Fix the bug (0 extension from int to int64) in ScaleARGBRowDownEven_NEON.

Reason of this bug is that ARM64 can't extend the sign bit of a 32-bit integer
to 64-bit integer automatically.
ScaleARGBRowDownEven_NEON is also enabled for ARM64.

BUG=319
TESTED=libyuv_unittest
R=fbarchard@google.com

Change-Id: Ib8d30a05156239247296aa8bb4faa94b4f69a9c3

Review URL: https://webrtc-codereview.appspot.com/32949004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1273 16f28f9a-4ce2-e073-06de-1de4eb20be90

source/rotate_argb.cc

@@ -31,7 +31,7 @@ void ScaleARGBRowDownEven_SSE2(const uint8* src_ptr, int src_stride,
                                uint8* dst_ptr, int dst_width);
 #endif
 #if !defined(LIBYUV_DISABLE_NEON) && !defined(__native_client__) && \
-    (defined(__ARM_NEON__) || defined(LIBYUV_NEON))
+    (defined(__ARM_NEON__) || defined(LIBYUV_NEON) || defined(__aarch64__))
 #define HAS_SCALEARGBROWDOWNEVEN_NEON
 void ScaleARGBRowDownEven_NEON(const uint8* src_ptr, int src_stride,
                                int src_stepx,

source/scale_neon64.cc

@@ -701,7 +701,7 @@ void ScaleARGBRowDownEven_NEON(const uint8* src_argb,  ptrdiff_t src_stride,
   : "+r"(src_argb),    // %0
     "+r"(dst_argb),    // %1
     "+r"(dst_width)    // %2
-  : "r"(src_stepx * 4) // %3
+  : "r"(static_cast<ptrdiff_t>(src_stepx * 4)) // %3
   : "memory", "cc", "v0"
   );
 }