Implement a NEON version of morphology. This is good for ~2.2X speedup on Tegra3.

src/opts/SkMorphology_opts_neon.cpp

Index: src/opts/SkMorphology_opts_neon.cpp
diff --git a/src/opts/SkMorphology_opts_SSE2.cpp b/src/opts/SkMorphology_opts_neon.cpp
similarity index 63%
copy from src/opts/SkMorphology_opts_SSE2.cpp
copy to src/opts/SkMorphology_opts_neon.cpp
index b58fced2c12718f8b84b42f948b8f85a6b914415..571b5c8b178b6463705a4b9b3b22e10423485667 100644
--- a/src/opts/SkMorphology_opts_SSE2.cpp
+++ b/src/opts/SkMorphology_opts_neon.cpp
@@ -7,11 +7,12 @@
 
 
 #include "SkColorPriv.h"
-#include "SkMorphology_opts_SSE2.h"
+#include "SkMorphology_opts.h"
+#include "SkMorphology_opts_neon.h"
 
-#include <emmintrin.h>
+#include <arm_neon.h>
 
-/* SSE2 version of dilateX, dilateY, erodeX, erodeY.
+/* neon version of dilateX, dilateY, erodeX, erodeY.
  * portable versions are in src/effects/SkMorphologyImageFilter.cpp.
  */
 
@@ -24,7 +25,7 @@ enum MorphDirection {
 };
 
 template<MorphType type, MorphDirection direction>
-static void SkMorph_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
+static void SkMorph_neon(const SkPMColor* src, SkPMColor* dst, int radius,
                          int width, int height, int srcStride, int dstStride)
 {
     const int srcStrideX = direction == kX ? 1 : srcStride;
@@ -38,12 +39,12 @@ static void SkMorph_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
         const SkPMColor* up = upperSrc;
         SkPMColor* dptr = dst;
         for (int y = 0; y < height; ++y) {
-            __m128i max = type == kDilate ? _mm_setzero_si128() : _mm_set1_epi32(0xFFFFFFFF);
+            uint8x8_t max = vdup_n_u8(type == kDilate ? 0 : 255);
             for (const SkPMColor* p = lp; p <= up; p += srcStrideX) {
-                __m128i src_pixel = _mm_cvtsi32_si128(*p);
-                max = type == kDilate ? _mm_max_epu8(src_pixel, max) : _mm_min_epu8(src_pixel, max);
+                uint8x8_t src_pixel = vreinterpret_u8_u32(vdup_n_u32(*p));
+                max = type == kDilate ? vmax_u8(src_pixel, max) : vmin_u8(src_pixel, max);
             }
-            *dptr = _mm_cvtsi128_si32(max);
+            *dptr = vget_lane_u32(vreinterpret_u32_u8(max), 0);
             dptr += dstStrideY;
             lp += srcStrideY;
             up += srcStrideY;
@@ -54,26 +55,26 @@ static void SkMorph_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
     }
 }
 
-void SkDilateX_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
+void SkDilateX_neon(const SkPMColor* src, SkPMColor* dst, int radius,
                     int width, int height, int srcStride, int dstStride)
 {
-    SkMorph_SSE2<kDilate, kX>(src, dst, radius, width, height, srcStride, dstStride);
+    SkMorph_neon<kDilate, kX>(src, dst, radius, width, height, srcStride, dstStride);
 }
 
-void SkErodeX_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
+void SkErodeX_neon(const SkPMColor* src, SkPMColor* dst, int radius,
                    int width, int height, int srcStride, int dstStride)
 {
-    SkMorph_SSE2<kErode, kX>(src, dst, radius, width, height, srcStride, dstStride);
+    SkMorph_neon<kErode, kX>(src, dst, radius, width, height, srcStride, dstStride);
 }
 
-void SkDilateY_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
+void SkDilateY_neon(const SkPMColor* src, SkPMColor* dst, int radius,
                     int width, int height, int srcStride, int dstStride)
 {
-    SkMorph_SSE2<kDilate, kY>(src, dst, radius, width, height, srcStride, dstStride);
+    SkMorph_neon<kDilate, kY>(src, dst, radius, width, height, srcStride, dstStride);
 }
 
-void SkErodeY_SSE2(const SkPMColor* src, SkPMColor* dst, int radius,
+void SkErodeY_neon(const SkPMColor* src, SkPMColor* dst, int radius,
                    int width, int height, int srcStride, int dstStride)
 {
-    SkMorph_SSE2<kErode, kY>(src, dst, radius, width, height, srcStride, dstStride);
+    SkMorph_neon<kErode, kY>(src, dst, radius, width, height, srcStride, dstStride);
 }