Refactor compound search loop
Refactors the two-loop compound search framework to create a
generic framework where any set of modes could be used in the
first pass and the rest in the second pass.
Also whether to use 2-pass or 1-pass is now a speed-feature.
When turned on, it gives 4% speed up with ~0.08% loss, as
tested on lowres, midres and AWCY-Objective-1-Fast.
Change-Id: Iaca3cc6ac36193cbfe34805795c51024df538ee8
diff --git a/av1/encoder/rdopt.c b/av1/encoder/rdopt.c
index e6e4da9..a505320 100644
--- a/av1/encoder/rdopt.c
+++ b/av1/encoder/rdopt.c
@@ -9736,7 +9736,7 @@
static int compound_type_rd(
const AV1_COMP *const cpi, MACROBLOCK *x, BLOCK_SIZE bsize, int mi_col,
- int mi_row, int_mv *cur_mv, int do_comp_distwtd, int masked_compound_used,
+ int mi_row, int_mv *cur_mv, int mode_search_mask, int masked_compound_used,
const BUFFER_SET *orig_dst, const BUFFER_SET *tmp_dst,
CompoundTypeRdBuffers *buffers, int *rate_mv, int64_t *rd,
RD_STATS *rd_stats, int64_t ref_best_rd, int *is_luma_interp_done) {
@@ -9773,10 +9773,10 @@
if (cur_type >= COMPOUND_WEDGE && !masked_compound_used) break;
if (!is_interinter_compound_used(cur_type, bsize)) continue;
if (cur_type == COMPOUND_DISTWTD &&
- (!do_comp_distwtd ||
- !cm->seq_params.order_hint_info.enable_dist_wtd_comp ||
+ (!cm->seq_params.order_hint_info.enable_dist_wtd_comp ||
cpi->sf.use_dist_wtd_comp_flag == DIST_WTD_COMP_DISABLED))
continue;
+ if (((1 << cur_type) & mode_search_mask) == 0) continue;
tmp_rate_mv = *rate_mv;
int64_t best_rd_cur = INT64_MAX;
mbmi->interinter_comp.type = cur_type;
@@ -9822,7 +9822,7 @@
}
}
// use spare buffer for following compound type try
- restore_dst_buf(xd, *tmp_dst, 1);
+ if (cur_type == COMPOUND_AVERAGE) restore_dst_buf(xd, *tmp_dst, 1);
} else {
mbmi->comp_group_idx = 1;
mbmi->compound_idx = 1;
@@ -9926,8 +9926,6 @@
int_mv mv;
} inter_mode_info;
-#define SEPARATE_COMP_DISTWTD_RD 1
-
static int64_t handle_inter_mode(
const AV1_COMP *const cpi, TileDataEnc *tile_data, MACROBLOCK *x,
BLOCK_SIZE bsize, RD_STATS *rd_stats, RD_STATS *rd_stats_y,
@@ -9981,14 +9979,11 @@
inter_mode_info mode_info[MAX_REF_MV_SERCH];
int comp_idx;
-#if SEPARATE_COMP_DISTWTD_RD
const int search_dist_wtd_comp =
is_comp_pred & cm->seq_params.order_hint_info.enable_dist_wtd_comp &
(mbmi->mode != GLOBAL_GLOBALMV) &
- (cpi->sf.use_dist_wtd_comp_flag != DIST_WTD_COMP_DISABLED);
-#else
- const int search_dist_wtd_comp = 0;
-#endif // SEPARATE_COMP_DISTWTD_RD
+ (cpi->sf.use_dist_wtd_comp_flag != DIST_WTD_COMP_DISABLED) &
+ cpi->sf.two_loop_comp_search;
// TODO(jingning): This should be deprecated shortly.
const int has_nearmv = have_nearmv_in_inter_mode(mbmi->mode) ? 1 : 0;
@@ -10044,12 +10039,13 @@
continue;
}
+ mbmi->comp_group_idx = 0;
+ mbmi->compound_idx = 1;
const RD_STATS backup_rd_stats = *rd_stats;
// If !search_dist_wtd_comp, we need to force mbmi->compound_idx = 1.
