rounding for arm filter

source/scale_common.cc

Index: source/scale_common.cc
diff --git a/source/scale_common.cc b/source/scale_common.cc
index baed70b9da1f9c6d705feff6f04bdb4e1c93cc92..3507aa4d9ff0cf006eba6f7ded4c124989318816 100644
--- a/source/scale_common.cc
+++ b/source/scale_common.cc
@@ -417,11 +417,9 @@ void ScaleColsUp2_16_C(uint16* dst_ptr, const uint16* src_ptr,
 }
 
 // (1-f)a + fb can be replaced with a + f(b-a)
-#if defined(__arm__)
-// arm uses 16 bit math with truncation.
-// TODO(fbarchard): add rounding.
+#if defined(__arm__) || defined(__aarch64__)
 #define BLENDER(a, b, f) (uint8)((int)(a) + \
-    (((int)((f)) * ((int)(b) - (int)(a))) >> 16))
+    ((((int)((f)) * ((int)(b) - (int)(a))) + 0x8000) >> 16))
 #else
 // inteluses 7 bit math with rounding.
 #define BLENDER(a, b, f) (uint8)((int)(a) + \
@@ -480,7 +478,7 @@ void ScaleFilterCols64_C(uint8* dst_ptr, const uint8* src_ptr,
 
 // Same as 8 bit arm blender but return is cast to uint16
 #define BLENDER(a, b, f) (uint16)((int)(a) + \
-    (((int)((f)) * ((int)(b) - (int)(a))) >> 16))
+    ((((int)((f)) * ((int)(b) - (int)(a))) + 0x8000) >> 16))
 
 void ScaleFilterCols_16_C(uint16* dst_ptr, const uint16* src_ptr,
                        int dst_width, int x, int dx) {
@@ -818,7 +816,7 @@ void ScaleARGBColsUp2_C(uint8* dst_argb, const uint8* src_argb,
   }
 }
 
-// TODO(fbarchard): Replace 0x7f ^ f with 128-f.  bug=605.
+// TODO(fbarchard): Replace 0x7f ^ f with 128-f.  bug=607.
 // Mimics SSSE3 blender
 #define BLENDER1(a, b, f) ((a) * (0x7f ^ f) + (b) * f) >> 7
 #define BLENDERC(a, b, f, s) (uint32)( \