ref: 06ec873f2c97d8b60b28a346d2e868badbd98d2d
parent: d8f68c245481c1511ab25db3d5cf3c9b9fea2d84
author: Jingning Han <[email protected]>
date: Fri Apr 4 11:59:10 EDT 2014
Select prediction filter type in non-RD mode decision This commit allows the non-RD mode decision flow to select prediction filter type in NEWMV mode. It provides 8.14% compression performance gains in both settings of AQ=0 and 3. The current speed impact is about 5% to 10% slower. Change-Id: Id66ecebf77abd8f90fb3f6a066c0e8dfb4bf1c42
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -3343,8 +3343,7 @@
} else {cpi->mb.e_mbd.lossless = cpi->oxcf.lossless;
cm->reference_mode = SINGLE_REFERENCE;
- // Force the usage of the BILINEAR interp_filter.
- cm->interp_filter = BILINEAR;
+ cm->interp_filter = SWITCHABLE;
encode_frame_internal(cpi);
}
}
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -214,6 +214,7 @@
struct macroblockd_plane *const pd = &xd->plane[0];
MB_PREDICTION_MODE this_mode, best_mode = ZEROMV;
MV_REFERENCE_FRAME ref_frame, best_ref_frame = LAST_FRAME;
+ INTERP_FILTER best_pred_filter = EIGHTTAP;
int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES];
struct buf_2d yv12_mb[4][MAX_MB_PLANE];
static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG,@@ -236,6 +237,7 @@
const int *const rd_thresh_freq_fact = cpi->rd_thresh_freq_fact[bsize];
// Mode index conversion form THR_MODES to MB_PREDICTION_MODE for a ref frame.
int mode_idx[MB_MODE_COUNT] = {0};+ INTERP_FILTER filter_ref = SWITCHABLE;
x->skip_encode = cpi->sf.skip_encode_frame && x->q_index < QIDX_SKIP_THRESH;
@@ -267,6 +269,11 @@
frame_mv[ZEROMV][ref_frame].as_int = 0;
}
+ if (xd->up_available)
+ filter_ref = xd->mi[-xd->mi_stride]->mbmi.interp_filter;
+ else if (xd->left_available)
+ filter_ref = xd->mi[-1]->mbmi.interp_filter;
+
for (ref_frame = LAST_FRAME; ref_frame <= LAST_FRAME ; ++ref_frame) {if (!(cpi->ref_frame_flags & flag_list[ref_frame]))
continue;
@@ -326,9 +333,63 @@
mbmi->mode = this_mode;
mbmi->mv[0].as_int = frame_mv[this_mode][ref_frame].as_int;
- vp9_build_inter_predictors_sby(xd, mi_row, mi_col, bsize);
- model_rd_for_sb_y(cpi, bsize, x, xd, &rate, &dist);
+ // Search for the best prediction filter type, when the resulting
+ // motion vector is at sub-pixel accuracy level for luma component, i.e.,
+ // the last three bits are all zeros.
+ if ((this_mode == NEWMV || filter_ref == SWITCHABLE) &&
+ ((mbmi->mv[0].as_mv.row & 0x07) != 0 ||
+ (mbmi->mv[0].as_mv.col & 0x07) != 0)) {+ int64_t tmp_rdcost1 = INT64_MAX;
+ int64_t tmp_rdcost2 = INT64_MAX;
+ int64_t tmp_rdcost3 = INT64_MAX;
+ int pf_rate[3];
+ int64_t pf_dist[3];
+
+ mbmi->interp_filter = EIGHTTAP;
+ vp9_build_inter_predictors_sby(xd, mi_row, mi_col, bsize);
+ model_rd_for_sb_y(cpi, bsize, x, xd, &pf_rate[EIGHTTAP],
+ &pf_dist[EIGHTTAP]);
+ tmp_rdcost1 = RDCOST(x->rdmult, x->rddiv,
+ vp9_get_switchable_rate(x) + pf_rate[EIGHTTAP],
+ pf_dist[EIGHTTAP]);
+
+ mbmi->interp_filter = EIGHTTAP_SHARP;
+ vp9_build_inter_predictors_sby(xd, mi_row, mi_col, bsize);
+ model_rd_for_sb_y(cpi, bsize, x, xd, &pf_rate[EIGHTTAP_SHARP],
+ &pf_dist[EIGHTTAP_SHARP]);
+ tmp_rdcost2 = RDCOST(x->rdmult, x->rddiv,
+ vp9_get_switchable_rate(x) + pf_rate[EIGHTTAP_SHARP],
+ pf_dist[EIGHTTAP_SHARP]);
+
+ mbmi->interp_filter = EIGHTTAP_SMOOTH;
