From 6253cc9279652fd7550bc103f85bb45525ff20f1 Mon Sep 17 00:00:00 2001
From: Paul Wilkins <paulwilkins@google.com>
Date: Tue, 1 Oct 2013 16:57:18 +0100
Subject: [PATCH] Speed setting review.
Substantial reworking of the speed vs quality trade offs for
speed 1 and 2.
In this patch I am attempting to freeze the "quality" meaning of
speeds 1 and 2 relative to speed 0 so that in future we can
better evaluate progress.
I am targeting :
Speed 1 quality ~-5% vs speed 0.
Speed 2 quality ~-10% vs speed 0
It is inevitable that quality will still fluctuate a little as we adjust
settings and add new features, but we will attempt to keep as
close as possible to these values. Above speed 2 things will remain
a bit more fluid for now.
In this patch speed 1 is approximately 4-5x as fast as speed 0. This
is similar to before but the quality hit is a lot less. Likewise speed 2
is approximately 2x as fast as speed 1 but is similar in quality to the
previous speed 1 configuration.
Also slight change to behavior of FLAG_EARLY_TERMINATE to insure
all reference frames get at least one rd test. Important for very low
variance regions.
WIP :- Added a new speed level with old speed 4 becoming speed 5.
Speed 3 and 4 tradeoffs still WIP
Change-Id: Ic7a38dd7b5b63ab1501f9352411972f480ac6264
---
vp9/encoder/vp9_encodeframe.c | 4 +-
vp9/encoder/vp9_onyx_if.c | 107 +++++++++++++++++-----------------
vp9/encoder/vp9_onyx_int.h | 10 +++-
vp9/encoder/vp9_rdopt.c | 5 +-
4 files changed, 70 insertions(+), 56 deletions(-)
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 0c1c30d681..957c57793f 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -1781,7 +1781,9 @@ static void encode_sb_row(VP9_COMP *cpi, int mi_row, TOKENEXTRA **tp,
|| cpi->common.show_frame == 0
|| cpi->common.frame_type == KEY_FRAME
|| cpi->is_src_frame_alt_ref
- || sb_has_motion(cpi, prev_mi_8x8)) {
+ || ((cpi->sf.use_lastframe_partitioning ==
+ LAST_FRAME_PARTITION_LOW_MOTION) &&
+ sb_has_motion(cpi, prev_mi_8x8))) {
// If required set upper and lower partition size limits
if (cpi->sf.auto_min_max_partition_size) {
set_offsets(cpi, mi_row, mi_col, BLOCK_64X64);
diff --git a/vp9/encoder/vp9_onyx_if.c b/vp9/encoder/vp9_onyx_if.c
index acad8e161a..f7d9869962 100644
--- a/vp9/encoder/vp9_onyx_if.c
+++ b/vp9/encoder/vp9_onyx_if.c
@@ -714,7 +714,7 @@ void vp9_set_speed_features(VP9_COMP *cpi) {
sf->max_step_search_steps = MAX_MVSEARCH_STEPS;
sf->comp_inter_joint_search_thresh = BLOCK_4X4;
sf->adaptive_rd_thresh = 0;
- sf->use_lastframe_partitioning = 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;
@@ -761,87 +761,88 @@ void vp9_set_speed_features(VP9_COMP *cpi) {
sf->static_segmentation = 0;
#endif
sf->use_avoid_tested_higherror = 1;
- sf->adaptive_rd_thresh = MIN((speed + 1), 4);
+ sf->adaptive_rd_thresh = 1;
if (speed == 1) {
- sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
- sf->less_rectangular_check = 1;
+ sf->use_square_partition_only = !(cpi->common.frame_type == KEY_FRAME ||
+ cpi->common.intra_only);
sf->tx_size_search_method = ((cpi->common.frame_type == KEY_FRAME ||
- cpi->common.intra_only ||
- cpi->common.show_frame == 0) ?
