vp9: Speed feature to adapt partition based on source_sad.

Move the source_sad feature to speed 6 (from speed 7), and
add speed feature to switch from the variance-based partition
to reference_partition (which uses nonrd-pickmode for bsize selection)
if source_sad is high.

Currently used only for speed 6 for resoln <= 360p.
About 4-5% improvement on 360p in RTC set.
Some speed slowdown, but still ~30% faster than speed 5.

Change-Id: Ib0330ee5fe9fdd2608aed91359a2a339d967491c

3 files changed, 24 insertions, 14 deletions

diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index d8115bc25..5220197fd 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -981,8 +981,8 @@ static void chroma_check(VP9_COMP *cpi, MACROBLOCK *x, int bsize,
   }
 }
 
-static void avg_source_sad(VP9_COMP *cpi, MACROBLOCK *x, int shift,
-                           int sb_offset) {
+static uint64_t avg_source_sad(VP9_COMP *cpi, MACROBLOCK *x, int shift,
+                               int sb_offset) {
   unsigned int tmp_sse;
   uint64_t tmp_sad;
   unsigned int tmp_variance;
@@ -994,7 +994,7 @@ static void avg_source_sad(VP9_COMP *cpi, MACROBLOCK *x, int shift,
   uint64_t avg_source_sad_threshold = 10000;
   uint64_t avg_source_sad_threshold2 = 12000;
 #if CONFIG_VP9_HIGHBITDEPTH
-  if (cpi->common.use_highbitdepth) return;
+  if (cpi->common.use_highbitdepth) return 0;
 #endif
   src_y += shift;
   last_src_y += shift;
@@ -1025,7 +1025,7 @@ static void avg_source_sad(VP9_COMP *cpi, MACROBLOCK *x, int shift,
       cpi->content_state_sb_fd[sb_offset] = 0;
     }
   }
-  return;
+  return tmp_sad;
 }
 
 // This function chooses partitioning based on the variance between source and
@@ -4167,7 +4167,9 @@ static void encode_nonrd_sb_row(VP9_COMP *cpi, ThreadData *td,
     if (cpi->compute_source_sad_onepass && cpi->sf.use_source_sad) {
       int shift = cpi->Source->y_stride * (mi_row << 3) + (mi_col << 3);
       int sb_offset2 = ((cm->mi_cols + 7) >> 3) * (mi_row >> 3) + (mi_col >> 3);
-      avg_source_sad(cpi, x, shift, sb_offset2);
+      int64_t source_sad = avg_source_sad(cpi, x, shift, sb_offset2);
+      if (sf->adapt_partition_source_sad && source_sad > 40000)
+        partition_search_type = REFERENCE_PARTITION;
     }
 
     // Set the partition type of the 64X64 block
diff --git a/vp9/encoder/vp9_speed_features.c b/vp9/encoder/vp9_speed_features.c
index d91142845..6488dfd53 100644
--- a/vp9/encoder/vp9_speed_features.c
+++ b/vp9/encoder/vp9_speed_features.c
@@ -364,6 +364,7 @@ static void set_rt_speed_feature_framesize_independent(
   sf->copy_partition_flag = 0;
   sf->use_source_sad = 0;
   sf->use_simple_block_yrd = 0;
+  sf->adapt_partition_source_sad = 0;
 
   if (speed >= 1) {
     sf->allow_txfm_domain_distortion = 1;
@@ -527,6 +528,17 @@ static void set_rt_speed_feature_framesize_independent(
     sf->mv.search_method = NSTEP;
     sf->mv.reduce_first_step_size = 1;
     sf->skip_encode_sb = 0;
+    if (!cpi->external_resize) sf->use_source_sad = 1;
+    if (sf->use_source_sad) {
+      if (cm->width * cm->height <= 640 * 360)
+        sf->adapt_partition_source_sad = 1;
+      if (cpi->content_state_sb_fd == NULL &&
+          (!cpi->use_svc ||
+           cpi->svc.spatial_layer_id == cpi->svc.number_spatial_layers - 1)) {
+        cpi->content_state_sb_fd = (uint8_t *)vpx_calloc(
+            (cm->mi_stride >> 3) * ((cm->mi_rows >> 3) + 1), sizeof(uint8_t));
+      }
+    }
     if (cpi->oxcf.rc_mode == VPX_CBR && content != VP9E_CONTENT_SCREEN) {
       // Enable short circuit for low temporal variance.
       sf->short_circuit_low_temp_var = 1;
@@ -539,6 +551,7 @@ static void set_rt_speed_feature_framesize_independent(
   }
 
   if (speed >= 7) {
+    sf->adapt_partition_source_sad = 0;
     sf->adaptive_rd_thresh = 3;
     sf->mv.search_method = FAST_DIAMOND;
     sf->mv.fullpel_search_step_param = 10;
@@ -552,15 +565,6 @@ static void set_rt_speed_feature_framesize_independent(
       if (cpi->svc.non_reference_frame)
         sf->mv.subpel_search_method = SUBPEL_TREE_PRUNED_EVENMORE;
     }
-    if (!cpi->external_resize) sf->use_source_sad = 1;
-    if (sf->use_source_sad) {
-      if (cpi->content_state_sb_fd == NULL &&
-          (!cpi->use_svc ||
-           cpi->svc.spatial_layer_id == cpi->svc.number_spatial_layers - 1)) {
-        cpi->content_state_sb_fd = (uint8_t *)vpx_calloc(
-            (cm->mi_stride >> 3) * ((cm->mi_rows >> 3) + 1), sizeof(uint8_t));
-      }
-    }
     // Enable partition copy. For SVC only enabled for top spatial resolution
     // layer.
     cpi->max_copied_frame = 0;
diff --git a/vp9/encoder/vp9_speed_features.h b/vp9/encoder/vp9_speed_features.h
index ee485a35f..11e3154a0 100644
--- a/vp9/encoder/vp9_speed_features.h
+++ b/vp9/encoder/vp9_speed_features.h
@@ -490,6 +490,10 @@ typedef struct SPEED_FEATURES {
   int use_source_sad;
 
   int use_simple_block_yrd;
+
+  // If source sad of superblock is high, will switch from VARIANCE_PARTITION to
+  // REFERENCE_PARTITION (which selects partition based on the nonrd-pickmode).
+  int adapt_partition_source_sad;
 } SPEED_FEATURES;
 
 struct VP9_COMP;