Select search_method based on current bsize
Performance:
---------+---------+----------+----------+---------+-------
SPD_SET | TESTSET | AVG_PSNR | OVR_PSNR | SSIM | SPD
---------+---------+----------+----------+---------+-------
| HDRES | +0.307% | +0.315% | +0.315% | +2.1%
6 | MIDRES | +0.210% | +0.191% | +0.362% | +1.5%
| LOWRES | -0.015% | -0.017% | +0.009% | +0.4%
---------+---------+----------+----------+---------+-------
STATS_CHANGED
Change-Id: I165e14fd318fceb0b2cd665bbc0029a67795b2d0
diff --git a/av1/encoder/mcomp.c b/av1/encoder/mcomp.c
index 7c98ae8..bdbff81 100644
--- a/av1/encoder/mcomp.c
+++ b/av1/encoder/mcomp.c
@@ -55,6 +55,26 @@
ms_buffers->obmc_mask = x->obmc_buffer.mask;
}
+static AOM_INLINE SEARCH_METHODS
+get_faster_search_method(SEARCH_METHODS search_method) {
+ // Note on search method's accuracy:
+ // 1. NSTEP
+ // 2. DIAMOND
+ // 3. BIGDIA \approx SQUARE
+ // 4. HEX.
+ // 5. FAST_HEX \approx FAST_DIAMOND
+ switch (search_method) {
+ case NSTEP: return DIAMOND;
+ case DIAMOND: return BIGDIA;
+ case BIGDIA: return HEX;
+ case SQUARE: return HEX;
+ case HEX: return FAST_HEX;
+ case FAST_HEX: return FAST_HEX;
+ case FAST_DIAMOND: return FAST_DIAMOND;
+ default: assert(0 && "Invalid search method!"); return DIAMOND;
+ }
+}
+
void av1_make_default_fullpel_ms_params(FULLPEL_MOTION_SEARCH_PARAMS *ms_params,
const struct AV1_COMP *cpi,
const MACROBLOCK *x, BLOCK_SIZE bsize,
@@ -62,26 +82,35 @@
const search_site_config *search_sites,
int fine_search_interval,
int fast_dia_search_level) {
+ const MV_SPEED_FEATURES *mv_sf = &cpi->sf.mv_sf;
+
// High level params
ms_params->bsize = bsize;
ms_params->vfp = &cpi->fn_ptr[bsize];
init_ms_buffers(&ms_params->ms_buffers, x);
- ms_params->search_method = cpi->sf.mv_sf.search_method;
+ ms_params->search_method = mv_sf->search_method;
+ if (mv_sf->use_bsize_dependent_search_method) {
+ const int min_dim = AOMMIN(block_size_wide[bsize], block_size_high[bsize]);
+ if (min_dim >= 32) {
+ ms_params->search_method =
+ get_faster_search_method(ms_params->search_method);
+ }
+ }
ms_params->search_sites = search_sites;
- ms_params->mesh_patterns[0] = cpi->sf.mv_sf.mesh_patterns;
- ms_params->mesh_patterns[1] = cpi->sf.mv_sf.intrabc_mesh_patterns;
- ms_params->force_mesh_thresh = cpi->sf.mv_sf.exhaustive_searches_thresh;
- ms_params->prune_mesh_search = cpi->sf.mv_sf.prune_mesh_search;
+ ms_params->mesh_patterns[0] = mv_sf->mesh_patterns;
+ ms_params->mesh_patterns[1] = mv_sf->intrabc_mesh_patterns;
+ ms_params->force_mesh_thresh = mv_sf->exhaustive_searches_thresh;
+ ms_params->prune_mesh_search = mv_sf->prune_mesh_search;
ms_params->run_mesh_search = 0;
ms_params->fine_search_interval = fine_search_interval;
ms_params->fast_dia_search_level = fast_dia_search_level;
ms_params->is_intra_mode = 0;
- ms_params->fast_obmc_search = cpi->sf.mv_sf.obmc_full_pixel_search_level;
