3-15% speedup to HardLight / Overlay xfermodes.

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 13 matching lines @@
 inline Sk4px::Wide Sk4px::widenLo() const {
     return Sk16h(_mm_unpacklo_epi8(this->fVec, _mm_setzero_si128()),
                  _mm_unpackhi_epi8(this->fVec, _mm_setzero_si128()));
 }
 
 inline Sk4px::Wide Sk4px::widenHi() const {
     return Sk16h(_mm_unpacklo_epi8(_mm_setzero_si128(), this->fVec),
                  _mm_unpackhi_epi8(_mm_setzero_si128(), this->fVec));
 }
 
+inline Sk4px::Wide Sk4px::widenLoHi() const {
+    return Sk16h(_mm_unpacklo_epi8(this->fVec, this->fVec),
+                 _mm_unpackhi_epi8(this->fVec, this->fVec));
+}
+
 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));
 }
 
 // Load4Alphas and Load2Alphas use possibly-unaligned loads (SkAlpha[] -> uint16_t or uint32_t).
@@ skipping 38 matching lines @@
 inline Sk4px Sk4px::zeroColors() const {
     return Sk16b(_mm_and_si128(_mm_set1_epi32(0xFF << SK_A32_SHIFT), this->fVec));
 }
 
 inline Sk4px Sk4px::zeroAlphas() const {
     // andnot(a,b) == ~a & b
     return Sk16b(_mm_andnot_si128(_mm_set1_epi32(0xFF << SK_A32_SHIFT), this->fVec));
 }
 
 }  // namespace