Xfermode: SSE2 implementation of a number of simple transfer modes

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 24ab6f73c8da13653fecf560f7ca8b4fdd08103f..960c48a023126ee304f0d0505d1b09054c542b46 100644
--- a/src/opts/SkColor_opts_SSE2.h
+++ b/src/opts/SkColor_opts_SSE2.h
@@ -10,6 +10,10 @@
 
 #include <emmintrin.h>
 
+static inline __m128i SkAlpha255To256_SSE2(const __m128i& alpha) {
+    return _mm_add_epi32(alpha, _mm_set1_epi32(1));
+}
+
 // See #define SkAlphaMulAlpha(a, b)  SkMulDiv255Round(a, b) in SkXfermode.cpp.
 static inline __m128i SkAlphaMulAlpha_SSE2(const __m128i& a,
                                            const __m128i& b) {
@@ -21,6 +25,27 @@ static inline __m128i SkAlphaMulAlpha_SSE2(const __m128i& a,
     return prod;
 }
 
+// Portable version SkAlphaMulQ is in SkColorPriv.h.
+static inline __m128i SkAlphaMulQ_SSE2(const __m128i& c, const __m128i& scale) {

[mtklein, 2014/04/11 18:41:54]

Is this still the best way to do this with SSE2?   You see we're doing a sort of
4x8-bit SIMD in 32-bit ints, but we have to split them in two because the 8*8
bit multiply results in 16 bits.  Any way to perhaps just do this directly in on
SSE register instead of splitting into rb and ag?  Or I guess that would need a
non-existent _mm_mullo_epi8?

[qiankun, 2014/04/14 02:33:01]

Unfortunately, there is no 8-bits multiply funciton in SSE2 intrinsics. So keep
the current SSE2 implementation now.

+    __m128i mask = _mm_set1_epi32(gMask_00FF00FF);
+    __m128i s = _mm_or_si128(_mm_slli_epi32(scale, 16), scale);
+
+    // uint32_t rb = ((c & mask) * scale) >> 8
+    __m128i rb = _mm_and_si128(mask, c);
+    rb = _mm_mullo_epi16(rb, s);
+    rb = _mm_srli_epi16(rb, 8);
+
+    // uint32_t ag = ((c >> 8) & mask) * scale
+    __m128i ag = _mm_srli_epi16(c, 8);
+    ag = _mm_and_si128(ag, mask);
+    ag = _mm_mullo_epi16(ag, s);
+
+    // (rb & mask) | (ag & ~mask)
+    rb = _mm_and_si128(mask, rb);
+    ag = _mm_andnot_si128(mask, ag);
+    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);