stages for most xfermodes

also assumes the clamp from https://codereview.chromium.org/2178793002/ is present

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2177103002

Review-Url: https://codereview.chromium.org/2177103002

2 files changed, 124 insertions, 0 deletions

diff --git a/src/core/SkXfermode.cpp b/src/core/SkXfermode.cpp
index 9246568699..64db9dfb56 100644
--- a/src/core/SkXfermode.cpp
+++ b/src/core/SkXfermode.cpp
@@ -11,6 +11,7 @@
 #include "SkMathPriv.h"
 #include "SkOnce.h"
 #include "SkOpts.h"
+#include "SkRasterPipeline.h"
 #include "SkReadBuffer.h"
 #include "SkString.h"
 #include "SkWriteBuffer.h"
@@ -1431,3 +1432,124 @@ bool SkXfermode::onAppendStages(SkRasterPipeline*) const {
 SK_DEFINE_FLATTENABLE_REGISTRAR_GROUP_START(SkXfermode)
     SK_DEFINE_FLATTENABLE_REGISTRAR_ENTRY(SkProcCoeffXfermode)
 SK_DEFINE_FLATTENABLE_REGISTRAR_GROUP_END
+
+static Sk4f inv(const Sk4f& x) { return 1.0f - x; }
+
+// Most of these modes apply the same logic kernel to each channel.
+template <Sk4f kernel(const Sk4f& s, const Sk4f& sa, const Sk4f& d, const Sk4f& da)>
+static void SK_VECTORCALL rgba(SkRasterPipeline::Stage* st, size_t x,
+                               Sk4f  r, Sk4f  g, Sk4f  b, Sk4f  a,
+                               Sk4f dr, Sk4f dg, Sk4f db, Sk4f da) {
+    r = kernel(r,a,dr,da);
+    g = kernel(g,a,dg,da);
+    b = kernel(b,a,db,da);
+    a = kernel(a,a,da,da);
+    st->next(x, r,g,b,a, dr,dg,db,da);
+}
+
+#define KERNEL(name) static Sk4f name(const Sk4f& s, const Sk4f& sa, const Sk4f& d, const Sk4f& da)
+KERNEL(clear)    { return 0.0f; }
+KERNEL(dst)      { return d; }
+KERNEL(dstover)  { return d + inv(da)*s; }
+
+KERNEL(srcin)    { return s * da; }
+KERNEL(srcout)   { return s * inv(da); }
+KERNEL(srcatop)  { return s*da + d*inv(sa); }
+KERNEL(dstin)    { return srcin  (d,da,s,sa); }
+KERNEL(dstout)   { return srcout (d,da,s,sa); }
+KERNEL(dstatop)  { return srcatop(d,da,s,sa); }
+
+KERNEL(modulate) { return s*d; }
+KERNEL(multiply) { return s*inv(da) + d*inv(sa) + s*d; }
+KERNEL(plus)     { return s + d; }
+KERNEL(screen)   { return s + d - s*d; }
+KERNEL(xor_)     { return s*inv(da) + d*inv(sa); }
+
+// Most of the rest apply the same logic to each color channel, and srcover's logic to alpha.
+// (darken and lighten can actually go either way, but they're a little faster this way.)
+template <Sk4f kernel(const Sk4f& s, const Sk4f& sa, const Sk4f& d, const Sk4f& da)>
+static void SK_VECTORCALL rgb_srcover(SkRasterPipeline::Stage* st, size_t x,
+                                      Sk4f  r, Sk4f  g, Sk4f  b, Sk4f  a,
+                                      Sk4f dr, Sk4f dg, Sk4f db, Sk4f da) {
+    r = kernel(r,a,dr,da);
+    g = kernel(g,a,dg,da);
+    b = kernel(b,a,db,da);
+    a = a + da*inv(a);
+    st->next(x, r,g,b,a, dr,dg,db,da);
+}
+
+KERNEL(colorburn) {
+    return (d == da  ).thenElse(d + s*inv(da),
+           (s == 0.0f).thenElse(s + d*inv(sa),
