Simpler version of Plus w/ AA. ~25% faster too.

src/core/Sk4pxXfermode.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 Sk4pxXfermode_DEFINED
 #define Sk4pxXfermode_DEFINED
 
@@ skipping 71 matching lines @@
     return Sk4px::Wide(noAA.mulWiden(aa) + d.mulWiden(Sk4px(aa).inv()))
         .div255RoundNarrow();
 }
 
 // For some transfermodes we specialize AA, either for correctness or performance.
 #ifndef SK_NO_SPECIALIZED_AA_XFERMODES
     #define XFERMODE_AA(Name) \
         template <> Sk4px xfer_aa<Name>(const Sk4px& s, const Sk4px& d, const Sk16b& aa)
 
     // Plus' clamp needs to happen after AA.  skia:3852
-    XFERMODE_AA(Plus) {  // [ clamp(D + AA*S) ]
-        // We implement this as D + Min(S*AA, (1-D)) to fit the arguments to Min in 16 bits.
-        return d +
-            Sk4px::Wide(Sk16h::Min(s.mulWiden(aa), d.inv().mul255Widen())).div255RoundNarrow();
+    XFERMODE_AA(Plus) {  // [ clamp( (1-AA)D + (AA)(S+D) ) == clamp(D + AA*S) ]
+        return d.saturatedAdd(s.fastMulDiv255Round(aa));
     }
 
     #undef XFERMODE_AA
 #endif
 
 template <typename ProcType>
 class SkT4pxXfermode : public SkProcCoeffXfermode {
 public:
     static SkProcCoeffXfermode* Create(const ProcCoeff& rec) {
         return SkNEW_ARGS(SkT4pxXfermode, (rec));
@@ skipping 41 matching lines @@
         case SkXfermode::kExclusion_Mode:  return SkT4pxXfermode<Exclusion>::Create(rec);
         default: break;
     }
 #endif
     return nullptr;
 }
 
 } // namespace
 
 #endif//Sk4pxXfermode_DEFINED