generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
0daadeb60c
vpx/vp9/encoder/vp9_speed_features.c
Jingning Han 0daadeb60c Enable motion field based mode seach skip 
This commit allows the encoder to check the above and left neighbor
blocks' reference frames and motion vectors. If they are all
consistent, skip checking the NEARMV and ZEROMV modes. This is
enabled in speed 3. The coding performance is improved:

pedestrian area 1080p at 2000 kbps,
from  74773 b/f, 41.101 dB, 198064 ms
to    74795 b/f, 41.099 dB, 193078 ms

park joy 1080p at 15000 kbps,
from 290727 b/f, 30.640 dB, 609113 ms
to   290558 b/f, 30.630 dB, 592815 ms

Overall compression performance of speed 3 is changed
derf  -0.171%
stdhd -0.168%

Change-Id: I8d47dd543a5f90d7a1c583f74035b926b6704b95
2014-08-13 12:15:13 -07:00

436 lines
15 KiB
C
Raw Blame History

/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
*
* 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.
*/
#include <limits.h>
#include "vp9/encoder/vp9_encoder.h"
#include "vp9/encoder/vp9_speed_features.h"
enum {
INTRA_ALL = (1 << DC_PRED) |
(1 << V_PRED) | (1 << H_PRED) |
(1 << D45_PRED) | (1 << D135_PRED) |
(1 << D117_PRED) | (1 << D153_PRED) |
(1 << D207_PRED) | (1 << D63_PRED) |
(1 << TM_PRED),
INTRA_DC = (1 << DC_PRED),
INTRA_DC_TM = (1 << DC_PRED) | (1 << TM_PRED),
INTRA_DC_H_V = (1 << DC_PRED) | (1 << V_PRED) | (1 << H_PRED),
INTRA_DC_TM_H_V = (1 << DC_PRED) | (1 << TM_PRED) | (1 << V_PRED) |
(1 << H_PRED)
};
enum {
INTER_ALL = (1 << NEARESTMV) | (1 << NEARMV) | (1 << ZEROMV) | (1 << NEWMV),
INTER_NEAREST = (1 << NEARESTMV),
INTER_NEAREST_NEAR_NEW = (1 << NEARESTMV) | (1 << NEARMV) | (1 << NEWMV)
};
enum {
DISABLE_ALL_INTER_SPLIT = (1 << THR_COMP_GA) |
(1 << THR_COMP_LA) |
(1 << THR_ALTR) |
(1 << THR_GOLD) |
(1 << THR_LAST),
DISABLE_ALL_SPLIT = (1 << THR_INTRA) | DISABLE_ALL_INTER_SPLIT,
DISABLE_COMPOUND_SPLIT = (1 << THR_COMP_GA) | (1 << THR_COMP_LA),
LAST_AND_INTRA_SPLIT_ONLY = (1 << THR_COMP_GA) |
(1 << THR_COMP_LA) |
(1 << THR_ALTR) |
(1 << THR_GOLD)
};
static void set_good_speed_feature(VP9_COMP *cpi, VP9_COMMON *cm,
SPEED_FEATURES *sf, int speed) {
sf->adaptive_rd_thresh = 1;
sf->recode_loop = (speed < 1) ? ALLOW_RECODE : ALLOW_RECODE_KFMAXBW;
sf->allow_skip_recode = 1;
if (speed >= 1) {
sf->use_square_partition_only = !frame_is_intra_only(cm);
sf->less_rectangular_check = 1;
sf->tx_size_search_method = frame_is_boosted(cpi) ? USE_FULL_RD
: USE_LARGESTALL;
if (MIN(cm->width, cm->height) >= 720)
sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
: DISABLE_ALL_INTER_SPLIT;
else
sf->disable_split_mask = DISABLE_COMPOUND_SPLIT;
sf->use_rd_breakout = 1;
sf->adaptive_motion_search = 1;
sf->mv.auto_mv_step_size = 1;
sf->adaptive_rd_thresh = 2;
sf->mv.subpel_iters_per_step = 1;
sf->mode_skip_start = 10;
sf->adaptive_pred_interp_filter = 1;
sf->recode_loop = ALLOW_RECODE_KFARFGF;
