fast sk4f <-> sk4i SSE methods

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 276 matching lines @@
     template <int k> uint8_t kth() const { return (*this)[k]; }
 
     SkNx thenElse(const SkNx& t, const SkNx& e) const {
         return _mm_or_si128(_mm_and_si128   (fVec, t.fVec),
                             _mm_andnot_si128(fVec, e.fVec));
     }
 
     __m128i fVec;
 };
 
+template<> /*static*/ inline Sk4f SkNx_cast<float, int>(const Sk4i& src) {
+    return _mm_cvtepi32_ps(src.fVec);
+}
+
+template <> /*static*/ inline Sk4i SkNx_cast<int, float>(const Sk4f& src) {
+    return _mm_cvttps_epi32(src.fVec);
+}
 
 template<> /*static*/ inline Sk4h SkNx_cast<uint16_t, float>(const Sk4f& src) {
     auto _32 = _mm_cvttps_epi32(src.fVec);
     // Ideally we'd use _mm_packus_epi32 here.  But that's SSE4.1+.
 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3
     // With SSSE3, we can just shuffle the low 2 bytes from each lane right into place.
     const int _ = ~0;
     return _mm_shuffle_epi8(_32, _mm_setr_epi8(0,1, 4,5, 8,9, 12,13, _,_,_,_,_,_,_,_));
 #else
     // With SSE2, we have to emulate _mm_packus_epi32 with _mm_packs_epi32:
@@ skipping 40 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