next frame in naturally occurring order that has not yet been coded.
int next_frame_in_order;
#endif
2010-05-18 17:58:33 +02:00
// Ambient reconstruction err target for force key frames
int ambient_err;
unsigned int mode_check_freq[MAX_MODES];
unsigned int mode_test_hit_counts[MAX_MODES];
unsigned int mode_chosen_counts[MAX_MODES];
2010-05-18 17:58:33 +02:00
int rd_thresh_mult[MAX_MODES];
int rd_baseline_thresh[BLOCK_SIZE_TYPES][MAX_MODES];
int rd_threshes[BLOCK_SIZE_TYPES][MAX_MODES];
int rd_thresh_freq_fact[BLOCK_SIZE_TYPES][MAX_MODES];
int64_t rd_comp_pred_diff[NB_PREDICTION_TYPES];
int rd_prediction_type_threshes[4][NB_PREDICTION_TYPES];
unsigned int intra_inter_count[INTRA_INTER_CONTEXTS][2];
unsigned int comp_inter_count[COMP_INTER_CONTEXTS][2];
unsigned int single_ref_count[REF_CONTEXTS][2][2];
unsigned int comp_ref_count[REF_CONTEXTS][2];
// FIXME contextualize
int64_t rd_tx_select_diff[NB_TXFM_MODES];
int rd_tx_select_threshes[4][NB_TXFM_MODES];
2010-05-18 17:58:33 +02:00
int RDMULT;
int RDDIV;
2010-05-18 17:58:33 +02:00
CODING_CONTEXT coding_context;
2010-05-18 17:58:33 +02:00
// Rate targetting variables
int this_frame_target;
int projected_frame_size;
int last_q[2]; // Separate values for Intra/Inter
int last_boosted_qindex; // Last boosted GF/KF/ARF q
2010-05-18 17:58:33 +02:00
double rate_correction_factor;
double key_frame_rate_correction_factor;
double gf_rate_correction_factor;
2010-05-18 17:58:33 +02:00
int frames_till_gf_update_due; // Count down till next GF
2010-05-18 17:58:33 +02:00
int gf_overspend_bits; // Total bits overspent becasue of GF boost (cumulative)
2010-05-18 17:58:33 +02:00
int non_gf_bitrate_adjustment; // Used in the few frames following a GF to recover the extra bits spent in that GF
2010-05-18 17:58:33 +02:00
int kf_overspend_bits; // Extra bits spent on key frames that need to be recovered on inter frames
int kf_bitrate_adjustment; // Current number of bit s to try and recover on each inter frame.
int max_gf_interval;
int baseline_gf_interval;
int active_arnr_frames; // <= cpi->oxcf.arnr_max_frames
int active_arnr_strength; // <= cpi->oxcf.arnr_max_strength
2010-05-18 17:58:33 +02:00
int64_t key_frame_count;
int prior_key_frame_distance[KEY_FRAME_CONTEXT];
int per_frame_bandwidth; // Current section per frame bandwidth target
int av_per_frame_bandwidth; // Average frame size target for clip
int min_frame_bandwidth; // Minimum allocation that should be used for any frame
int inter_frame_target;
double output_frame_rate;
int64_t last_time_stamp_seen;
int64_t last_end_time_stamp_seen;
int64_t first_time_stamp_ever;
2010-05-18 17:58:33 +02:00
int ni_av_qi;
int ni_tot_qi;
int ni_frames;
int avg_frame_qindex;
double tot_q;
double avg_q;
2010-05-18 17:58:33 +02:00
int zbin_mode_boost;
int zbin_mode_boost_enabled;
2010-05-18 17:58:33 +02:00
int64_t total_byte_count;
2010-05-18 17:58:33 +02:00
int buffered_mode;
2010-05-18 17:58:33 +02:00
int buffer_level;
int bits_off_target;
2010-05-18 17:58:33 +02:00
int rolling_target_bits;
int rolling_actual_bits;
2010-05-18 17:58:33 +02:00
int long_rolling_target_bits;
int long_rolling_actual_bits;
2010-05-18 17:58:33 +02:00
int64_t total_actual_bits;
int total_target_vs_actual; // debug stats
2010-05-18 17:58:33 +02:00
int worst_quality;
int active_worst_quality;
int best_quality;
int active_best_quality;
2010-05-18 17:58:33 +02:00
int cq_target_quality;
int y_mode_count[4][VP9_INTRA_MODES];
int y_uv_mode_count[VP9_INTRA_MODES][VP9_INTRA_MODES];
unsigned int partition_count[NUM_PARTITION_CONTEXTS][PARTITION_TYPES];