+ vp9_build_inter_predictors_sby(xd, mi_row, mi_col, bsize);
+ model_rd_for_sb_y(cpi, bsize, x, xd, &pf_rate[EIGHTTAP_SMOOTH],
+ &pf_dist[EIGHTTAP_SMOOTH]);
+ tmp_rdcost3 = RDCOST(x->rdmult, x->rddiv,
+ vp9_get_switchable_rate(x) + pf_rate[EIGHTTAP_SMOOTH],
+ pf_dist[EIGHTTAP_SMOOTH]);
+
+ if (tmp_rdcost2 < tmp_rdcost1) {+ if (tmp_rdcost2 < tmp_rdcost3)
+ mbmi->interp_filter = EIGHTTAP_SHARP;
+ else
+ mbmi->interp_filter = EIGHTTAP_SMOOTH;
+ } else {+ if (tmp_rdcost1 < tmp_rdcost3)
+ mbmi->interp_filter = EIGHTTAP;
+ else
+ mbmi->interp_filter = EIGHTTAP_SMOOTH;
+ }
+
+ rate = pf_rate[mbmi->interp_filter];
+ dist = pf_dist[mbmi->interp_filter];
+ } else {+ mbmi->interp_filter = (filter_ref == SWITCHABLE) ? EIGHTTAP: filter_ref;
+ vp9_build_inter_predictors_sby(xd, mi_row, mi_col, bsize);
+ model_rd_for_sb_y(cpi, bsize, x, xd, &rate, &dist);
+ }
+
rate += rate_mv;
rate += x->inter_mode_cost[mbmi->mode_context[ref_frame]]
[INTER_OFFSET(this_mode)];
@@ -339,6 +400,7 @@
*returnrate = rate;
*returndistortion = dist;
best_mode = this_mode;
+ best_pred_filter = mbmi->interp_filter;
best_ref_frame = ref_frame;
}
}
@@ -345,6 +407,7 @@
}
mbmi->mode = best_mode;
+ mbmi->interp_filter = best_pred_filter;
mbmi->ref_frame[0] = best_ref_frame;
mbmi->mv[0].as_int = frame_mv[best_mode][best_ref_frame].as_int;
xd->mi[0]->bmi[0].as_mv[0].as_int = mbmi->mv[0].as_int;
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -2358,7 +2358,7 @@
return (scaled_idx != ref_idx) ? &cm->frame_bufs[scaled_idx].buf : NULL;
}
-static INLINE int get_switchable_rate(const MACROBLOCK *x) {+int vp9_get_switchable_rate(const MACROBLOCK *x) {const MACROBLOCKD *const xd = &x->e_mbd;
const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
const int ctx = vp9_get_pred_context_switchable_interp(xd);
@@ -2815,7 +2815,7 @@
int j;
int64_t rs_rd;
mbmi->interp_filter = i;
- rs = get_switchable_rate(x);
+ rs = vp9_get_switchable_rate(x);
rs_rd = RDCOST(x->rdmult, x->rddiv, rs, 0);
if (i > 0 && intpel_mv) {@@ -2885,7 +2885,7 @@
// Set the appropriate filter
mbmi->interp_filter = cm->interp_filter != SWITCHABLE ?
cm->interp_filter : *best_filter;
- rs = cm->interp_filter == SWITCHABLE ? get_switchable_rate(x) : 0;
+ rs = cm->interp_filter == SWITCHABLE ? vp9_get_switchable_rate(x) : 0;
if (pred_exists) { if (best_needs_copy) {@@ -2915,7 +2915,7 @@
}
if (cm->interp_filter == SWITCHABLE)
- *rate2 += get_switchable_rate(x);
+ *rate2 += vp9_get_switchable_rate(x);
if (!is_comp_pred) { if (!x->in_active_map) {@@ -4072,7 +4072,7 @@
if (tmp_rd == INT64_MAX)
continue;
- rs = get_switchable_rate(x);
+ rs = vp9_get_switchable_rate(x);
rs_rd = RDCOST(x->rdmult, x->rddiv, rs, 0);
cpi->rd_filter_cache[switchable_filter_index] = tmp_rd;
cpi->rd_filter_cache[SWITCHABLE_FILTERS] =
@@ -4153,7 +4153,7 @@
distortion2 += distortion;
if (cm->interp_filter == SWITCHABLE)
- rate2 += get_switchable_rate(x);
+ rate2 += vp9_get_switchable_rate(x);
if (!mode_excluded)
mode_excluded = comp_pred ? cm->reference_mode == SINGLE_REFERENCE
--- a/vp9/encoder/vp9_rdopt.h
+++ b/vp9/encoder/vp9_rdopt.h
@@ -40,6 +40,8 @@
unsigned int qstep, int *rate,
int64_t *dist);
+int vp9_get_switchable_rate(const MACROBLOCK *x);
+
void vp9_setup_buffer_inter(VP9_COMP *cpi, MACROBLOCK *x,
const TileInfo *const tile,
MV_REFERENCE_FRAME ref_frame,