Improve naive SkColorXform to half floats

src/opts/SkNx_sse.h

 /*
  * Copyright 2015 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #ifndef SkNx_sse_DEFINED
 #define SkNx_sse_DEFINED
 
@@ skipping 385 matching lines @@
 }
 
 template<> /*static*/ inline Sk4b SkNx_cast<uint8_t, int>(const Sk4i& src) {
     return _mm_packus_epi16(_mm_packus_epi16(src.fVec, src.fVec), src.fVec);
 }
 
 static inline Sk4i Sk4f_round(const Sk4f& x) {
     return _mm_cvtps_epi32(x.fVec);
 }
 
+static inline void Sk4h_store4(void* dst, const Sk4h& r, const Sk4h& g, const Sk4h& b,
+                               const Sk4h& a) {
+    __m128i rg = _mm_unpacklo_epi16(r.fVec, g.fVec);
+    __m128i ba = _mm_unpacklo_epi16(b.fVec, a.fVec);
+    __m128i lo = _mm_unpacklo_epi32(rg, ba);
+    __m128i hi = _mm_unpackhi_epi32(rg, ba);
+    _mm_storeu_si128(((__m128i*) dst) + 0, lo);
+    _mm_storeu_si128(((__m128i*) dst) + 1, hi);
+}
+
 #endif//SkNx_sse_DEFINED