nmv_context_counts NMVcount;
2010-05-18 17:58:33 +02:00
vp9_coeff_count coef_counts[TX_SIZE_MAX_SB][BLOCK_TYPES];
vp9_coeff_probs_model frame_coef_probs[TX_SIZE_MAX_SB][BLOCK_TYPES];
vp9_coeff_stats frame_branch_ct[TX_SIZE_MAX_SB][BLOCK_TYPES];
32x32 transform for superblocks. This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds code all over the place to wrap that in the bitstream/encoder/decoder/RD. Some implementation notes (these probably need careful review): - token range is extended by 1 bit, since the value range out of this transform is [-16384,16383]. - the coefficients coming out of the FDCT are manually scaled back by 1 bit, or else they won't fit in int16_t (they are 17 bits). Because of this, the RD error scoring does not right-shift the MSE score by two (unlike for 4x4/8x8/16x16). - to compensate for this loss in precision, the quantizer is halved also. This is currently a little hacky. - FDCT and IDCT is double-only right now. Needs a fixed-point impl. - There are no default probabilities for the 32x32 transform yet; I'm simply using the 16x16 luma ones. A future commit will add newly generated probabilities for all transforms. - No ADST version. I don't think we'll add one for this level; if an ADST is desired, transform-size selection can scale back to 16x16 or lower, and use an ADST at that level. Additional notes specific to Debargha's DWT/DCT hybrid: - coefficient scale is different for the top/left 16x16 (DCT-over-DWT) block than for the rest (DWT pixel differences) of the block. Therefore, RD error scoring isn't easily scalable between coefficient and pixel domain. Thus, unfortunately, we need to compute the RD distortion in the pixel domain until we figure out how to scale these appropriately. Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
2012-12-07 23:45:05 +01:00
int gfu_boost;
int last_boost;
int kf_boost;
int kf_zeromotion_pct;
2010-05-18 17:58:33 +02:00
int64_t target_bandwidth;
struct vpx_codec_pkt_list *output_pkt_list;
2010-05-18 17:58:33 +02:00
#if 0
// Experimental code for lagged and one pass
ONEPASS_FRAMESTATS one_pass_frame_stats[MAX_LAG_BUFFERS];
int one_pass_frame_index;
2010-05-18 17:58:33 +02:00
#endif
MBGRAPH_FRAME_STATS mbgraph_stats[MAX_LAG_BUFFERS];
int mbgraph_n_frames; // number of frames filled in the above
int static_mb_pct; // % forced skip mbs by segmentation
int seg0_progress, seg0_idx, seg0_cnt;
int decimation_factor;
int decimation_count;
// for real time encoding
int avg_encode_time; // microsecond
int avg_pick_mode_time; // microsecond
int speed;
unsigned int cpu_freq; // Mhz
int compressor_speed;
int interquantizer;
int goldfreq;
int auto_worst_q;
int cpu_used;
int pass;
2010-05-18 17:58:33 +02:00
vp9_prob last_skip_false_probs[3][MBSKIP_CONTEXTS];
int last_skip_probs_q[3];
2010-05-18 17:58:33 +02:00
int ref_frame_flags;
2010-05-18 17:58:33 +02:00
SPEED_FEATURES sf;
int error_bins[1024];
2010-05-18 17:58:33 +02:00
unsigned int max_mv_magnitude;
// Data used for real time conferencing mode to help determine if it would be good to update the gf
int inter_zz_count;
int gf_bad_count;
int gf_update_recommended;
2010-05-18 17:58:33 +02:00
unsigned char *segmentation_map;
// segment threashold for encode breakout
int segment_encode_breakout[MAX_MB_SEGMENTS];
2010-05-18 17:58:33 +02:00
unsigned char *active_map;
unsigned int active_map_enabled;
2010-05-18 17:58:33 +02:00
fractional_mv_step_fp *find_fractional_mv_step;
vp9_full_search_fn_t full_search_sad;
vp9_refining_search_fn_t refining_search_sad;
