one level less of tx size search for blocks larger than 64

3~5% encoding speedup for speed 0; no quality loss.

Change-Id: I0e31755f45253e5e99d8d9eed0d7a6fe6050f49f

av1/encoder/rdopt.c

diff --git a/av1/encoder/rdopt.c b/av1/encoder/rdopt.c
index ac2bd16..44deb05 100644
--- a/av1/encoder/rdopt.c
+++ b/av1/encoder/rdopt.c
@@ -2493,8 +2493,12 @@
 static int get_search_init_depth(int mi_width, int mi_height,
                                  const SPEED_FEATURES *sf) {
   if (sf->tx_size_search_method == USE_LARGESTALL) return MAX_VARTX_DEPTH;
-  return (mi_height != mi_width) ? sf->tx_size_search_init_depth_rect
-                                 : sf->tx_size_search_init_depth_sqr;
+  int init_depth = (mi_height != mi_width) ? sf->tx_size_search_init_depth_rect
+                                           : sf->tx_size_search_init_depth_sqr;
+  // If either dimension is 128, search one level less.
+  if (mi_width == 32 || mi_height == 32)
+    init_depth = AOMMIN(init_depth + 1, MAX_TX_DEPTH);
+  return init_depth;
 }
 
 static void choose_tx_size_type_from_rd(const AV1_COMP *const cpi,