vrshrn by 8 for neon. vector rounding shift right with narrowing from u16 to u8.

BUG=183
TEST=try bots
Review URL: https://webrtc-codereview.appspot.com/1269007

git-svn-id: http://libyuv.googlecode.com/svn/trunk@628 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 627
+Version: 628
 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 627
+#define LIBYUV_VERSION 628
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/row_neon.cc

@@ -2491,14 +2491,10 @@ void ARGBMultiplyRow_NEON(const uint8* src_argb0, const uint8* src_argb1,
     "vmull.u8   q1, d2, d3                     \n"  // multiply G
     "vmull.u8   q2, d4, d5                     \n"  // multiply R
     "vmull.u8   q3, d6, d7                     \n"  // multiply A
-    "vrshr.u16  q0, q0, #8                     \n"  // low 8 bits B
-    "vrshr.u16  q1, q1, #8                     \n"  // low 8 bits G
-    "vrshr.u16  q2, q2, #8                     \n"  // low 8 bits R
-    "vrshr.u16  q3, q3, #8                     \n"  // low 8 bits A
-    "vmovn.u16  d0, q0                         \n"  // 16 bit to 8 bit B
-    "vmovn.u16  d1, q1                         \n"  // 16 bit to 8 bit G
-    "vmovn.u16  d2, q2                         \n"  // 16 bit to 8 bit R
-    "vmovn.u16  d3, q3                         \n"  // 16 bit to 8 bit A
+    "vrshrn.u16 d0, q0, #8                     \n"  // 16 bit to 8 bit B
+    "vrshrn.u16 d1, q1, #8                     \n"  // 16 bit to 8 bit G
+    "vrshrn.u16 d2, q2, #8                     \n"  // 16 bit to 8 bit R
+    "vrshrn.u16 d3, q3, #8                     \n"  // 16 bit to 8 bit A
     "vst4.8     {d0, d1, d2, d3}, [%2]!        \n"  // store 8 ARGB pixels.
     "bgt        1b                             \n"