Xfermode: SSE2 implementation of a number of simple transfer modes

These modes share some common code and not very complex, so group them
together. This CL yields about 50% performance improvement on desktop
i7-3770. Here are the data:
before:
   Xfermode_Screen   8888:  cmsecs =     30.25   565:  cmsecs =     46.81
 Xfermode_Modulate   8888:  cmsecs =     22.48   565:  cmsecs =     40.06
     Xfermode_Plus   8888:  cmsecs =     21.04   565:  cmsecs =     37.51
      Xfermode_Xor   8888:  cmsecs =     37.18   565:  cmsecs =     52.53
  Xfermode_DstATop   8888:  cmsecs =     28.97   565:  cmsecs =     46.42
  Xfermode_SrcATop   8888:  cmsecs =     29.74   565:  cmsecs =     46.25
   Xfermode_DstOut   8888:  cmsecs =      5.34   565:  cmsecs =     24.53
   Xfermode_SrcOut   8888:  cmsecs =     12.25   565:  cmsecs =     24.39
    Xfermode_DstIn   8888:  cmsecs =      5.30   565:  cmsecs =     24.50
    Xfermode_SrcIn   8888:  cmsecs =     12.05   565:  cmsecs =     25.40
  Xfermode_DstOver   8888:  cmsecs =     12.45   565:  cmsecs =      0.15
  Xfermode_SrcOver   8888:  cmsecs =      2.68   565:  cmsecs =      4.42
after:
   Xfermode_Screen   8888:  cmsecs =     13.68   565:  cmsecs =     21.73
 Xfermode_Modulate   8888:  cmsecs =     13.25   565:  cmsecs =     20.97
     Xfermode_Plus   8888:  cmsecs =      9.77   565:  cmsecs =     16.71
      Xfermode_Xor   8888:  cmsecs =     17.64   565:  cmsecs =     25.62
  Xfermode_DstATop   8888:  cmsecs =     15.99   565:  cmsecs =     23.74
  Xfermode_SrcATop   8888:  cmsecs =     15.69   565:  cmsecs =     23.40
   Xfermode_DstOut   8888:  cmsecs =      4.77   565:  cmsecs =     11.85
   Xfermode_SrcOut   8888:  cmsecs =      4.98   565:  cmsecs =     11.84
    Xfermode_DstIn   8888:  cmsecs =      4.68   565:  cmsecs =     11.72
    Xfermode_SrcIn   8888:  cmsecs =      4.93   565:  cmsecs =     11.79
  Xfermode_DstOver   8888:  cmsecs =      5.04   565:  cmsecs =      0.15
  Xfermode_SrcOver   8888:  cmsecs =      2.69   565:  cmsecs =      4.42

BUG=skia:
R=mtklein@google.com

Author: qiankun.miao@intel.com

Review URL: https://codereview.chromium.org/232793002

git-svn-id: http://skia.googlecode.com/svn/trunk@14176 2bbb7eff-a529-9590-31e7-b0007b416f81

2 files changed, 182 insertions, 12 deletions

diff --git a/src/opts/SkColor_opts_SSE2.h b/src/opts/SkColor_opts_SSE2.h
index 24ab6f73c8..960c48a023 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) {
+    __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);
diff --git a/src/opts/SkXfermode_opts_SSE2.cpp b/src/opts/SkXfermode_opts_SSE2.cpp
index 01c782b9ef..d1a2632fdb 100644
--- a/src/opts/SkXfermode_opts_SSE2.cpp
+++ b/src/opts/SkXfermode_opts_SSE2.cpp
@@ -17,6 +17,15 @@ static inline __m128i SkDiv255Round_SSE2(const __m128i& a) {
     return prod;
 }
 
+static inline __m128i saturated_add_SSE2(const __m128i& a, const __m128i& b) {
+    __m128i sum = _mm_add_epi32(a, b);
+    __m128i cmp = _mm_cmpgt_epi32(sum, _mm_set1_epi32(255));
+
+    sum = _mm_or_si128(_mm_and_si128(cmp, _mm_set1_epi32(255)),
+                       _mm_andnot_si128(cmp, sum));
+    return sum;
+}
+
 static inline __m128i clamp_div255round_SSE2(const __m128i& prod) {
     // test if > 0
     __m128i cmp1 = _mm_cmpgt_epi32(prod, _mm_setzero_si128());
@@ -38,6 +47,130 @@ static inline __m128i clamp_div255round_SSE2(const __m128i& prod) {
     return ret;
 }
 