vp9_diamond_search_fn_t diamond_search_sad;
vp9_variance_fn_ptr_t fn_ptr[BLOCK_MAX_SB_SEGMENTS];
uint64_t time_receive_data;
uint64_t time_compress_data;
uint64_t time_pick_lpf;
uint64_t time_encode_mb_row;
2010-05-18 17:58:33 +02:00
struct twopass_rc {
unsigned int section_intra_rating;
unsigned int next_iiratio;
unsigned int this_iiratio;
FIRSTPASS_STATS total_stats;
FIRSTPASS_STATS this_frame_stats;
FIRSTPASS_STATS *stats_in, *stats_in_end, *stats_in_start;
FIRSTPASS_STATS total_left_stats;
int first_pass_done;
int64_t bits_left;
int64_t clip_bits_total;
double avg_iiratio;
double modified_error_total;
double modified_error_used;
double modified_error_left;
double kf_intra_err_min;
double gf_intra_err_min;
int frames_to_key;
int maxq_max_limit;
int maxq_min_limit;
int static_scene_max_gf_interval;
int kf_bits;
// Remaining error from uncoded frames in a gf group. Two pass use only
int64_t gf_group_error_left;
// Projected total bits available for a key frame group of frames
int64_t kf_group_bits;
// Error score of frames still to be coded in kf group
int64_t kf_group_error_left;
// Projected Bits available for a group of frames including 1 GF or ARF
int64_t gf_group_bits;
// Bits for the golden frame or ARF - 2 pass only
int gf_bits;
int alt_extra_bits;
int sr_update_lag;
double est_max_qcorrection_factor;
} twopass;
2010-05-18 17:58:33 +02:00
YV12_BUFFER_CONFIG alt_ref_buffer;
YV12_BUFFER_CONFIG *frames[MAX_LAG_BUFFERS];
int fixed_divide[512];
2010-05-18 17:58:33 +02:00
#if CONFIG_INTERNAL_STATS
int count;
double total_y;
double total_u;
double total_v;
double total;
double total_sq_error;
double totalp_y;
double totalp_u;
double totalp_v;
double totalp;
double total_sq_error2;
int bytes;
double summed_quality;
double summed_weights;
double summedp_quality;
double summedp_weights;
unsigned int tot_recode_hits;
double total_ssimg_y;
double total_ssimg_u;
double total_ssimg_v;
double total_ssimg_all;
int b_calculate_ssimg;
2010-05-18 17:58:33 +02:00
#endif
int b_calculate_psnr;
// Per MB activity measurement
unsigned int activity_avg;
unsigned int *mb_activity_map;
int *mb_norm_activity_map;
int output_partition;
New ways of passing encoded data between encoder and decoder. With this commit frames can be received partition-by-partition from the encoder and passed partition-by-partition to the decoder. At the encoder-side this makes it easier to split encoded frames at partition boundaries, useful when packetizing frames. When VPX_CODEC_USE_OUTPUT_PARTITION is enabled, several VPX_CODEC_CX_FRAME_PKT packets will be returned from vpx_codec_get_cx_data(), containing one partition each. The partition_id (starting at 0) specifies the decoding order of the partitions. All partitions but the last has the VPX_FRAME_IS_FRAGMENT flag set. At the decoder this opens up the possibility of decoding partition N even though partition N-1 was lost (given that independent partitioning has been enabled in the encoder) if more info about the missing parts of the stream is available through external signaling. Each partition is passed to the decoder through the vpx_codec_decode() function, with the data pointer pointing to the start of the partition, and with data_sz equal to the size of the partition. Missing partitions can be signaled to the decoder by setting data != NULL and data_sz = 0. When all partitions have been given to the decoder "end of data" should be signaled by calling vpx_codec_decode() with data = NULL and data_sz = 0. The first partition is the first partition according to the VP8 bitstream + the uncompressed data chunk + DCT address offsets if multiple residual partitions are used. Change-Id: I5bc0682b9e4112e0db77904755c694c3c7ac6e74
