vp9: Reuse motion from choose_partitioning in NEWMV search.

When int_pro_motion_estimation is done for superblock in
choose_partitioning, use it to avoid the full_pixel_search
for NEWMV mode, if bsize is >= 32X32.

For speed > 7.
Small/neutral change on RTC metrics.
~1-2% speedup on arm on high motion clip.

Change-Id: I3cfe6833ff4bf75d4afa83eaf058ad45729de85b

diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index 2ab948955df0720df3694ee1aa80aa3f482cc853..e64cb1b4b4ff20ee2819a34ea4790e01647844bc 100644 (file)
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -172,6 +172,12 @@ struct macroblock {
 
   uint8_t last_sb_high_content;
 
+  int sb_use_mv_part;
+
+  int sb_mvcol_part;
+
+  int sb_mvrow_part;
+
   // For each superblock: saves the content value (e.g., low/high sad/sumdiff)
   // based on source sad, prior to encoding the frame.
   uint8_t content_state_sb;

diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 948d01bc3b57c46315435c945adceb21845b9760..7cc6e20a33a449634825cedf980299b28f05547b 100644 (file)
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -1090,6 +1090,7 @@ static int choose_partitioning(VP9_COMP *cpi, const TileInfo *const tile,
     // If source_sad is low copy the partition without computing the y_sad.
     if (x->skip_low_source_sad && cpi->sf.copy_partition_flag &&
         copy_partitioning(cpi, x, xd, mi_row, mi_col, segment_id, sb_offset)) {
+      x->sb_use_mv_part = 1;
       return 0;
     }
   }
@@ -1166,12 +1167,17 @@ static int choose_partitioning(VP9_COMP *cpi, const TileInfo *const tile,
     mi->mv[0].as_int = 0;
     mi->interp_filter = BILINEAR;
 
-    if (cpi->oxcf.speed >= 8 && !low_res && x->content_state_sb != kVeryHighSad)
+    if (cpi->oxcf.speed >= 8 && !low_res &&
+        x->content_state_sb != kVeryHighSad) {
       y_sad = cpi->fn_ptr[bsize].sdf(
           x->plane[0].src.buf, x->plane[0].src.stride, xd->plane[0].pre[0].buf,
           xd->plane[0].pre[0].stride);
-    else
+    } else {
       y_sad = vp9_int_pro_motion_estimation(cpi, x, bsize, mi_row, mi_col);
+      x->sb_use_mv_part = 1;
+      x->sb_mvcol_part = mi->mv[0].as_mv.col;
+      x->sb_mvrow_part = mi->mv[0].as_mv.row;
+    }
 
     y_sad_last = y_sad;
     // Pick ref frame for partitioning, bias last frame when y_sad_g and y_sad
@@ -4145,6 +4151,9 @@ static void encode_nonrd_sb_row(VP9_COMP *cpi, ThreadData *td,
     x->skip_low_source_sad = 0;
     x->lowvar_highsumdiff = 0;
     x->content_state_sb = 0;
+    x->sb_use_mv_part = 0;
+    x->sb_mvcol_part = 0;
+    x->sb_mvrow_part = 0;
 
     if (seg->enabled) {
       const uint8_t *const map =

diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index f340ef2082726f9d3f3e0be303307e5b3ee6962b..25ae6c442cd1245d34949f1fae5a4eab5264300d 100644 (file)
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -193,9 +193,14 @@ static int combined_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
   else
     center_mv = tmp_mv->as_mv;
 
-  vp9_full_pixel_search(
-      cpi, x, bsize, &mvp_full, step_param, cpi->sf.mv.search_method, sadpb,
-      cond_cost_list(cpi, cost_list), &center_mv, &tmp_mv->as_mv, INT_MAX, 0);
+  if (x->sb_use_mv_part) {
+    tmp_mv->as_mv.row = x->sb_mvrow_part >> 3;
+    tmp_mv->as_mv.col = x->sb_mvcol_part >> 3;
+  } else {
+    vp9_full_pixel_search(
+        cpi, x, bsize, &mvp_full, step_param, cpi->sf.mv.search_method, sadpb,
+        cond_cost_list(cpi, cost_list), &center_mv, &tmp_mv->as_mv, INT_MAX, 0);
+  }
 
   x->mv_limits = tmp_mv_limits;
 
@@ -1628,6 +1633,8 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data,
     }
   }
 
+  if (cpi->oxcf.speed <= 7 || bsize < BLOCK_32X32) x->sb_use_mv_part = 0;
+
   for (idx = 0; idx < RT_INTER_MODES; ++idx) {
     int rate_mv = 0;
     int mode_rd_thresh;