Extend prune_zero_mv_with_sse speed feature to speed 3

This patch extends SF prune_zero_mv_with_sse to speed 3
with reduced aggressiveness.

cpu-used   Instruction Count       BD-Rate Loss(%)
              Reduction(%)    avg.psnr   ovr.psnr   ssim
    3           1.093         -0.0162    -0.0093   -0.0038

STATS_CHANGED

Change-Id: I95bfe546653722db870201578642e74fca9ac415

av1/encoder/rdopt.c

diff --git a/av1/encoder/rdopt.c b/av1/encoder/rdopt.c
index 7cd8e54..acf1db3 100644
--- a/av1/encoder/rdopt.c
+++ b/av1/encoder/rdopt.c
@@ -2468,11 +2468,13 @@
  * \param[in]     bsize             The current block_size.
  * \param[in]     args              The args to handle_inter_mode, used to track
  *                                  the best SSE.
+ * \param[in]    prune_zero_mv_with_sse  The argument holds speed feature
+ *                                       prune_zero_mv_with_sse value
  * \return Returns 1 if zero_mv is pruned, 0 otherwise.
  */
 static AOM_INLINE int prune_zero_mv_with_sse(
     const aom_variance_fn_ptr_t *fn_ptr, const MACROBLOCK *x, BLOCK_SIZE bsize,
-    const HandleInterModeArgs *args) {
+    const HandleInterModeArgs *args, int prune_zero_mv_with_sse) {
   const MACROBLOCKD *xd = &x->e_mbd;
   const MB_MODE_INFO *mbmi = xd->mi[0];
 
@@ -2513,7 +2515,9 @@
     const unsigned int best_sse = args->best_single_sse_in_refs[refs[idx]];
     best_sse_sum += best_sse;
   }
-  if (this_sse_sum > best_sse_sum) {
+
+  const double mul = prune_zero_mv_with_sse > 1 ? 1.00 : 1.25;
+  if ((double)this_sse_sum > (mul * (double)best_sse_sum)) {
     return 1;
   }
 
@@ -2931,7 +2935,8 @@
     if (cpi->sf.gm_sf.prune_zero_mv_with_sse &&
         cpi->sf.gm_sf.gm_search_type == GM_DISABLE_SEARCH &&
         (this_mode == GLOBALMV || this_mode == GLOBAL_GLOBALMV)) {
-      if (prune_zero_mv_with_sse(cpi->ppi->fn_ptr, x, bsize, args)) {
+      if (prune_zero_mv_with_sse(cpi->ppi->fn_ptr, x, bsize, args,
+                                 cpi->sf.gm_sf.prune_zero_mv_with_sse)) {
         continue;
       }
     }

av1/encoder/speed_features.c

diff --git a/av1/encoder/speed_features.c b/av1/encoder/speed_features.c
index e004bb6..e6199fa 100644
--- a/av1/encoder/speed_features.c
+++ b/av1/encoder/speed_features.c
@@ -983,6 +983,7 @@
     sf->hl_sf.high_precision_mv_usage = CURRENT_Q;
 
     sf->gm_sf.gm_search_type = GM_DISABLE_SEARCH;
+    sf->gm_sf.prune_zero_mv_with_sse = 1;
 
     sf->part_sf.less_rectangular_check_level = 2;
     sf->part_sf.simple_motion_search_prune_agg =
@@ -1055,7 +1056,7 @@
   }
 
   if (speed >= 4) {
-    sf->gm_sf.prune_zero_mv_with_sse = 1;
+    sf->gm_sf.prune_zero_mv_with_sse = 2;
 
     sf->mv_sf.subpel_search_method = SUBPEL_TREE_PRUNED_MORE;
 

av1/encoder/speed_features.h

diff --git a/av1/encoder/speed_features.h b/av1/encoder/speed_features.h
index b09599f..087e56d 100644
--- a/av1/encoder/speed_features.h
+++ b/av1/encoder/speed_features.h
@@ -477,6 +477,9 @@
 
   // When the current GM type is set to ZEROMV, prune ZEROMV if its performance
   // is worse than NEWMV under SSE metric.
+  // 0 : no pruning
+  // 1 : conservative pruning
+  // 2 : aggressive pruning
   int prune_zero_mv_with_sse;
 
   // Disable global motion estimation based on stats of previous frames in the