+ ms_params->fast_obmc_search = mv_sf->obmc_full_pixel_search_level;
ms_params->mv_limits = x->mv_limits;
av1_set_mv_search_range(&ms_params->mv_limits, ref_mv);
@@ -991,7 +1020,7 @@
// passed into this function
static int pattern_search(
FULLPEL_MV start_mv, const FULLPEL_MOTION_SEARCH_PARAMS *ms_params,
- const int search_step, const int do_init_search,
+ int search_step, const int do_init_search,
const int num_candidates[MAX_PATTERN_SCALES],
const MV candidates[MAX_PATTERN_SCALES][MAX_PATTERN_CANDIDATES],
int *cost_list, FULLPEL_MV *best_mv) {
@@ -1009,7 +1038,8 @@
int thissad;
int k = -1;
const MV_COST_PARAMS *mv_cost_params = &ms_params->mv_cost_params;
- assert(search_step < MAX_MVSEARCH_STEPS);
+ search_step = AOMMIN(search_step, MAX_MVSEARCH_STEPS - 1);
+ assert(search_step >= 0);
int best_init_s = search_steps[search_step];
// adjust ref_mv to make sure it is within MV range
clamp_fullmv(&start_mv, &ms_params->mv_limits);
diff --git a/av1/encoder/speed_features.c b/av1/encoder/speed_features.c
index 4cf9afa..697b183 100644
--- a/av1/encoder/speed_features.c
+++ b/av1/encoder/speed_features.c
@@ -595,7 +595,6 @@
// TODO(any): The following features give really bad quality/speed trade
// off. Needs to be re-worked.
- // sf->mv_sf.search_method = BIGDIA;
// sf->inter_sf.adaptive_rd_thresh = 4;
// sf->rd_sf.tx_domain_dist_level = 2;
// sf->rt_sf.mode_search_skip_flags =
@@ -636,7 +635,7 @@
if (speed >= 6) {
sf->mv_sf.simple_motion_subpel_force_stop = FULL_PEL;
- sf->rd_sf.perform_coeff_opt = is_boosted_arf2_bwd_type ? 4 : 6;
+ sf->mv_sf.use_bsize_dependent_search_method = 1;
sf->tpl_sf.disable_gop_length_decision = 1;
sf->tpl_sf.subpel_force_stop = FULL_PEL;
@@ -646,6 +645,8 @@
sf->tx_sf.use_intra_txb_hash = 1;
sf->tx_sf.tx_type_search.prune_tx_type_est_rd = 0;
+ sf->rd_sf.perform_coeff_opt = is_boosted_arf2_bwd_type ? 4 : 6;
+
sf->winner_mode_sf.multi_winner_mode_type = MULTI_WINNER_MODE_OFF;
}
@@ -1017,6 +1018,7 @@
mv_sf->subpel_iters_per_step = 2;
mv_sf->subpel_search_method = SUBPEL_TREE;
mv_sf->use_accurate_subpel_search = USE_8_TAPS;
+ mv_sf->use_bsize_dependent_search_method = 0;
mv_sf->use_fullpel_costlist = 0;
}
diff --git a/av1/encoder/speed_features.h b/av1/encoder/speed_features.h
index 1c0e58f..a2d24fb 100644
--- a/av1/encoder/speed_features.h
+++ b/av1/encoder/speed_features.h
@@ -477,6 +477,12 @@
// Motion search method (Diamond, NSTEP, Hex, Big Diamond, Square, etc).
SEARCH_METHODS search_method;
+ // Enable the use of faster, less accurate mv search method on bsize >=
+ // BLOCK_32X32.
+ // TODO(chiyotsai@google.com): Take the clip's resolution and mv activity into
+ // account.
+ int use_bsize_dependent_search_method;
+
// If this is set to 1, we limit the motion search range to 2 times the
// largest motion vector found in the last frame.
int auto_mv_step_size;