+static __m128i srcover_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
+    __m128i isa = _mm_sub_epi32(_mm_set1_epi32(256), SkGetPackedA32_SSE2(src));
+    return _mm_add_epi32(src, SkAlphaMulQ_SSE2(dst, isa));
+}
+
+static __m128i dstover_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
+    __m128i ida = _mm_sub_epi32(_mm_set1_epi32(256), SkGetPackedA32_SSE2(dst));
+    return _mm_add_epi32(dst, SkAlphaMulQ_SSE2(src, ida));
+}
+
+static __m128i srcin_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
+    __m128i da = SkGetPackedA32_SSE2(dst);
+    return SkAlphaMulQ_SSE2(src, SkAlpha255To256_SSE2(da));
+}
+
+static __m128i dstin_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
+    __m128i sa = SkGetPackedA32_SSE2(src);
+    return SkAlphaMulQ_SSE2(dst, SkAlpha255To256_SSE2(sa));
+}
+
+static __m128i srcout_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
+    __m128i ida = _mm_sub_epi32(_mm_set1_epi32(256), SkGetPackedA32_SSE2(dst));
+    return SkAlphaMulQ_SSE2(src, ida);
+}
+
+static __m128i dstout_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
+    __m128i isa = _mm_sub_epi32(_mm_set1_epi32(256), SkGetPackedA32_SSE2(src));
+    return SkAlphaMulQ_SSE2(dst, isa);
+}
+
+static __m128i srcatop_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
+    __m128i sa = SkGetPackedA32_SSE2(src);
+    __m128i da = SkGetPackedA32_SSE2(dst);
+    __m128i isa = _mm_sub_epi32(_mm_set1_epi32(255), sa);
+
+    __m128i a = da;
+
+    __m128i r1 = SkAlphaMulAlpha_SSE2(da, SkGetPackedR32_SSE2(src));
+    __m128i r2 = SkAlphaMulAlpha_SSE2(isa, SkGetPackedR32_SSE2(dst));
+    __m128i r = _mm_add_epi32(r1, r2);
+
+    __m128i g1 = SkAlphaMulAlpha_SSE2(da, SkGetPackedG32_SSE2(src));
+    __m128i g2 = SkAlphaMulAlpha_SSE2(isa, SkGetPackedG32_SSE2(dst));
+    __m128i g = _mm_add_epi32(g1, g2);
+
+    __m128i b1 = SkAlphaMulAlpha_SSE2(da, SkGetPackedB32_SSE2(src));
+    __m128i b2 = SkAlphaMulAlpha_SSE2(isa, SkGetPackedB32_SSE2(dst));
+    __m128i b = _mm_add_epi32(b1, b2);
+
+    return SkPackARGB32_SSE2(a, r, g, b);
+}
+
+static __m128i dstatop_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
+    __m128i sa = SkGetPackedA32_SSE2(src);
+    __m128i da = SkGetPackedA32_SSE2(dst);
+    __m128i ida = _mm_sub_epi32(_mm_set1_epi32(255), da);
+
+    __m128i a = sa;
+
+    __m128i r1 = SkAlphaMulAlpha_SSE2(ida, SkGetPackedR32_SSE2(src));
+    __m128i r2 = SkAlphaMulAlpha_SSE2(sa, SkGetPackedR32_SSE2(dst));
+    __m128i r = _mm_add_epi32(r1, r2);
+
+    __m128i g1 = SkAlphaMulAlpha_SSE2(ida, SkGetPackedG32_SSE2(src));
+    __m128i g2 = SkAlphaMulAlpha_SSE2(sa, SkGetPackedG32_SSE2(dst));
+    __m128i g = _mm_add_epi32(g1, g2);
+
+    __m128i b1 = SkAlphaMulAlpha_SSE2(ida, SkGetPackedB32_SSE2(src));
+    __m128i b2 = SkAlphaMulAlpha_SSE2(sa, SkGetPackedB32_SSE2(dst));
+    __m128i b = _mm_add_epi32(b1, b2);
+
+    return SkPackARGB32_SSE2(a, r, g, b);
+}
+
+static __m128i xor_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
+    __m128i sa = SkGetPackedA32_SSE2(src);
+    __m128i da = SkGetPackedA32_SSE2(dst);
+    __m128i isa = _mm_sub_epi32(_mm_set1_epi32(255), sa);
+    __m128i ida = _mm_sub_epi32(_mm_set1_epi32(255), da);
+
+    __m128i a1 = _mm_add_epi32(sa, da);
+    __m128i a2 = SkAlphaMulAlpha_SSE2(sa, da);
+    a2 = _mm_slli_epi32(a2, 1);
+    __m128i a = _mm_sub_epi32(a1, a2);
+
+    __m128i r1 = SkAlphaMulAlpha_SSE2(ida, SkGetPackedR32_SSE2(src));
+    __m128i r2 = SkAlphaMulAlpha_SSE2(isa, SkGetPackedR32_SSE2(dst));
+    __m128i r = _mm_add_epi32(r1, r2);
+
+    __m128i g1 = SkAlphaMulAlpha_SSE2(ida, SkGetPackedG32_SSE2(src));
+    __m128i g2 = SkAlphaMulAlpha_SSE2(isa, SkGetPackedG32_SSE2(dst));
+    __m128i g = _mm_add_epi32(g1, g2);
+
+    __m128i b1 = SkAlphaMulAlpha_SSE2(ida, SkGetPackedB32_SSE2(src));
+    __m128i b2 = SkAlphaMulAlpha_SSE2(isa, SkGetPackedB32_SSE2(dst));
+    __m128i b = _mm_add_epi32(b1, b2);
+
+    return SkPackARGB32_SSE2(a, r, g, b);
+}
+
+static __m128i plus_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
+    __m128i b = saturated_add_SSE2(SkGetPackedB32_SSE2(src),
+                                   SkGetPackedB32_SSE2(dst));
+    __m128i g = saturated_add_SSE2(SkGetPackedG32_SSE2(src),
+                                   SkGetPackedG32_SSE2(dst));
+    __m128i r = saturated_add_SSE2(SkGetPackedR32_SSE2(src),
+                                   SkGetPackedR32_SSE2(dst));
+    __m128i a = saturated_add_SSE2(SkGetPackedA32_SSE2(src),
+                                   SkGetPackedA32_SSE2(dst));
+    return SkPackARGB32_SSE2(a, r, g, b);
+}
+
+static __m128i modulate_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
+    __m128i a = SkAlphaMulAlpha_SSE2(SkGetPackedA32_SSE2(src),
+                                     SkGetPackedA32_SSE2(dst));
+    __m128i r = SkAlphaMulAlpha_SSE2(SkGetPackedR32_SSE2(src),
+                                     SkGetPackedR32_SSE2(dst));
+    __m128i g = SkAlphaMulAlpha_SSE2(SkGetPackedG32_SSE2(src),
+                                     SkGetPackedG32_SSE2(dst));
+    __m128i b = SkAlphaMulAlpha_SSE2(SkGetPackedB32_SSE2(src),
+                                     SkGetPackedB32_SSE2(dst));
+    return SkPackARGB32_SSE2(a, r, g, b);
+}
+
 static inline __m128i srcover_byte_SSE2(const __m128i& a, const __m128i& b) {
     // a + b - SkAlphaMulAlpha(a, b);
     return _mm_sub_epi32(_mm_add_epi32(a, b), SkAlphaMulAlpha_SSE2(a, b));
@@ -84,6 +217,18 @@ static __m128i multiply_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
     return SkPackARGB32_SSE2(a, r, g, b);
 }
 
