Merge changes I320e160e,Iddd99733 into experimental

* changes:
  Removing rounding from UV MV calculation
  make buid_inter_predictors block size agnostic (luma)

vp9/encoder/vp9_mcomp.c

diff --git a/vp9/encoder/vp9_mcomp.c b/vp9/encoder/vp9_mcomp.c
index caba2ea..17dafe6 100644
--- a/vp9/encoder/vp9_mcomp.c
+++ b/vp9/encoder/vp9_mcomp.c
@@ -56,9 +56,9 @@
   MV v;
   v.row = mv->as_mv.row - ref->as_mv.row;
   v.col = mv->as_mv.col - ref->as_mv.col;
-  return ((mvjcost[vp9_get_mv_joint(&v)] +
-           mvcost[0][v.row] +
-           mvcost[1][v.col]) * weight) >> 7;
+  return ROUND_POWER_OF_TWO((mvjcost[vp9_get_mv_joint(&v)] +
+                             mvcost[0][v.row] +
+                             mvcost[1][v.col]) * weight, 7);
 }
 
 static int mv_err_cost(int_mv *mv, int_mv *ref, int *mvjcost, int *mvcost[2],