could not resist: fast sse float <--> u16

src/opts/SkNx_sse.h

Index: src/opts/SkNx_sse.h
diff --git a/src/opts/SkNx_sse.h b/src/opts/SkNx_sse.h
index 71ecbfd89d9664ae6a4bded0cc44dac4dcd18618..ee6fdc5654b26838bf2e9f649741476d8d648055 100644
--- a/src/opts/SkNx_sse.h
+++ b/src/opts/SkNx_sse.h
@@ -348,6 +348,11 @@ template<> inline Sk4i SkNx_cast<int, float, 4>(const Sk4f& src) {
     return _mm_cvttps_epi32(src.fVec);
 }
 
+template<> inline Sk4h SkNx_cast<uint16_t, float, 4>(const Sk4f& src) {
+    auto _32 = _mm_cvttps_epi32(src.fVec);
+    return _mm_packus_epi16(_32, _32);
+}
+
 template<> inline Sk4b SkNx_cast<uint8_t, float, 4>(const Sk4f& src) {
     auto _32 = _mm_cvttps_epi32(src.fVec);
 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3
@@ -370,6 +375,11 @@ template<> inline Sk4f SkNx_cast<float, uint8_t, 4>(const Sk4b& src) {
     return _mm_cvtepi32_ps(_32);
 }
 
+template<> inline Sk4f SkNx_cast<float, uint16_t, 4>(const Sk4h& src) {
+    auto _32 = _mm_unpacklo_epi16(src.fVec, _mm_setzero_si128());
+    return _mm_cvtepi32_ps(_32);
+}
+
 static inline void Sk4f_ToBytes(uint8_t bytes[16],
                                 const Sk4f& a, const Sk4f& b, const Sk4f& c, const Sk4f& d) {
     _mm_storeu_si128((__m128i*)bytes,