Revert of Revert of SSE4 opaque blend using intrinsics instead of assembly.

src/opts/SkColor_opts_SSE2.h

Index: src/opts/SkColor_opts_SSE2.h
diff --git a/src/opts/SkColor_opts_SSE2.h b/src/opts/SkColor_opts_SSE2.h
index 970abb859b980dff84dd4f1085ddd967ac2b7d3e..feb1d98f8df3c894b912ea7279c15c12b6e51995 100644
--- a/src/opts/SkColor_opts_SSE2.h
+++ b/src/opts/SkColor_opts_SSE2.h
@@ -206,7 +206,14 @@
     return d_pixel;
 }
 
-// Portable version SkBlendARGB32 is in SkColorPriv.h.
+// Portable version is SkPMSrcOver in SkColorPriv.h.
+static inline __m128i SkPMSrcOver_SSE2(const __m128i& src, const __m128i& dst) {
+    return _mm_add_epi32(src,
+                         SkAlphaMulQ_SSE2(dst, _mm_sub_epi32(_mm_set1_epi32(256),
+                                                             SkGetPackedA32_SSE2(src))));
+}
+
+// Portable version is SkBlendARGB32 in SkColorPriv.h.
 static inline __m128i SkBlendARGB32_SSE2(const __m128i& src, const __m128i& dst,
                                          const __m128i& aa) {
     __m128i src_scale = SkAlpha255To256_SSE2(aa);