The compiler can generate smulbb perfectly well nowadays.

include/core/SkColorPriv.h

Index: include/core/SkColorPriv.h
diff --git a/include/core/SkColorPriv.h b/include/core/SkColorPriv.h
index f9c5d928a0b06b842741ca112fe395f269a6ca40..3dec49b73e06846f21d13e43cb2082af9e345244 100644
--- a/include/core/SkColorPriv.h
+++ b/include/core/SkColorPriv.h
@@ -193,7 +193,7 @@ static inline unsigned Sk255To256(U8CPU value) {
 /** Multiplify value by 0..256, and shift the result down 8
     (i.e. return (value * alpha256) >> 8)
  */
-#define SkAlphaMul(value, alpha256)     (SkMulS16(value, alpha256) >> 8)
+#define SkAlphaMul(value, alpha256)     (((value) * (alpha256)) >> 8)
 
 //  The caller may want negative values, so keep all params signed (int)
 //  so we don't accidentally slip into unsigned math and lose the sign
@@ -213,7 +213,7 @@ static inline int SkAlphaBlend255(S16CPU src, S16CPU dst, U8CPU alpha) {
     SkASSERT((int16_t)dst == dst);
     SkASSERT((uint8_t)alpha == alpha);
 
-    int prod = SkMulS16(src - dst, alpha) + 128;
+    int prod = (src - dst) * alpha + 128;
     prod = (prod + (prod >> 8)) >> 8;
     return dst + prod;
 }