Optimize RGB16 blitH 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 8caf90831231fdcc61cd6c166e193a9eab1ee281..e3b418edc68b5d10de0a061077ba629bf2d30b92 100644
--- a/src/core/SkBlitter_RGB16.cpp
+++ b/src/core/SkBlitter_RGB16.cpp
@@ -27,6 +27,10 @@ extern void SkRGB16BlitterBlitV_neon(uint16_t* device,
                                      size_t deviceRB,
                                      unsigned scale,
                                      uint32_t src32);
+extern void SkRGB16BlitterBlitH_neon(uint16_t* device,
+                                     int height,
+                                     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
@@ -338,10 +342,14 @@ void SkRGB16_Opaque_Blitter::blitAntiH(int x, int y,
                 uint32_t src32 = srcExpanded * scale5;
                 scale5 = 32 - scale5; // now we can use it on the device
                 int n = count;
+#if SK_ARM_NEON_IS_ALWAYS && defined(SK_CPU_LENDIAN)
+                SkRGB16BlitterBlitH_neon(device, n, scale5, src32);
+#else
                 do {
                     uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
                     *device++ = SkCompact_rgb_16((src32 + dst32) >> 5);
                 } while (--n != 0);
+#endif
                 goto DONE;
             }
         }
@@ -601,10 +609,14 @@ void SkRGB16_Blitter::blitAntiH(int x, int y,
             unsigned scale5 = SkAlpha255To256(aa) * scale >> (8 + 3);
             uint32_t src32 =  srcExpanded * scale5;
             scale5 = 32 - scale5;
+#if SK_ARM_NEON_IS_ALWAYS && defined(SK_CPU_LENDIAN)

[mtklein, 2015/07/29 14:07:11]

This is fine, but we basically make no attempt to support big-endian anymore. 
I'm not sure it'd even compile.

[yang.zhang, 2015/07/30 05:35:34]

I have verified that this NEON function can be compiled.

[mtklein, 2015/07/30 12:29:05]

(Sorry, I meant "whether Skia'd even compile for big-endian".  Anyway, no big
deal.)

+            SkRGB16BlitterBlitH_neon(device, count, scale5, src32);
+#else
             do {
                 uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
                 *device++ = SkCompact_rgb_16((src32 + dst32) >> 5);
             } while (--count != 0);
+#endif
             continue;
         }
         device += count;