sf->intra_y_mode_mask[TX_32X32] = INTRA_DC_H_V;
sf->intra_uv_mode_mask[TX_32X32] = INTRA_DC_H_V;
sf->intra_y_mode_mask[TX_16X16] = INTRA_DC_H_V;
sf->intra_uv_mode_mask[TX_16X16] = INTRA_DC_H_V;
}
if (speed >= 2) {
if (MIN(cm->width, cm->height) >= 720) {
sf->lf_motion_threshold = LOW_MOTION_THRESHOLD;
sf->last_partitioning_redo_frequency = 3;
sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
: DISABLE_ALL_INTER_SPLIT;
sf->adaptive_pred_interp_filter = 0;
} else {
sf->disable_split_mask = LAST_AND_INTRA_SPLIT_ONLY;
sf->last_partitioning_redo_frequency = 2;
sf->lf_motion_threshold = NO_MOTION_THRESHOLD;
}
sf->reference_masking = 1;
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
FLAG_SKIP_INTRA_LOWVAR;
sf->disable_filter_search_var_thresh = 100;
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
sf->auto_min_max_partition_size = RELAXED_NEIGHBORING_MIN_MAX;
sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_LOW_MOTION;
sf->adjust_partitioning_from_last_frame = 1;
}
if (speed >= 3) {
sf->tx_size_search_method = frame_is_intra_only(cm) ? USE_FULL_RD
: USE_LARGESTALL;
if (MIN(cm->width, cm->height) >= 720) {
sf->disable_split_mask = DISABLE_ALL_SPLIT;
sf->cb_partition_search = frame_is_boosted(cpi) ? 0 : 1;
} else {
sf->disable_split_mask = DISABLE_ALL_INTER_SPLIT;
}
sf->adaptive_pred_interp_filter = 0;
sf->cb_pred_filter_search = 1;
sf->motion_field_mode_search = frame_is_boosted(cpi) ? 0 : 1;
sf->lf_motion_threshold = LOW_MOTION_THRESHOLD;
sf->last_partitioning_redo_frequency = 3;
sf->recode_loop = ALLOW_RECODE_KFMAXBW;
sf->adaptive_rd_thresh = 3;
sf->mode_skip_start = 6;
sf->use_fast_coef_updates = ONE_LOOP_REDUCED;
sf->use_fast_coef_costing = 1;
}
if (speed >= 4) {
sf->use_square_partition_only = 1;
sf->tx_size_search_method = USE_LARGESTALL;
sf->disable_split_mask = DISABLE_ALL_SPLIT;
sf->adaptive_rd_thresh = 4;
sf->mode_search_skip_flags |= FLAG_SKIP_COMP_REFMISMATCH |
FLAG_EARLY_TERMINATE;
sf->disable_filter_search_var_thresh = 200;
sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_ALL;
sf->use_lp32x32fdct = 1;
}
if (speed >= 5) {
int i;
sf->partition_search_type = FIXED_PARTITION;
sf->optimize_coefficients = 0;
sf->mv.search_method = HEX;
sf->disable_filter_search_var_thresh = 500;
for (i = 0; i < TX_SIZES; ++i) {
sf->intra_y_mode_mask[i] = INTRA_DC;
sf->intra_uv_mode_mask[i] = INTRA_DC;
}
cpi->allow_encode_breakout = ENCODE_BREAKOUT_ENABLED;
}
if (speed >= 6) {
sf->mv.reduce_first_step_size = 1;
}
}
static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf,
int speed, vp9e_tune_content content) {
VP9_COMMON *const cm = &cpi->common;
const int frames_since_key =
cm->frame_type == KEY_FRAME ? 0 : cpi->rc.frames_since_key;
sf->static_segmentation = 0;
sf->adaptive_rd_thresh = 1;
sf->use_fast_coef_costing = 1;
if (speed >= 1) {
sf->use_square_partition_only = !frame_is_intra_only(cm);
sf->less_rectangular_check = 1;
