Fix aligned version of SSSE3 alpha blend and re-enable SSE2 alpha blend for…

Fix aligned version of SSSE3 alpha blend and re-enable SSE2 alpha blend for performance evaluation on low end CPUs where pshufb is slow.
BUG=none
TEST=none
Review URL: https://webrtc-codereview.appspot.com/728009

git-svn-id: http://libyuv.googlecode.com/svn/trunk@322 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 321
+Version: 322
 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 321
+#define LIBYUV_VERSION 322
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/row.h

@@ -80,6 +80,9 @@ extern "C" {
 #define HAS_CUMULATIVESUMTOAVERAGE_SSE2
 #define HAS_ARGBSHADE_SSE2
 #define HAS_ARGBAFFINEROW_SSE2
+// HAS_ARGBBLENDROW_SSE2 may be faster than SSSE3 version on some CPUs, so
+// enable it here instead of LIBYUV_SSSE3_ONLY section.
+#define HAS_ARGBBLENDROW_SSE2
 #endif
 
 // The following are Windows only:
@@ -96,7 +99,6 @@ extern "C" {
     !defined(LIBYUV_SSSE3_ONLY)
 #define HAS_MIRRORROW_SSE2
 #define HAS_ARGBATTENUATE_SSE2
-#define HAS_ARGBBLENDROW_SSE2
 #endif
 
 // The following are available on Neon platforms

source/row_posix.cc

@@ -2548,16 +2548,16 @@ void ARGBBlendRow_SSSE3(const uint8* src_argb0, const uint8* src_argb1,
     // 4 pixel loop.
     ".p2align  2                               \n"
   "40:                                         \n"
-    "movdqu    (%0),%%xmm3                     \n"
+    "movdqa    (%0),%%xmm3                     \n"
     "lea       0x10(%0),%0                     \n"
     "movdqa    %%xmm3,%%xmm0                   \n"
     "pxor      %%xmm4,%%xmm3                   \n"
-    "movdqu    (%1),%%xmm2                     \n"
+    "movdqa    (%1),%%xmm2                     \n"
     "pshufb    %4,%%xmm3                       \n"
     "pand      %%xmm6,%%xmm2                   \n"
     "paddw     %%xmm7,%%xmm3                   \n"
     "pmullw    %%xmm3,%%xmm2                   \n"
-    "movdqu    (%1),%%xmm1                     \n"
+    "movdqa    (%1),%%xmm1                     \n"
     "lea       0x10(%1),%1                     \n"
     "psrlw     $0x8,%%xmm1                     \n"
     "por       %%xmm4,%%xmm0                   \n"