2011-06-13 16:42:27 +02:00
/* force next frame to intra when kf_auto says so */
int force_next_frame_intra;
int droppable;
int dummy_packing; /* flag to indicate if packing is dummy */
unsigned int switchable_interp_count[VP9_SWITCHABLE_FILTERS + 1]
[VP9_SWITCHABLE_FILTERS];
unsigned int best_switchable_interp_count[VP9_SWITCHABLE_FILTERS];
Tx size selection enhancements (1) Refines the modeling function and uses that to add some speed features. Specifically, intead of using a flag use_largest_txfm as a speed feature, an enum tx_size_search_method is used, of which two of the types are USE_FULL_RD and USE_LARGESTALL. Two other new types are added: USE_LARGESTINTRA (use largest only for intra) USE_LARGESTINTRA_MODELINTER (use largest for intra, and model for inter) (2) Another change is that the framework for deciding transform type is simplified to use a heuristic count based method rather than an rd based method using txfm_cache. In practice the new method is found to work just as well - with derf only -0.01 down. The new method is more compatible with the new framework where certain rd costs are based on full rd and certain others are based on modeled rd or are not computed. In this patch the existing rd based method is still kept for use in the USE_FULL_RD mode. In the other modes, the count based method is used. However the recommendation is to remove it eventually since the benefit is limited, and will remove a lot of complications in the code (3) Finally a bug is fixed with the existing use_largest_txfm speed feature that causes mismatches when the lossless mode and 4x4 WH transform is forced. Results on derf: USE_FULL_RD: +0.03% (due to change in the tables), 0% encode time reduction USE_LARGESTINTRA: -0.21%, 15% encode time reduction (this one is a pretty good compromise) USE_LARGESTINTRA_MODELINTER: -0.98%, 22% encode time reduction (currently the benefit of modeling is limited for txfm size selection, but keeping this enum as a placeholder) . USE_LARGESTALL: -1.05%, 27% encode-time reduction (same as existing use_largest_txfm speed feature). Change-Id: I4d60a5f9ce78fbc90cddf2f97ed91d8bc0d4f936
2013-06-22 01:31:12 +02:00
unsigned int txfm_stepdown_count[TX_SIZE_MAX_SB];
int initial_width;
int initial_height;
#if CONFIG_MULTIPLE_ARF
// ARF tracking variables.
int multi_arf_enabled;
unsigned int frame_coding_order_period;
unsigned int new_frame_coding_order_period;
int frame_coding_order[MAX_LAG_BUFFERS * 2];
int arf_buffer_idx[MAX_LAG_BUFFERS * 3 / 2];
int arf_weight[MAX_LAG_BUFFERS];
int arf_buffered;
int this_frame_weight;
int max_arf_level;
#endif
#ifdef ENTROPY_STATS
int64_t mv_ref_stats[INTER_MODE_CONTEXTS][VP9_INTER_MODES - 1][2];
#endif
} VP9_COMP;
2010-05-18 17:58:33 +02:00
static int get_ref_frame_idx(VP9_COMP *cpi, MV_REFERENCE_FRAME ref_frame) {
if (ref_frame == LAST_FRAME) {
return cpi->lst_fb_idx;
} else if (ref_frame == GOLDEN_FRAME) {
return cpi->gld_fb_idx;
} else {
return cpi->alt_fb_idx;
}
}
void vp9_encode_frame(VP9_COMP *cpi);
2010-05-18 17:58:33 +02:00
void vp9_pack_bitstream(VP9_COMP *cpi, unsigned char *dest,
unsigned long *size);
2010-05-18 17:58:33 +02:00
void vp9_activity_masking(VP9_COMP *cpi, MACROBLOCK *x);
void vp9_set_speed_features(VP9_COMP *cpi);
extern int vp9_calc_ss_err(YV12_BUFFER_CONFIG *source,
YV12_BUFFER_CONFIG *dest);
extern void vp9_alloc_compressor_data(VP9_COMP *cpi);
#endif // VP9_ENCODER_VP9_ONYX_INT_H_