sf->tx_size_search_method = frame_is_intra_only(cm) ? USE_FULL_RD
: USE_LARGESTALL;
if (MIN(cm->width, cm->height) >= 720)
sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
: DISABLE_ALL_INTER_SPLIT;
else
sf->disable_split_mask = DISABLE_COMPOUND_SPLIT;
sf->use_rd_breakout = 1;
sf->adaptive_motion_search = 1;
sf->adaptive_pred_interp_filter = 1;
sf->mv.auto_mv_step_size = 1;
sf->adaptive_rd_thresh = 2;
sf->intra_y_mode_mask[TX_32X32] = INTRA_DC_H_V;
sf->intra_uv_mode_mask[TX_32X32] = INTRA_DC_H_V;
sf->intra_uv_mode_mask[TX_16X16] = INTRA_DC_H_V;
}
if (speed >= 2) {
if (MIN(cm->width, cm->height) >= 720)
sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
: DISABLE_ALL_INTER_SPLIT;
else
sf->disable_split_mask = LAST_AND_INTRA_SPLIT_ONLY;
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
FLAG_SKIP_INTRA_LOWVAR;
sf->adaptive_pred_interp_filter = 2;
sf->reference_masking = 1;
sf->disable_filter_search_var_thresh = 50;
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
sf->auto_min_max_partition_size = RELAXED_NEIGHBORING_MIN_MAX;
sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_LOW_MOTION;
sf->lf_motion_threshold = LOW_MOTION_THRESHOLD;
sf->adjust_partitioning_from_last_frame = 1;
sf->last_partitioning_redo_frequency = 3;
sf->use_lp32x32fdct = 1;
sf->mode_skip_start = 11;
sf->intra_y_mode_mask[TX_16X16] = INTRA_DC_H_V;
}
if (speed >= 3) {
sf->use_square_partition_only = 1;
sf->disable_filter_search_var_thresh = 100;
sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_ALL;
sf->constrain_copy_partition = 1;
sf->use_uv_intra_rd_estimate = 1;
sf->skip_encode_sb = 1;
sf->mv.subpel_iters_per_step = 1;
sf->use_fast_coef_updates = ONE_LOOP_REDUCED;
sf->adaptive_rd_thresh = 4;
sf->mode_skip_start = 6;
sf->allow_skip_recode = 0;
sf->optimize_coefficients = 0;
sf->disable_split_mask = DISABLE_ALL_SPLIT;
sf->lpf_pick = LPF_PICK_FROM_Q;
}
if (speed >= 4) {
int i;
sf->last_partitioning_redo_frequency = 4;
sf->adaptive_rd_thresh = 5;
sf->use_fast_coef_costing = 0;
sf->auto_min_max_partition_size = STRICT_NEIGHBORING_MIN_MAX;
sf->adjust_partitioning_from_last_frame =
cm->last_frame_type != cm->frame_type || (0 ==
(frames_since_key + 1) % sf->last_partitioning_redo_frequency);
sf->mv.subpel_force_stop = 1;
for (i = 0; i < TX_SIZES; i++) {
sf->intra_y_mode_mask[i] = INTRA_DC_H_V;
sf->intra_uv_mode_mask[i] = INTRA_DC;
}
sf->intra_y_mode_mask[TX_32X32] = INTRA_DC;
sf->frame_parameter_update = 0;
sf->mv.search_method = FAST_HEX;
sf->inter_mode_mask[BLOCK_32X32] = INTER_NEAREST_NEAR_NEW;
sf->inter_mode_mask[BLOCK_32X64] = INTER_NEAREST;
sf->inter_mode_mask[BLOCK_64X32] = INTER_NEAREST;
sf->inter_mode_mask[BLOCK_64X64] = INTER_NEAREST;
sf->max_intra_bsize = BLOCK_32X32;
sf->allow_skip_recode = 1;
}
if (speed >= 5) {
sf->use_quant_fp = cm->frame_type == KEY_FRAME ? 0 : 1;
sf->auto_min_max_partition_size = (cm->frame_type == KEY_FRAME) ?
