Blends ought to be fast

include/core/SkColorPriv.h

 /*
  * Copyright 2006 The Android Open Source Project
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #ifndef SkColorPriv_DEFINED
 #define SkColorPriv_DEFINED
 
@@ skipping 506 matching lines @@
     SkB32Assert(b);
 
     if (a != 255) {
         r = SkMulDiv255Round(r, a);
         g = SkMulDiv255Round(g, a);
         b = SkMulDiv255Round(b, a);
     }
     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;
     uint32_t ag = ((c >> 8) & mask) * scale;
     return (rb & mask) | (ag & ~mask);
 }
 
 static inline SkPMColor SkPMSrcOver(SkPMColor src, SkPMColor dst) {
     return src + SkAlphaMulQ(dst, SkAlpha255To256(255 - SkGetPackedA32(src)));
 }
@@ skipping 510 matching lines @@
     int srcG = SkColorGetG(src);
     int srcB = SkColorGetB(src);
 
     for (int i = 0; i < width; i++) {
         dst[i] = SkBlendLCD16Opaque(srcR, srcG, srcB, dst[i], mask[i],
                                     opaqueDst);
     }
 }
 
 #endif