div255(x) as ((x+128)*257)>>16 with SSE

src/opts/Sk4px_SSE2.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.
  */
 
 namespace { // See Sk4px.h
 
 inline Sk4px Sk4px::DupPMColor(SkPMColor px) { return Sk16b(_mm_set1_epi32(px)); }
@@ skipping 27 matching lines @@
 
 inline Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const {
     return this->widenLo() * Sk4px(other).widenLo();
 }
 
 inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
     Sk4px::Wide r = (*this + other) >> 8;
     return Sk4px(_mm_packus_epi16(r.fLo.fVec, r.fHi.fVec));
 }
 
+inline Sk4px Sk4px::Wide::div255() const {
+    // (x + 127) / 255 == ((x+128) * 257)>>16,
+    // and _mm_mulhi_epu16 makes the (_ * 257)>>16 part very convenient.
+    const __m128i _128 = _mm_set1_epi16(128),
+                  _257 = _mm_set1_epi16(257);
+    return Sk4px(_mm_packus_epi16(_mm_mulhi_epu16(_mm_add_epi16(fLo.fVec, _128), _257),
+                                  _mm_mulhi_epu16(_mm_add_epi16(fHi.fVec, _128), _257)));
+}
+
 // Load4Alphas and Load2Alphas use possibly-unaligned loads (SkAlpha[] -> uint16_t or uint32_t).
 // These are safe on x86, often with no speed penalty.
 
 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3
     inline Sk4px Sk4px::alphas() const {
         static_assert(SK_A32_SHIFT == 24, "Intel's always little-endian.");
         __m128i splat = _mm_set_epi8(15,15,15,15, 11,11,11,11, 7,7,7,7, 3,3,3,3);
         return Sk16b(_mm_shuffle_epi8(this->fVec, splat));
     }
 
@@ skipping 103 matching lines @@
     uint32_t dst2 = _mm_cvtsi128_si32(narrow_to_565(this->fVec));
     dst[0] = dst2;
     dst[1] = dst2 >> 16;
 }
 inline void Sk4px::store1(SkPMColor16 dst[1]) const {
     uint32_t dst2 = _mm_cvtsi128_si32(narrow_to_565(this->fVec));
     dst[0] = dst2;
 }
 
 }  // namespace