- USE_FULL_RD :
- USE_LARGESTALL);
+ cpi->common.intra_only)
+ ? USE_FULL_RD : USE_LARGESTALL);
+ sf->disable_splitmv =
+ (MIN(cpi->common.width, cpi->common.height) >= 720)? 1 : 0;
+
+ sf->use_rd_breakout = 1;
+ sf->adaptive_motion_search = 1;
+ sf->auto_mv_step_size = 1;
+ sf->adaptive_rd_thresh = 2;
+ }
+ if (speed == 2) {
sf->use_square_partition_only = !(cpi->common.frame_type == KEY_FRAME ||
- cpi->common.intra_only ||
- cpi->common.show_frame == 0);
+ cpi->common.intra_only);
+ sf->tx_size_search_method = ((cpi->common.frame_type == KEY_FRAME ||
+ cpi->common.intra_only)
+ ? USE_FULL_RD : USE_LARGESTALL);
sf->disable_splitmv =
- (MIN(cpi->common.width, cpi->common.height) >= 720)? 1 : 0;
+ (MIN(cpi->common.width, cpi->common.height) >= 720)? 1 : 0;
+
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
FLAG_SKIP_INTRA_LOWVAR;
- sf->use_uv_intra_rd_estimate = 1;
+
sf->use_rd_breakout = 1;
- sf->skip_encode_sb = 1;
- sf->use_lp32x32fdct = 1;
sf->adaptive_motion_search = 1;
sf->auto_mv_step_size = 1;
- sf->auto_min_max_partition_size = 1;
- // FIXME(jingning): temporarily turn off disable_split_var_thresh
- // during refactoring process. will get this back after finishing
- // the main framework of partition search type.
- sf->disable_split_var_thresh = 0;
sf->disable_filter_search_var_thresh = 16;
+ sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
+
+ sf->auto_min_max_partition_size = 1;
+ sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_LOW_MOTION;
+ sf->adjust_partitioning_from_last_frame = 1;
+ sf->last_partitioning_redo_frequency = 3;
- sf->intra_y_mode_mask = INTRA_DC_TM_H_V;
- sf->intra_uv_mode_mask = INTRA_DC_TM_H_V;
- sf->use_fast_coef_updates = 1;
+ sf->adaptive_rd_thresh = 2;
sf->mode_skip_start = 11;
}
- if (speed == 2) {
- sf->less_rectangular_check = 1;
+ if (speed == 3) {
sf->use_square_partition_only = 1;
- sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
- sf->use_lastframe_partitioning = 1;
- sf->adjust_partitioning_from_last_frame = 1;
- sf->last_partitioning_redo_frequency = 3;
- sf->tx_size_search_method = ((cpi->common.frame_type == KEY_FRAME ||
- cpi->common.intra_only ||
- cpi->common.show_frame == 0) ?
- USE_FULL_RD :
- USE_LARGESTALL);
+ sf->tx_size_search_method = USE_LARGESTALL;
+ sf->disable_splitmv =
+ (MIN(cpi->common.width, cpi->common.height) >= 720)? 1 : 0;
+
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
- FLAG_SKIP_COMP_REFMISMATCH |
- FLAG_SKIP_INTRA_LOWVAR |
- FLAG_EARLY_TERMINATE;
- sf->intra_y_mode_mask = INTRA_DC_TM;
- sf->intra_uv_mode_mask = INTRA_DC_TM;
- sf->use_uv_intra_rd_estimate = 1;
+ FLAG_SKIP_INTRA_LOWVAR;
+
sf->use_rd_breakout = 1;
- sf->skip_encode_sb = 1;
- sf->use_lp32x32fdct = 1;
sf->adaptive_motion_search = 1;
- sf->using_small_partition_info = 0;
- sf->disable_splitmv =
- (MIN(cpi->common.width, cpi->common.height) >= 720)? 1 : 0;
sf->auto_mv_step_size = 1;
- sf->search_method = SQUARE;
- sf->subpel_iters_per_step = 1;
- sf->use_fast_lpf_pick = 1;
+
+ sf->disable_filter_search_var_thresh = 16;
+ sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
+
sf->auto_min_max_partition_size = 1;
- sf->disable_split_var_thresh = 32;
- sf->disable_filter_search_var_thresh = 32;
+ sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_ALL;