+static __m128i screen_modeproc_SSE2(const __m128i& src, const __m128i& dst) {
+    __m128i a = srcover_byte_SSE2(SkGetPackedA32_SSE2(src),
+                                  SkGetPackedA32_SSE2(dst));
+    __m128i r = srcover_byte_SSE2(SkGetPackedR32_SSE2(src),
+                                  SkGetPackedR32_SSE2(dst));
+    __m128i g = srcover_byte_SSE2(SkGetPackedG32_SSE2(src),
+                                  SkGetPackedG32_SSE2(dst));
+    __m128i b = srcover_byte_SSE2(SkGetPackedB32_SSE2(src),
+                                  SkGetPackedB32_SSE2(dst));
+    return SkPackARGB32_SSE2(a, r, g, b);
+}
+
 ////////////////////////////////////////////////////////////////////////////////
 
 typedef __m128i (*SkXfermodeProcSIMD)(const __m128i& src, const __m128i& dst);
@@ -226,18 +371,18 @@ SkXfermodeProcSIMD gSSE2XfermodeProcs[] = {
     NULL, // kClear_Mode
     NULL, // kSrc_Mode
     NULL, // kDst_Mode
-    NULL, // kSrcOver_Mode
-    NULL, // kDstOver_Mode
-    NULL, // kSrcIn_Mode
-    NULL, // kDstIn_Mode
-    NULL, // kSrcOut_Mode
-    NULL, // kDstOut_Mode
-    NULL, // kSrcATop_Mode
-    NULL, // kDstATop_Mode
-    NULL, // kXor_Mode
-    NULL, // kPlus_Mode
-    NULL, // kModulate_Mode
-    NULL, // kScreen_Mode
+    srcover_modeproc_SSE2,
+    dstover_modeproc_SSE2,
+    srcin_modeproc_SSE2,
+    dstin_modeproc_SSE2,
+    srcout_modeproc_SSE2,
+    dstout_modeproc_SSE2,
+    srcatop_modeproc_SSE2,
+    dstatop_modeproc_SSE2,
+    xor_modeproc_SSE2,
+    plus_modeproc_SSE2,
+    modulate_modeproc_SSE2,
+    screen_modeproc_SSE2,
 
     NULL, // kOverlay_Mode
     NULL, // kDarken_Mode