Neon RGBToUV more accurate coefficients and subsample averaging. Instead of…

Neon RGBToUV more accurate coefficients and subsample averaging.  Instead of adding 4 pixels and making coefficients 4x smaller, this makes the coefficients 2x small and does a shift, for best accuracy.
BUG=297
TESTED=try bots
R=tpsiaki@google.com, yunqingwang@google.com

Review URL: https://webrtc-codereview.appspot.com/6309004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@920 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 919
+Version: 920
 License: BSD
 License File: LICENSE
 

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 919
+#define LIBYUV_VERSION 920
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/scale_neon.cc

@@ -21,6 +21,7 @@ extern "C" {
 // NEON downscalers with interpolation.
 // Provided by Fritz Koenig
 
+// Read 32x1 throw away even pixels, and write 16x1.
 void ScaleRowDown2_NEON(const uint8* src_ptr, ptrdiff_t /* src_stride */,
                         uint8* dst, int dst_width) {
   asm volatile (
@@ -39,6 +40,7 @@ void ScaleRowDown2_NEON(const uint8* src_ptr, ptrdiff_t /* src_stride */,
   );
 }
 
+// Read 32x2 average down and write 16x1.
 void ScaleRowDown2Box_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
                            uint8* dst, int dst_width) {
   asm volatile (