+ sf->adjust_partitioning_from_last_frame = 1;
+ sf->last_partitioning_redo_frequency = 3;
+
+ sf->use_uv_intra_rd_estimate = 1;
+ sf->skip_encode_sb = 1;
+ sf->use_lp32x32fdct = 1;
+ sf->subpel_iters_per_step = 1;
sf->use_fast_coef_updates = 2;
+
+ sf->adaptive_rd_thresh = 4;
sf->mode_skip_start = 6;
}
- if (speed == 3) {
+ if (speed == 4) {
sf->less_rectangular_check = 1;
sf->use_square_partition_only = 1;
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
- sf->use_lastframe_partitioning = 1;
+ sf->use_lastframe_partitioning = LAST_FRAME_PARTITION_ALL;
sf->adjust_partitioning_from_last_frame = 1;
sf->last_partitioning_redo_frequency = 3;
sf->tx_size_search_method = ((cpi->common.frame_type == KEY_FRAME ||
@@ -872,9 +873,10 @@ void vp9_set_speed_features(VP9_COMP *cpi) {
sf->disable_split_var_thresh = 64;
sf->disable_filter_search_var_thresh = 64;
sf->use_fast_coef_updates = 2;
+ sf->adaptive_rd_thresh = 4;
sf->mode_skip_start = 6;
}
- if (speed == 4) {
+ if (speed == 5) {
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
sf->use_one_partition_size_always = 1;
sf->always_this_block_size = BLOCK_16X16;
@@ -904,6 +906,7 @@ void vp9_set_speed_features(VP9_COMP *cpi) {
sf->intra_y_mode_mask = INTRA_DC_ONLY;
sf->intra_uv_mode_mask = INTRA_DC_ONLY;
sf->use_fast_coef_updates = 2;
+ sf->adaptive_rd_thresh = 4;
sf->mode_skip_start = 6;
}
break;
diff --git a/vp9/encoder/vp9_onyx_int.h b/vp9/encoder/vp9_onyx_int.h
index 3dbeb98e2a..db8b484cc8 100644
--- a/vp9/encoder/vp9_onyx_int.h
+++ b/vp9/encoder/vp9_onyx_int.h
@@ -36,7 +36,7 @@
#define DISABLE_RC_LONG_TERM_MEM 0
#endif
-// #define MODE_TEST_HIT_STATS
+#define MODE_TEST_HIT_STATS
// #define SPEEDSTATS 1
#if CONFIG_MULTIPLE_ARF
@@ -229,6 +229,12 @@ typedef enum {
#define INTRA_DC_TM ((1 << TM_PRED) | (1 << DC_PRED))
#define INTRA_DC_TM_H_V (INTRA_DC_TM | (1 << V_PRED) | (1 << H_PRED))
+typedef enum {
+ LAST_FRAME_PARTITION_OFF = 0,
+ LAST_FRAME_PARTITION_LOW_MOTION = 1,
+ LAST_FRAME_PARTITION_ALL = 2
+} LAST_FRAME_PARTITION_METHOD;
+
typedef struct {
int RD;
SEARCH_METHODS search_method;
@@ -246,7 +252,7 @@ typedef struct {
int adaptive_rd_thresh;
int skip_encode_sb;
int skip_encode_frame;
- int use_lastframe_partitioning;
+ LAST_FRAME_PARTITION_METHOD use_lastframe_partitioning;
TX_SIZE_SEARCH_METHOD tx_size_search_method;
int use_lp32x32fdct;
int use_avoid_tested_higherror;
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 83cd612261..96ed6f7952 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -52,6 +52,8 @@ DECLARE_ALIGNED(16, extern const uint8_t,
#define GOLDEN_FRAME_MODE_MASK 0xFFB5A3BB0
#define ALT_REF_MODE_MASK 0xFF8C648D0
+#define MIN_EARLY_TERM_INDEX 3
+
const MODE_DEFINITION vp9_mode_order[MAX_MODES] = {
{RD_NEARESTMV, LAST_FRAME, NONE},
{RD_NEARESTMV, ALTREF_FRAME, NONE},
@@ -3851,7 +3853,8 @@ int64_t vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
// TODO(debargha): enhance this test with a better distortion prediction
// based on qp, activity mask and history
- if (cpi->sf.mode_search_skip_flags & FLAG_EARLY_TERMINATE) {
+ if ((cpi->sf.mode_search_skip_flags & FLAG_EARLY_TERMINATE) &&
+ (mode_index > MIN_EARLY_TERM_INDEX)) {
const int qstep = xd->plane[0].dequant[1];
// TODO(debargha): Enhance this by specializing for each mode_index
int scale = 4;
--
GitLab