Support sRGB dsts in opt code

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 106 matching lines @@
     float operator[](int k) const {
         SkASSERT(0 <= k && k < 4);
         union { __m128 v; float fs[4]; } pun = {fVec};
         return pun.fs[k&3];
     }
 
     bool allTrue() const { return 0xffff == _mm_movemask_epi8(_mm_castps_si128(fVec)); }
     bool anyTrue() const { return 0x0000 != _mm_movemask_epi8(_mm_castps_si128(fVec)); }
 
     SkNx thenElse(const SkNx& t, const SkNx& e) const {
+#if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSE41
+        return _mm_blendv_ps(e.fVec, t.fVec, fVec);
+#else
         return _mm_or_ps(_mm_and_ps   (fVec, t.fVec),
                          _mm_andnot_ps(fVec, e.fVec));
+#endif
     }
 
     __m128 fVec;
 };
 
 template <>
 class SkNx<4, int> {
 public:
     SkNx(const __m128i& vec) : fVec(vec) {}
 
@@ skipping 228 matching lines @@
 
 template<> /*static*/ inline Sk4h SkNx_cast<uint16_t, uint8_t>(const Sk4b& src) {
     return _mm_unpacklo_epi8(src.fVec, _mm_setzero_si128());
 }
 
 template<> /*static*/ inline Sk4b SkNx_cast<uint8_t, uint16_t>(const Sk4h& src) {
     return _mm_packus_epi16(src.fVec, src.fVec);
 }
 
 #endif//SkNx_sse_DEFINED