Cleanup with SkAlphaMulQ_SSE2()

src/opts/SkColor_opts_SSE2.h

Index: src/opts/SkColor_opts_SSE2.h
diff --git a/src/opts/SkColor_opts_SSE2.h b/src/opts/SkColor_opts_SSE2.h
index 741d1ab77732cdfb41c1a8f8a4aa2f8ca426484b..c14b9ada900f6e26cabac1c2ef6b52c0ab993a74 100644
--- a/src/opts/SkColor_opts_SSE2.h
+++ b/src/opts/SkColor_opts_SSE2.h
@@ -63,6 +63,23 @@ static inline __m128i SkAlphaMulQ_SSE2(const __m128i& c, const __m128i& scale) {
     return _mm_or_si128(rb, ag);
 }
 
+// Fast path for colors scale the same factor.

[mtklein, 2014/11/24 16:48:23]

Maybe  // Fast path for SkAlphaMulQ_SSE2 with a constant scale factor.

[qiankun, 2014/11/25 05:33:52]

Done.

+static inline __m128i SkAlphaMulQ_SSE2(const __m128i& c, const unsigned scale) {
+    __m128i mask = _mm_set1_epi32(0xFF00FF);
+    __m128i s = _mm_set1_epi16(scale << 8); // Move scale factor to upper byte of word.
+
+    // With mulhi, red and blue values are already in the right place and
+    // don't need to be divided by 256.
+    __m128i rb = _mm_and_si128(mask, c);
+    rb = _mm_mulhi_epu16(rb, s);
+
+    __m128i ag = _mm_srli_epi16(c, 8);
+    ag = _mm_mulhi_epu16(ag, s); // Alpha and green values are in the lower byte of each word.
+    ag = _mm_slli_epi16(ag, 8);  // Shift alpha and green to higher byte of each word.
+
+    return _mm_or_si128(rb, ag);
+}
+
 static inline __m128i SkGetPackedA32_SSE2(const __m128i& src) {
     __m128i a = _mm_slli_epi32(src, (24 - SK_A32_SHIFT));
     return _mm_srli_epi32(a, 24);