Turn on ML partition search breakout on speed 0

Enable ML based partition search breakout on speed 0 when frame
resolution is less then 720p and bitdepth is 8.

Compression performance change is neutral.
Tested encoding speed over 20 480p sequences:

Speed gain(%)	QP=30	QP=40	QP=50	QP=60
  max     	14.4	18.6	17.8	24.4
average	         4.6	 9.0	 8.0	13.2

Change-Id: Ia0d2947030ac774dc1533eb27ffc57f5b788a6ce

diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index d7faa4be9cd5601220f30088193912843bc6e41a..bfe34ac45d48f9d0e44da6e0eceaba6c634d36e8 100644 (file)
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -3662,33 +3662,7 @@ static void rd_pick_partition(VP9_COMP *cpi, ThreadData *td,
         best_rdc = this_rdc;
         if (bsize >= BLOCK_8X8) pc_tree->partitioning = PARTITION_NONE;
 
-        if (!cpi->sf.ml_partition_search_early_termination) {
-          // If all y, u, v transform blocks in this partition are skippable,
-          // and the dist & rate are within the thresholds, the partition search
-          // is terminated for current branch of the partition search tree.
-          if (!x->e_mbd.lossless && ctx->skippable) {
-            int use_ml_based_breakout =
-                cpi->sf.use_ml_partition_search_breakout &&
-                cm->base_qindex >= 100;
-#if CONFIG_VP9_HIGHBITDEPTH
-            if (x->e_mbd.cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
-              use_ml_based_breakout = 0;
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-            if (use_ml_based_breakout) {
-              if (ml_predict_breakout(cpi, bsize, x, &this_rdc)) {
-                do_split = 0;
-                do_rect = 0;
-              }
-            } else {
-              if ((best_rdc.dist < (dist_breakout_thr >> 2)) ||
-                  (best_rdc.dist < dist_breakout_thr &&
-                   best_rdc.rate < rate_breakout_thr)) {
-                do_split = 0;
-                do_rect = 0;
-              }
-            }
-          }
-        } else {
+        if (cpi->sf.ml_partition_search_early_termination) {
           // Currently, the machine-learning based partition search early
           // termination is only used while bsize is 16x16, 32x32 or 64x64,
           // VPXMIN(cm->width, cm->height) >= 480, and speed = 0.
@@ -3702,6 +3676,31 @@ static void rd_pick_partition(VP9_COMP *cpi, ThreadData *td,
           }
         }
 
+        if ((do_split || do_rect) && !x->e_mbd.lossless && ctx->skippable) {
+          int use_ml_based_breakout =
+              cpi->sf.use_ml_partition_search_breakout &&
+              cm->base_qindex >= 100;
+#if CONFIG_VP9_HIGHBITDEPTH
+          if (x->e_mbd.cur_buf->flags & YV12_FLAG_HIGHBITDEPTH)
+            use_ml_based_breakout = 0;
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+          if (use_ml_based_breakout) {
+            if (ml_predict_breakout(cpi, bsize, x, &this_rdc)) {
+              do_split = 0;
+              do_rect = 0;
+            }
+          } else {
+            if (!cpi->sf.ml_partition_search_early_termination) {
+              if ((best_rdc.dist < (dist_breakout_thr >> 2)) ||
+                  (best_rdc.dist < dist_breakout_thr &&
+                   best_rdc.rate < rate_breakout_thr)) {
+                do_split = 0;
+                do_rect = 0;
+              }
+            }
+          }
+        }
+
 #if CONFIG_FP_MB_STATS
         // Check if every 16x16 first pass block statistics has zero
         // motion and the corresponding first pass residue is small enough.

diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c
index b9f8055bc4cd296b77c62bdff43c70f86f4ddab3..75401131f1ad5f2b1b7cf50e07ba15afb87ba678 100644 (file)
--- a/vp9/encoder/vp9_speed_features.c
+++ b/vp9/encoder/vp9_speed_features.c
@@ -61,6 +61,9 @@ static void set_good_speed_feature_framesize_dependent(VP9_COMP *cpi,
                                                        SPEED_FEATURES *sf,
                                                        int speed) {
   VP9_COMMON *const cm = &cpi->common;
+  const int is_480p_or_larger = VPXMIN(cm->width, cm->height) >= 480;
+  const int is_720p_or_larger = VPXMIN(cm->width, cm->height) >= 720;
+  const int is_2160p_or_larger = VPXMIN(cm->width, cm->height) >= 2160;
 
