Xfermode: SSE2 implementation of darken&lighten mode

src/opts/SkXfermode_opts_SSE2.cpp

Index: src/opts/SkXfermode_opts_SSE2.cpp
diff --git a/src/opts/SkXfermode_opts_SSE2.cpp b/src/opts/SkXfermode_opts_SSE2.cpp
index 01c782b9ef3df48267b4a040a1a17ec51f971997..b6eca716b2b7677359d7e2fe7317f6c91a8068e7 100644
--- a/src/opts/SkXfermode_opts_SSE2.cpp
+++ b/src/opts/SkXfermode_opts_SSE2.cpp
@@ -84,6 +84,72 @@ static __m128i multiply_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
     return SkPackARGB32_SSE2(a, r, g, b);
 }
 
+static inline __m128i darken_byte_SSE2(const __m128i& sc, const __m128i& dc,
+                                       const __m128i& sa, const __m128i& da) {
+    __m128i sd = _mm_mullo_epi16(sc, da);
+    __m128i ds = _mm_mullo_epi16(dc, sa);
+
+    __m128i cmp = _mm_cmplt_epi32(sd, ds);
+
+    __m128i ret1 = _mm_add_epi32(sc, dc);

[mtklein, 2014/04/24 17:55:12]

I'm not sure why the original code was written so oddly, but we can probably be
pretty sure that _mm_add_epi32(sc, dc) and _mm_add_epi32(dc, sc) are the same
value.  I figure it's worth saving that in its own register here rather than
recalculating?

[qiankun, 2014/04/25 14:13:24]

Done.

+    ret1 = _mm_sub_epi32(ret1, SkDiv255Round_SSE2(ds));
+
+    __m128i ret2 = _mm_add_epi32(dc, sc);
+    ret2 = _mm_sub_epi32(ret2, SkDiv255Round_SSE2(sd));
+
+    __m128i ret = _mm_or_si128(_mm_and_si128(cmp, ret1),
+                               _mm_andnot_si128(cmp, ret2));
+
+    return ret;
+}
+
+static __m128i darken_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
+    __m128i sa = SkGetPackedA32_SSE2(src);
+    __m128i da = SkGetPackedA32_SSE2(dst);
+
+    __m128i a = srcover_byte_SSE2(sa, da);
+    __m128i r = darken_byte_SSE2(SkGetPackedR32_SSE2(src),
+                                 SkGetPackedR32_SSE2(dst), sa, da);
+    __m128i g = darken_byte_SSE2(SkGetPackedG32_SSE2(src),
+                                 SkGetPackedG32_SSE2(dst), sa, da);
+    __m128i b = darken_byte_SSE2(SkGetPackedB32_SSE2(src),
+                                 SkGetPackedB32_SSE2(dst), sa, da);
+    return SkPackARGB32_SSE2(a, r, g, b);
+}
+
+static inline __m128i lighten_byte_SSE2(const __m128i& sc, const __m128i& dc,
+                                        const __m128i& sa, const __m128i& da) {
+    __m128i sd = _mm_mullo_epi16(sc, da);
+    __m128i ds = _mm_mullo_epi16(dc, sa);
+
+    __m128i cmp = _mm_cmpgt_epi32(sd, ds);
+
+    __m128i ret1 = _mm_add_epi32(sc, dc);

[mtklein, 2014/04/24 17:55:12]

Same here.

[qiankun, 2014/04/25 14:13:24]

Done.

+    ret1 = _mm_sub_epi32(ret1, SkDiv255Round_SSE2(ds));
+
+    __m128i ret2 = _mm_add_epi32(dc, sc);
+    ret2 = _mm_sub_epi32(ret2, SkDiv255Round_SSE2(sd));
+
+    __m128i ret = _mm_or_si128(_mm_and_si128(cmp, ret1),
+                               _mm_andnot_si128(cmp, ret2));
+
+    return ret;
+}
+
+static __m128i lighten_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
+    __m128i sa = SkGetPackedA32_SSE2(src);
+    __m128i da = SkGetPackedA32_SSE2(dst);
+
+    __m128i a = srcover_byte_SSE2(sa, da);
+    __m128i r = lighten_byte_SSE2(SkGetPackedR32_SSE2(src),
+                                  SkGetPackedR32_SSE2(dst), sa, da);
+    __m128i g = lighten_byte_SSE2(SkGetPackedG32_SSE2(src),
+                                  SkGetPackedG32_SSE2(dst), sa, da);
+    __m128i b = lighten_byte_SSE2(SkGetPackedB32_SSE2(src),
+                                  SkGetPackedB32_SSE2(dst), sa, da);
+    return SkPackARGB32_SSE2(a, r, g, b);
+}
+
 ////////////////////////////////////////////////////////////////////////////////
 
 typedef __m128i (*SkXfermodeProcSIMD)(const __m128i& src, const __m128i& dst);
@@ -240,8 +306,8 @@ SkXfermodeProcSIMD gSSE2XfermodeProcs[] = {
     NULL, // kScreen_Mode
 
     NULL, // kOverlay_Mode
-    NULL, // kDarken_Mode
-    NULL, // kLighten_Mode
+    darken_modeproc_SSE2,
+    lighten_modeproc_SSE2,
     NULL, // kColorDodge_Mode
     NULL, // kColorBurn_Mode
     NULL, // kHardLight_Mode