Blends ought to be fast

include/core/SkColorPriv.h

Index: include/core/SkColorPriv.h
diff --git a/include/core/SkColorPriv.h b/include/core/SkColorPriv.h
index 8fd1e57fa7ffdf18653eda421f3cad634b37f8f8..21e17836dd0ee04d72dc79b57a86b8bd0ce87ff1 100644
--- a/include/core/SkColorPriv.h
+++ b/include/core/SkColorPriv.h
@@ -524,7 +524,9 @@ SkPMColor SkPremultiplyARGBInline(U8CPU a, U8CPU r, U8CPU g, U8CPU b) {
     return SkPackARGB32(a, r, g, b);
 }
 
-static inline uint32_t SkAlphaMulQ(uint32_t c, unsigned scale) {
+// When Android is compiled optimizing for size, SkAlphaMulQ doesn't get
+// inlined; forcing inlining significantly improves performance.
+static SK_ALWAYS_INLINE uint32_t SkAlphaMulQ(uint32_t c, unsigned scale) {
     uint32_t mask = 0xFF00FF;
 
     uint32_t rb = ((c & mask) * scale) >> 8;