RELAXED_NEIGHBORING_MIN_MAX : STRICT_NEIGHBORING_MIN_MAX;
sf->max_partition_size = BLOCK_32X32;
sf->min_partition_size = BLOCK_8X8;
sf->partition_check =
(frames_since_key % sf->last_partitioning_redo_frequency == 1);
sf->force_frame_boost = cm->frame_type == KEY_FRAME ||
(frames_since_key %
(sf->last_partitioning_redo_frequency << 1) == 1);
sf->max_delta_qindex = (cm->frame_type == KEY_FRAME) ? 20 : 15;
sf->partition_search_type = REFERENCE_PARTITION;
sf->use_nonrd_pick_mode = 1;
sf->allow_skip_recode = 0;
}
if (speed >= 6) {
if (content == VP9E_CONTENT_SCREEN) {
int i;
// Allow fancy modes at all sizes since SOURCE_VAR_BASED_PARTITION is used
for (i = 0; i < BLOCK_SIZES; ++i)
sf->inter_mode_mask[i] = INTER_ALL;
}
// Adaptively switch between SOURCE_VAR_BASED_PARTITION and FIXED_PARTITION.
sf->partition_search_type = SOURCE_VAR_BASED_PARTITION;
sf->search_type_check_frequency = 50;
sf->tx_size_search_method = (cm->frame_type == KEY_FRAME) ?
USE_LARGESTALL : USE_TX_8X8;
sf->max_intra_bsize = BLOCK_8X8;
// This feature is only enabled when partition search is disabled.
sf->reuse_inter_pred_sby = 1;
// Increase mode checking threshold for NEWMV.
sf->elevate_newmv_thresh = 2000;
sf->mv.reduce_first_step_size = 1;
}
if (speed >= 7) {
sf->mv.search_method = FAST_DIAMOND;
sf->mv.fullpel_search_step_param = 10;
sf->lpf_pick = LPF_PICK_MINIMAL_LPF;
sf->encode_breakout_thresh = (MIN(cm->width, cm->height) >= 720) ?
800 : 300;
sf->elevate_newmv_thresh = 2500;
}
if (speed >= 12) {
sf->elevate_newmv_thresh = 4000;
sf->mv.subpel_force_stop = 2;
}
if (speed >= 13) {
int i;
sf->max_intra_bsize = BLOCK_32X32;
for (i = 0; i < BLOCK_SIZES; ++i)
sf->inter_mode_mask[i] = INTER_NEAREST;
}
}
void vp9_set_speed_features(VP9_COMP *cpi) {
SPEED_FEATURES *const sf = &cpi->sf;
VP9_COMMON *const cm = &cpi->common;
const VP9EncoderConfig *const oxcf = &cpi->oxcf;
int i;
// best quality defaults
sf->frame_parameter_update = 1;
sf->mv.search_method = NSTEP;
sf->recode_loop = ALLOW_RECODE;
sf->mv.subpel_search_method = SUBPEL_TREE;
sf->mv.subpel_iters_per_step = 2;
sf->mv.subpel_force_stop = 0;
sf->optimize_coefficients = !is_lossless_requested(&cpi->oxcf);
sf->mv.reduce_first_step_size = 0;
sf->mv.auto_mv_step_size = 0;
sf->mv.fullpel_search_step_param = 6;
sf->comp_inter_joint_search_thresh = BLOCK_4X4;
sf->adaptive_rd_thresh = 0;
sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_OFF;
sf->tx_size_search_method = USE_FULL_RD;
sf->use_lp32x32fdct = 0;
sf->adaptive_motion_search = 0;
sf->adaptive_pred_interp_filter = 0;
sf->cb_pred_filter_search = 0;
sf->cb_partition_search = 0;
sf->motion_field_mode_search = 0;
sf->use_quant_fp = 0;
