Add SSE optimization of Color32A_D565

src/core/SkBlitRow_D16.cpp

Index: src/core/SkBlitRow_D16.cpp
diff --git a/src/core/SkBlitRow_D16.cpp b/src/core/SkBlitRow_D16.cpp
index 6bf4cea9974f5d4488a65bd6aec532ccbe8096e5..5c4cce4a1ace0214542386e601f58d08be4765fb 100644
--- a/src/core/SkBlitRow_D16.cpp
+++ b/src/core/SkBlitRow_D16.cpp
@@ -9,6 +9,7 @@
 #include "SkColorPriv.h"
 #include "SkDither.h"
 #include "SkMathPriv.h"
+#include "SkUtils.h"
 
 ///////////////////////////////////////////////////////////////////////////////
 
@@ -211,6 +212,11 @@ static uint32_t pmcolor_to_expand16(SkPMColor c) {
     return (g << 24) | (r << 13) | (b << 2);
 }
 
+static void Color32_D565(uint16_t dst[], SkPMColor src, int count, int x, int y) {

[mtklein, 2015/01/30 18:17:36]

The edits to this file seem unambiguously good and orthogonal to the rest of the
change.  Why don't you spin it off so we can land it while reviewing the rest?

[henrik.smiding, 2015/02/10 15:11:51]

Done.

+    SkASSERT(count > 0);
+    sk_memset16(dst, SkPixel32ToPixel16_ToU16(src), count);
+}
+
 static void Color32A_D565(uint16_t dst[], SkPMColor src, int count, int x, int y) {
     SkASSERT(count > 0);
     uint32_t src_expand = pmcolor_to_expand16(src);
@@ -260,10 +266,10 @@ static const SkBlitRow::ColorProc16 gDefault_565_ColorProcs[] = {
     Color32_D565_Dither,
     Color32A_D565_Dither
 #else
-    // TODO: stop cheating and fill in the above specializations!
-    Color32A_D565,
-    Color32A_D565,
+    // TODO: stop cheating and fill dither from the above specializations! Use sk_dither_memset16?
+    Color32_D565,
     Color32A_D565,
+    Color32_D565,
     Color32A_D565,
 #endif
 };