inline version of alpha blend

BUG=none
TEST=none
Review URL: https://webrtc-codereview.appspot.com/446006

git-svn-id: http://libyuv.googlecode.com/svn/trunk@216 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 214
+Version: 216
 License: BSD
 License File: LICENSE
 

include/libyuv/version.h

@@ -11,7 +11,7 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 214
+#define LIBYUV_VERSION 216
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_
 

source/planar_functions.cc

@@ -170,6 +170,10 @@ int ARGBBlend(const uint8* src_argb, int src_stride_argb,
 #if defined(HAS_ARGBBLENDROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     ARGBBlendRow = ARGBBlendRow_SSE2;
+    if (IS_ALIGNED(width, 4) &&
+        IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
+      ARGBBlendRow = ARGBBlendRow_Aligned_SSE2;
+    }
   }
 #endif
 

source/row.h

@@ -240,10 +240,13 @@ void YToARGBRow_SSE2(const uint8* y_buf,
                      uint8* rgb_buf,
                      int width);
 
+// ARGB preattenuated alpha blend.
+void ARGBBlendRow_Aligned_SSE2(const uint8* src_argb, uint8* dst_argb,
+                               int width);
 void ARGBBlendRow_SSE2(const uint8* src_argb, uint8* dst_argb, int width);
 void ARGBBlendRow_C(const uint8* src_argb, uint8* dst_argb, int width);
 
-// 'Any' wrappers use memcpy()
+// 'Any' functions handle any size and alignment.
 void I420ToARGBRow_Any_SSSE3(const uint8* y_buf,
                              const uint8* u_buf,
                              const uint8* v_buf,

source/row_common.cc

@@ -452,7 +452,7 @@ void UYVYToYRow_C(const uint8* src_yuy2, uint8* dst_y, int width) {
   }
 }
 
-#define BLENDER(f, b, a) (f * a + b * (a ^ 0xff) + 0x80) >> 8
+#define BLENDER(f, b, a) (((256 - a) * b) >> 8) + f
 void ARGBBlendRow_C(const uint8* src_argb, uint8* dst_argb, int width) {
   for (int x = 0; x < width - 1; x += 2) {
     uint32 a = src_argb[3];