Only search the largest tx_size for cbs above 64

If any dimension of a coding block size is above 64, only search
the largest transform block size, since the max transform block
size supported is 64x64.

Bring back this under a speed 1 feature. The encoding goes 2.5%
faster.

Stats change expected.

Change-Id: I50ed7565cd4f83ff78c83ea6c80d56f5ef3cbb6d
diff --git a/av1/encoder/rdopt.c b/av1/encoder/rdopt.c
index e94c797..51403f8 100644
--- a/av1/encoder/rdopt.c
+++ b/av1/encoder/rdopt.c
@@ -2455,6 +2455,13 @@
 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;
+
+  if (sf->tx_size_search_lgr_block) {
+    if (mi_width > mi_size_wide[BLOCK_64X64] ||
+        mi_height > mi_size_high[BLOCK_64X64])
+      return MAX_VARTX_DEPTH;
+  }
+
   return (mi_height != mi_width) ? sf->tx_size_search_init_depth_rect
                                  : sf->tx_size_search_init_depth_sqr;
 }
diff --git a/av1/encoder/speed_features.c b/av1/encoder/speed_features.c
index 072b8e7..21cbff7 100644
--- a/av1/encoder/speed_features.c
+++ b/av1/encoder/speed_features.c
@@ -146,6 +146,7 @@
     sf->selective_ref_frame = 1;
     sf->tx_size_search_init_depth_rect = 1;
     sf->tx_size_search_init_depth_sqr = 1;
+    sf->tx_size_search_lgr_block = 1;
     sf->two_pass_partition_search = 1;
     sf->prune_ext_partition_types_search = 1;
     sf->use_fast_interpolation_filter_search = 1;
@@ -411,6 +412,7 @@
   sf->tx_size_search_method = USE_FULL_RD;
   sf->tx_size_search_init_depth_sqr = 0;
   sf->tx_size_search_init_depth_rect = 0;
+  sf->tx_size_search_lgr_block = 0;
   sf->reduce_inter_modes = 0;
   sf->adaptive_motion_search = 0;
   sf->adaptive_pred_interp_filter = 0;
diff --git a/av1/encoder/speed_features.h b/av1/encoder/speed_features.h
index 6d82381..6e52618 100644
--- a/av1/encoder/speed_features.h
+++ b/av1/encoder/speed_features.h
@@ -346,6 +346,9 @@
   // 0 - search full tree, 1: search 1 level, 2: search the highest level only
   int tx_size_search_init_depth_sqr;
   int tx_size_search_init_depth_rect;
+  // If any dimension of a coding block size above 64, always search the
+  // largest transform only, since the largest transform block size is 64x64.
+  int tx_size_search_lgr_block;
 
   // After looking at the first set of modes (set by index here), skip
   // checking modes for reference frames that don't match the reference frame