Plus xfermode using Sk4px.

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 278 matching lines @@
     explicit SkNi(uint8_t val) : fVec(_mm_set1_epi8(val)) {}
     static SkNi Load(const uint8_t vals[16]) { return _mm_loadu_si128((const __m128i*)vals); }
     SkNi(uint8_t a, uint8_t b, uint8_t c, uint8_t d,
          uint8_t e, uint8_t f, uint8_t g, uint8_t h,
          uint8_t i, uint8_t j, uint8_t k, uint8_t l,
          uint8_t m, uint8_t n, uint8_t o, uint8_t p)
         : fVec(_mm_setr_epi8(a,b,c,d, e,f,g,h, i,j,k,l, m,n,o,p)) {}
 
     void store(uint8_t vals[16]) const { _mm_storeu_si128((__m128i*)vals, fVec); }
 
+    SkNi saturatedAdd(const SkNi& o) const { return _mm_adds_epu8(fVec, o.fVec); }
+
     SkNi operator + (const SkNi& o) const { return _mm_add_epi8(fVec, o.fVec); }
     SkNi operator - (const SkNi& o) const { return _mm_sub_epi8(fVec, o.fVec); }
 
     // SSE cannot multiply or shift vectors of uint8_t.
     SkNi operator * (const SkNi& o) const { SkASSERT(false); return fVec; }
     SkNi operator << (int bits) const { SkASSERT(false); return fVec; }
     SkNi operator >> (int bits) const { SkASSERT(false); return fVec; }
 
     template <int k> uint8_t kth() const {
         SkASSERT(0 <= k && k < 16);
         // SSE4.1 would just `return _mm_extract_epi8(fVec, k)`.  We have to read 16-bits instead.
         int pair = _mm_extract_epi16(fVec, k/2);
         return k % 2 == 0 ? pair : (pair >> 8);
     }
 
     __m128i fVec;
 };
 
 #endif//SkNx_sse_DEFINED