Plus xfermode using Sk4px.

src/core/SkNx.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_DEFINED
 #define SkNx_DEFINED
 
@@ skipping 39 matching lines @@
     SkNi(T a, T b, T c, T d,  T e, T f, T g, T h) : fLo(a,b,c,d), fHi(e,f,g,h) { REQUIRE(N==8); }
     SkNi(T a, T b, T c, T d,  T e, T f, T g, T h,
          T i, T j, T k, T l,  T m, T n, T o, T p)
         : fLo(a,b,c,d, e,f,g,h), fHi(i,j,k,l, m,n,o,p) { REQUIRE(N==16); }
 
     void store(T vals[N]) const {
         fLo.store(vals);
         fHi.store(vals+N/2);
     }
 
+    SkNi saturatedAdd(const SkNi& o) const {
+        return SkNi(fLo.saturatedAdd(o.fLo), fHi.saturatedAdd(o.fHi));
+    }
+
     SkNi operator + (const SkNi& o) const { return SkNi(fLo + o.fLo, fHi + o.fHi); }
     SkNi operator - (const SkNi& o) const { return SkNi(fLo - o.fLo, fHi - o.fHi); }
     SkNi operator * (const SkNi& o) const { return SkNi(fLo * o.fLo, fHi * o.fHi); }
 
     SkNi operator << (int bits) const { return SkNi(fLo << bits, fHi << bits); }
     SkNi operator >> (int bits) const { return SkNi(fLo >> bits, fHi >> bits); }
 
     // TODO: comparisons, min, max?
 
     template <int k> T kth() const {
@@ skipping 89 matching lines @@
 
 template <typename T>
 class SkNi<1,T> {
 public:
     SkNi() {}
     explicit SkNi(T val) : fVal(val) {}
     static SkNi Load(const T vals[1]) { return SkNi(vals[0]); }
 
     void store(T vals[1]) const { vals[0] = fVal; }
 
+    SkNi saturatedAdd(const SkNi& o) const {
+        SkASSERT((T)(~0) > 0); // TODO: support signed T
+        T sum = fVal + o.fVal;
+        return SkNi(sum > fVal ? sum : (T)(~0));
+    }
+
     SkNi operator + (const SkNi& o) const { return SkNi(fVal + o.fVal); }
     SkNi operator - (const SkNi& o) const { return SkNi(fVal - o.fVal); }
     SkNi operator * (const SkNi& o) const { return SkNi(fVal * o.fVal); }
 
     SkNi operator << (int bits) const { return SkNi(fVal << bits); }
     SkNi operator >> (int bits) const { return SkNi(fVal >> bits); }
 
     template <int k> T kth() const {
         SkASSERT(0 == k);
         return fVal;
@@ skipping 90 matching lines @@
 typedef SkNi<4,  uint16_t> Sk4h;
 typedef SkNi<8,  uint16_t> Sk8h;
 typedef SkNi<16, uint16_t> Sk16h;
 
 typedef SkNi<16, uint8_t> Sk16b;
 
 typedef SkNi<4,  int32_t> Sk4i;
 typedef SkNi<4, uint32_t> Sk4u;
 
 #endif//SkNx_DEFINED