sf->reference_masking = 0;
sf->partition_search_type = SEARCH_PARTITION;
sf->less_rectangular_check = 0;
sf->use_square_partition_only = 0;
sf->auto_min_max_partition_size = NOT_IN_USE;
sf->max_partition_size = BLOCK_64X64;
sf->min_partition_size = BLOCK_4X4;
sf->adjust_partitioning_from_last_frame = 0;
sf->last_partitioning_redo_frequency = 4;
sf->constrain_copy_partition = 0;
sf->disable_split_mask = 0;
sf->mode_search_skip_flags = 0;
sf->force_frame_boost = 0;
sf->max_delta_qindex = 0;
sf->disable_split_var_thresh = 0;
sf->disable_filter_search_var_thresh = 0;
for (i = 0; i < TX_SIZES; i++) {
sf->intra_y_mode_mask[i] = INTRA_ALL;
sf->intra_uv_mode_mask[i] = INTRA_ALL;
}
sf->use_rd_breakout = 0;
sf->skip_encode_sb = 0;
sf->use_uv_intra_rd_estimate = 0;
sf->allow_skip_recode = 0;
sf->lpf_pick = LPF_PICK_FROM_FULL_IMAGE;
sf->use_fast_coef_updates = TWO_LOOP;
sf->use_fast_coef_costing = 0;
sf->mode_skip_start = MAX_MODES; // Mode index at which mode skip mask set
sf->use_nonrd_pick_mode = 0;
for (i = 0; i < BLOCK_SIZES; ++i)
sf->inter_mode_mask[i] = INTER_ALL;
sf->max_intra_bsize = BLOCK_64X64;
sf->reuse_inter_pred_sby = 0;
// This setting only takes effect when partition_search_type is set
// to FIXED_PARTITION.
sf->always_this_block_size = BLOCK_16X16;
sf->search_type_check_frequency = 50;
sf->encode_breakout_thresh = 0;
sf->elevate_newmv_thresh = 0;
// Recode loop tolerence %.
sf->recode_tolerance = 25;
sf->default_interp_filter = SWITCHABLE;
switch (oxcf->mode) {
case ONE_PASS_BEST:
case TWO_PASS_SECOND_BEST: // This is the best quality mode.
cpi->diamond_search_sad = vp9_full_range_search;
break;
case TWO_PASS_FIRST:
case ONE_PASS_GOOD:
case TWO_PASS_SECOND_GOOD:
set_good_speed_feature(cpi, cm, sf, oxcf->speed);
break;
case REALTIME:
set_rt_speed_feature(cpi, sf, oxcf->speed, oxcf->content);
break;
}
// Slow quant, dct and trellis not worthwhile for first pass
// so make sure they are always turned off.
if (oxcf->pass == 1)
sf->optimize_coefficients = 0;
// No recode for 1 pass.
if (oxcf->pass == 0) {
sf->recode_loop = DISALLOW_RECODE;
sf->optimize_coefficients = 0;
}
if (sf->mv.subpel_search_method == SUBPEL_TREE) {
cpi->find_fractional_mv_step = vp9_find_best_sub_pixel_tree;
}
cpi->mb.optimize = sf->optimize_coefficients == 1 && oxcf->pass != 1;
if (sf->disable_split_mask == DISABLE_ALL_SPLIT)
sf->adaptive_pred_interp_filter = 0;
if (!cpi->oxcf.frame_periodic_boost) {
sf->max_delta_qindex = 0;
}
if (cpi->encode_breakout && oxcf->mode == REALTIME &&
sf->encode_breakout_thresh > cpi->encode_breakout)
cpi->encode_breakout = sf->encode_breakout_thresh;
}
Reference in New Issue View Git Blame Copy Permalink