   // speed 0 features
   sf->partition_search_breakout_thr.dist = (1 << 20);
@@ -68,21 +71,27 @@ static void set_good_speed_feature_framesize_dependent(VP9_COMP *cpi,
 
   // Currently, the machine-learning based partition search early termination
   // is only used while VPXMIN(cm->width, cm->height) >= 480 and speed = 0.
-  if (VPXMIN(cm->width, cm->height) >= 480) {
+  if (is_480p_or_larger) {
     sf->ml_partition_search_early_termination = 1;
   }
 
+  if (!is_720p_or_larger) {
+    sf->use_ml_partition_search_breakout = 1;
+    sf->ml_partition_search_breakout_thresh[0] = 2.5f;
+    sf->ml_partition_search_breakout_thresh[1] = 1.5f;
+    sf->ml_partition_search_breakout_thresh[2] = 1.5f;
+  }
+
   if (speed >= 1) {
     sf->ml_partition_search_early_termination = 0;
 
-    if (VPXMIN(cm->width, cm->height) >= 720) {
+    if (is_720p_or_larger) {
       sf->disable_split_mask =
           cm->show_frame ? DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT;
       sf->partition_search_breakout_thr.dist = (1 << 23);
     } else {
       sf->disable_split_mask = DISABLE_COMPOUND_SPLIT;
       sf->partition_search_breakout_thr.dist = (1 << 21);
-      sf->use_ml_partition_search_breakout = 1;
       sf->ml_partition_search_breakout_thresh[0] = 0.0f;
       sf->ml_partition_search_breakout_thresh[1] = 0.0f;
       sf->ml_partition_search_breakout_thresh[2] = 0.0f;
@@ -90,7 +99,7 @@ static void set_good_speed_feature_framesize_dependent(VP9_COMP *cpi,
   }
 
   if (speed >= 2) {
-    if (VPXMIN(cm->width, cm->height) >= 720) {
+    if (is_720p_or_larger) {
       sf->disable_split_mask =
           cm->show_frame ? DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT;
       sf->adaptive_pred_interp_filter = 0;
@@ -107,7 +116,7 @@ static void set_good_speed_feature_framesize_dependent(VP9_COMP *cpi,
     sf->rd_auto_partition_min_limit = set_partition_min_limit(cm);
 
     // Use a set of speed features for 4k videos.
-    if (VPXMIN(cm->width, cm->height) >= 2160) {
+    if (is_2160p_or_larger) {
       sf->use_square_partition_only = 1;
       sf->intra_y_mode_mask[TX_32X32] = INTRA_DC;
       sf->intra_uv_mode_mask[TX_32X32] = INTRA_DC;
@@ -120,7 +129,7 @@ static void set_good_speed_feature_framesize_dependent(VP9_COMP *cpi,
 
   if (speed >= 3) {
     sf->use_ml_partition_search_breakout = 0;
-    if (VPXMIN(cm->width, cm->height) >= 720) {
+    if (is_720p_or_larger) {
       sf->disable_split_mask = DISABLE_ALL_SPLIT;
       sf->schedule_mode_search = cm->base_qindex < 220 ? 1 : 0;
       sf->partition_search_breakout_thr.dist = (1 << 25);
@@ -145,7 +154,7 @@ static void set_good_speed_feature_framesize_dependent(VP9_COMP *cpi,
 
   if (speed >= 4) {
     sf->partition_search_breakout_thr.rate = 300;
-    if (VPXMIN(cm->width, cm->height) >= 720) {
+    if (is_720p_or_larger) {
       sf->partition_search_breakout_thr.dist = (1 << 26);
     } else {
       sf->partition_search_breakout_thr.dist = (1 << 24);