for (comp_idx = 1; comp_idx >= !search_dist_wtd_comp; --comp_idx) {
int rs = 0;
int compmode_interinter_cost = 0;
- mbmi->compound_idx = comp_idx;
if (is_comp_pred && comp_idx == 0) *rd_stats = backup_rd_stats;
@@ -10070,10 +10066,7 @@
newmv_ret_val = args->single_newmv_valid[ref_mv_idx][ref0] ? 0 : 1;
cur_mv[0] = args->single_newmv[ref_mv_idx][ref0];
rate_mv = args->single_newmv_rate[ref_mv_idx][ref0];
- } else if (!(search_dist_wtd_comp &&
- (cpi->sf.use_dist_wtd_comp_flag ==
- DIST_WTD_COMP_SKIP_MV_SEARCH) &&
- comp_idx == 0)) {
+ } else if (comp_idx == 1) {
newmv_ret_val = handle_newmv(cpi, x, bsize, cur_mv, mi_row, mi_col,
&rate_mv, args);
@@ -10179,17 +10172,30 @@
int skip_build_pred = 0;
if (is_comp_pred) {
- if (comp_idx == 0) {
- mbmi->interinter_comp.type = COMPOUND_DISTWTD;
+ int mode_search_mask;
+ if (cpi->sf.two_loop_comp_search) {
+ mode_search_mask = comp_idx ? (1 << COMPOUND_AVERAGE) |
+ (1 << COMPOUND_WEDGE) |
+ (1 << COMPOUND_DIFFWTD)
+ : (1 << COMPOUND_DISTWTD);
+ } else {
+ mode_search_mask = (1 << COMPOUND_AVERAGE) | (1 << COMPOUND_WEDGE) |
+ (1 << COMPOUND_DIFFWTD) | (1 << COMPOUND_DISTWTD);
+ }
+ if (mode_search_mask == (1 << COMPOUND_DISTWTD) ||
+ mode_search_mask == (1 << COMPOUND_AVERAGE)) {
+ mbmi->interinter_comp.type =
+ (mode_search_mask == (1 << COMPOUND_AVERAGE)) ? COMPOUND_AVERAGE
+ : COMPOUND_DISTWTD;
mbmi->num_proj_ref = 0;
mbmi->motion_mode = SIMPLE_TRANSLATION;
mbmi->comp_group_idx = 0;
+ mbmi->compound_idx = (mode_search_mask == (1 << COMPOUND_AVERAGE));
const int comp_index_ctx = get_comp_index_context(cm, xd);
compmode_interinter_cost +=
x->comp_idx_cost[comp_index_ctx][mbmi->compound_idx];
- }
- if (comp_idx) {
+ } else {
// Find matching interp filter or set to default interp filter
const int need_search =
av1_is_interp_needed(xd) && av1_is_interp_search_needed(xd);
@@ -10204,13 +10210,13 @@
}
int64_t best_rd_compound;
- const int do_comp_distwtd = !SEPARATE_COMP_DISTWTD_RD;
compmode_interinter_cost = compound_type_rd(
- cpi, x, bsize, mi_col, mi_row, cur_mv, do_comp_distwtd,
+ cpi, x, bsize, mi_col, mi_row, cur_mv, mode_search_mask,
masked_compound_used, &orig_dst, &tmp_dst, rd_buffers, &rate_mv,
&best_rd_compound, rd_stats, ref_best_rd, &is_luma_interp_done);
if (ref_best_rd < INT64_MAX &&
- (best_rd_compound >> 4) * (11 + 2 * SEPARATE_COMP_DISTWTD_RD) >
+ (best_rd_compound >> 4) *
+ (11 + 2 * cpi->sf.two_loop_comp_search) >
ref_best_rd) {
restore_dst_buf(xd, orig_dst, num_planes);
continue;
@@ -10266,8 +10272,7 @@
}
rd_stats->rate += compmode_interinter_cost;
- if (search_dist_wtd_comp && cpi->sf.dist_wtd_comp_fast_tx_search &&
- comp_idx == 0) {
+ if (cpi->sf.second_loop_comp_fast_tx_search && comp_idx == 0) {
// TODO(chengchen): this speed feature introduces big loss.
// Need better estimation of rate distortion.
int dummy_rate;
diff --git a/av1/encoder/speed_features.c b/av1/encoder/speed_features.c
index 5e9eda3..b5121f0 100644
--- a/av1/encoder/speed_features.c
+++ b/av1/encoder/speed_features.c
@@ -719,7 +719,8 @@
sf->use_inter_txb_hash = 0;
sf->use_mb_rd_hash = 1;
sf->optimize_b_precheck = 0;
- sf->dist_wtd_comp_fast_tx_search = 0;
+ sf->two_loop_comp_search = 1;
+ sf->second_loop_comp_fast_tx_search = 0;
sf->use_dist_wtd_comp_flag = DIST_WTD_COMP_ENABLED;
sf->reuse_inter_intra_mode = 0;
sf->intra_angle_estimation = 0;
diff --git a/av1/encoder/speed_features.h b/av1/encoder/speed_features.h
index 66c096d..f0816f8 100644
--- a/av1/encoder/speed_features.h
+++ b/av1/encoder/speed_features.h
@@ -562,8 +562,11 @@
// Calculate RD cost before doing optimize_b, and skip if the cost is large.
int optimize_b_precheck;
- // Use model rd instead of transform search in dist_wtd_comp
- int dist_wtd_comp_fast_tx_search;
+ // Use two-loop compound search
+ int two_loop_comp_search;
+
+ // Use model rd instead of transform search in second loop of compound search
+ int second_loop_comp_fast_tx_search;
// Decide when and how to use joint_comp.
DIST_WTD_COMP_FLAG use_dist_wtd_comp_flag;
