Optimize RGB16 blitV functions with NEON for ARM platform.

src/core/SkBlitter_RGB16.cpp

Index: src/core/SkBlitter_RGB16.cpp
diff --git a/src/core/SkBlitter_RGB16.cpp b/src/core/SkBlitter_RGB16.cpp
index 6724b12ace423493cbb02871dc3c3307783439fa..8caf90831231fdcc61cd6c166e193a9eab1ee281 100644
--- a/src/core/SkBlitter_RGB16.cpp
+++ b/src/core/SkBlitter_RGB16.cpp
@@ -22,6 +22,11 @@ extern void blitmask_d565_opaque_mips(int width, int height, uint16_t* device,
 
 #if SK_ARM_NEON_IS_ALWAYS && defined(SK_CPU_LENDIAN)
     #include <arm_neon.h>
+extern void SkRGB16BlitterBlitV_neon(uint16_t* device,
+                                     int height,
+                                     size_t deviceRB,
+                                     unsigned scale,
+                                     uint32_t src32);
 #else
     // if we don't have neon, then our black blitter is worth the extra code
     #define USE_BLACK_BLITTER
@@ -484,11 +489,15 @@ void SkRGB16_Opaque_Blitter::blitV(int x, int y, int height, SkAlpha alpha) {
     unsigned scale5 = SkAlpha255To256(alpha) >> 3;
     uint32_t src32 =  fExpandedRaw16 * scale5;
     scale5 = 32 - scale5;
+#if SK_ARM_NEON_IS_ALWAYS && defined(SK_CPU_LENDIAN)
+    SkRGB16BlitterBlitV_neon(device, height, deviceRB, scale5, src32);
+#else
     do {
         uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
         *device = SkCompact_rgb_16((src32 + dst32) >> 5);
         device = (uint16_t*)((char*)device + deviceRB);
     } while (--height != 0);
+#endif
 }
 
 void SkRGB16_Opaque_Blitter::blitRect(int x, int y, int width, int height) {
@@ -659,11 +668,15 @@ void SkRGB16_Blitter::blitV(int x, int y, int height, SkAlpha alpha) {
     unsigned scale5 = SkAlpha255To256(alpha) * fScale >> (8 + 3);
     uint32_t src32 =  fExpandedRaw16 * scale5;
     scale5 = 32 - scale5;
+#if SK_ARM_NEON_IS_ALWAYS && defined(SK_CPU_LENDIAN)
+    SkRGB16BlitterBlitV_neon(device, height, deviceRB, scale5, src32);
+#else
     do {
         uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
         *device = SkCompact_rgb_16((src32 + dst32) >> 5);
         device = (uint16_t*)((char*)device + deviceRB);
     } while (--height != 0);
+#endif
 }
 
 void SkRGB16_Blitter::blitRect(int x, int y, int width, int height) {