Use whole pixel only at speed 8 screen content.

+5.857% BD-RATE on SCREEN_CONTENT

Leaving this off for non-screen content because:
+25.300% on TWITCH120
+37.833% BD-RATE on RTC

Change-Id: Ie0a312182d6cc859fb04298e4cd81d02b39e23fe

diff --git a/vp9/encoder/vp9_mcomp.c b/vp9/encoder/vp9_mcomp.c
index 8b7825e7b69ddabdbf2e09bd051615fe57dbe159..4669145187a2d2947a5406a52f7df5c38de3afa2 100644 (file)
--- a/vp9/encoder/vp9_mcomp.c
+++ b/vp9/encoder/vp9_mcomp.c
@@ -383,6 +383,51 @@ static void get_cost_surf_min(int *cost_list, int *ir, int *ic,
                          (cost_list[4] - 2 * cost_list[0] + cost_list[2]));
 }
 
+int vp9_skip_sub_pixel_tree(
+    const MACROBLOCK *x,
+    MV *bestmv, const MV *ref_mv,
+    int allow_hp,
+    int error_per_bit,
+    const vp9_variance_fn_ptr_t *vfp,
+    int forced_stop,
+    int iters_per_step,
+    int *cost_list,
+    int *mvjcost, int *mvcost[2],
+    int *distortion,
+    unsigned int *sse1,
+    const uint8_t *second_pred,
+    int w, int h) {
+  SETUP_SUBPEL_SEARCH;
+  besterr = setup_center_error(xd, bestmv, ref_mv, error_per_bit, vfp,
+                               z, src_stride, y, y_stride, second_pred,
+                               w, h, offset, mvjcost, mvcost,
+                               sse1, distortion);
+  (void) halfiters;
+  (void) quarteriters;
+  (void) eighthiters;
+  (void) whichdir;
+  (void) allow_hp;
+  (void) forced_stop;
+  (void) hstep;
+  (void) rr;
+  (void) rc;
+  (void) minr;
+  (void) minc;
+  (void) maxr;
+  (void) maxc;
+  (void) tr;
+  (void) tc;
+  (void) sse;
+  (void) thismse;
+  (void) cost_list;
+
+  if ((abs(bestmv->col - ref_mv->col) > (MAX_FULL_PEL_VAL << 3)) ||
+      (abs(bestmv->row - ref_mv->row) > (MAX_FULL_PEL_VAL << 3)))
+    return INT_MAX;
+
+  return besterr;
+}
+
 int vp9_find_best_sub_pixel_tree_pruned_evenmore(
     const MACROBLOCK *x,
     MV *bestmv, const MV *ref_mv,

diff --git a/vp9/encoder/vp9_mcomp.h b/vp9/encoder/vp9_mcomp.h
index 1c101f2e208b16ade39c743a74849f928480ebf3..1b0c860c5701b1e32dd0595aec9050931de9d996 100644 (file)
--- a/vp9/encoder/vp9_mcomp.h
+++ b/vp9/encoder/vp9_mcomp.h
@@ -92,6 +92,7 @@ extern fractional_mv_step_fp vp9_find_best_sub_pixel_tree;
 extern fractional_mv_step_fp vp9_find_best_sub_pixel_tree_pruned;
 extern fractional_mv_step_fp vp9_find_best_sub_pixel_tree_pruned_more;
 extern fractional_mv_step_fp vp9_find_best_sub_pixel_tree_pruned_evenmore;
+extern fractional_mv_step_fp vp9_skip_sub_pixel_tree;
 
 typedef int (*vp9_full_search_fn_t)(const MACROBLOCK *x,
                                     const MV *ref_mv, int sad_per_bit,

diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c
index f6845078a67af163266dffa7b92aae66d3d0a7f9..c014ca174bda7f64fe66268fb92fe5cab94d4129 100644 (file)
--- a/vp9/encoder/vp9_speed_features.c
+++ b/vp9/encoder/vp9_speed_features.c
@@ -441,7 +441,7 @@ static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf,
   }
   if (speed >= 8) {
     sf->adaptive_rd_thresh = 4;
-    sf->mv.subpel_force_stop = 2;
+    sf->mv.subpel_force_stop = (content == VP9E_CONTENT_SCREEN) ? 3 : 2;
     sf->lpf_pick = LPF_PICK_MINIMAL_LPF;
   }
 }
@@ -607,7 +607,10 @@ void vp9_set_speed_features_framesize_independent(VP9_COMP *cpi) {
     sf->optimize_coefficients = 0;
   }
 
-  if (sf->mv.subpel_search_method == SUBPEL_TREE) {
+  if (sf->mv.subpel_force_stop == 3) {
+    // Whole pel only
+    cpi->find_fractional_mv_step = vp9_skip_sub_pixel_tree;
+  } else if (sf->mv.subpel_search_method == SUBPEL_TREE) {
     cpi->find_fractional_mv_step = vp9_find_best_sub_pixel_tree;
   } else if (sf->mv.subpel_search_method == SUBPEL_TREE_PRUNED) {
     cpi->find_fractional_mv_step = vp9_find_best_sub_pixel_tree_pruned;

diff --git a/vp9/encoder/vp9_speed_features.h b/vp9/encoder/vp9_speed_features.h
index fa2f79d31eac55f1be13f590cd95aa2978a27146..90b32164b64e1d07467858f19ff83f30e275a753 100644 (file)
--- a/vp9/encoder/vp9_speed_features.h
+++ b/vp9/encoder/vp9_speed_features.h
@@ -188,7 +188,11 @@ typedef struct MV_SPEED_FEATURES {
   // Maximum number of steps in logarithmic subpel search before giving up.
   int subpel_iters_per_step;
 
-  // Control when to stop subpel search
+  // Control when to stop subpel search:
+  // 0: Full subpel search.
+  // 1: Stop at quarter pixel.
+  // 2: Stop at half pixel.
+  // 3: Stop at full pixel.
   int subpel_force_stop;
 
   // This variable sets the step_param used in full pel motion search.