+                                sa*(da - Sk4f::Min(da, (da-d)*sa/s)) + s*inv(da) + d*inv(sa)));
+}
+KERNEL(colordodge) {
+    return (d == 0.0f).thenElse(d + s*inv(da),
+           (s == sa  ).thenElse(s + d*inv(sa),
+                                sa*Sk4f::Min(da, (d*sa)/(sa - s)) + s*inv(da) + d*inv(sa)));
+}
+KERNEL(darken)     { return s + d - Sk4f::Max(s*da, d*sa); }
+KERNEL(difference) { return s + d - 2.0f*Sk4f::Min(s*da,d*sa); }
+KERNEL(exclusion)  { return s + d - 2.0f*s*d; }
+KERNEL(hardlight) {
+    return s*inv(da) + d*inv(sa)
+         + (2.0f*s <= sa).thenElse(2.0f*s*d, sa*da - 2.0f*(da-d)*(sa-s));
+}
+KERNEL(lighten) { return s + d - Sk4f::Min(s*da, d*sa); }
+KERNEL(overlay) { return hardlight(d,da,s,sa); }
+KERNEL(softlight) {
+    Sk4f m  = (da > 0.0f).thenElse(d / da, 0.0f),
+         s2 = 2.0f*s,
+         m4 = 4.0f*m;
+
+    // The logic forks three ways:
+    //    1. dark src?
+    //    2. light src, dark dst?
+    //    3. light src, light dst?
+    Sk4f darkSrc = d*(sa + (s2 - sa)*(1.0f - m)),     // Used in case 1.
+         darkDst = (m4*m4 + m4)*(m - 1.0f) + 7.0f*m,  // Used in case 2.
+         liteDst = m.rsqrt().invert() - m,            // Used in case 3.
+         liteSrc = d*sa + da*(s2 - sa) * (4.0f*d <= da).thenElse(darkDst, liteDst);  // 2 or 3?
+    return s*inv(da) + d*inv(sa) + (s2 <= sa).thenElse(darkSrc, liteSrc);  // 1 or (2 or 3)?
+}
+#undef KERNEL
+
+bool SkProcCoeffXfermode::onAppendStages(SkRasterPipeline* p) const {
+    switch (fMode) {
+        case kSrcOver_Mode: SkASSERT(false); return false;  // Well how did we get here?
+
+        case kSrc_Mode:    /*This stage is a no-op.*/ return true;
+        case kDst_Mode:     p->append(rgba<dst>);     return true;
+        case kSrcATop_Mode: p->append(rgba<srcatop>); return true;
+        case kDstATop_Mode: p->append(rgba<dstatop>); return true;
+        case kSrcIn_Mode:   p->append(rgba<srcin>);   return true;
+        case kDstIn_Mode:   p->append(rgba<dstin>);   return true;
+        case kSrcOut_Mode:  p->append(rgba<srcout>);  return true;
+        case kDstOut_Mode:  p->append(rgba<dstout>);  return true;
+        case kDstOver_Mode: p->append(rgba<dstover>); return true;
+
+        case kClear_Mode:    p->append(rgba<clear>);    return true;
+        case kModulate_Mode: p->append(rgba<modulate>); return true;
+        case kMultiply_Mode: p->append(rgba<multiply>); return true;
+        case kPlus_Mode:     p->append(rgba<plus>);     return true;
+        case kScreen_Mode:   p->append(rgba<screen>);   return true;
+        case kXor_Mode:      p->append(rgba<xor_>);     return true;
+
+        case kColorBurn_Mode:  p->append(rgb_srcover<colorburn>);  return true;
+        case kColorDodge_Mode: p->append(rgb_srcover<colordodge>); return true;
+        case kDarken_Mode:     p->append(rgb_srcover<darken>);     return true;
+        case kDifference_Mode: p->append(rgb_srcover<difference>); return true;
+        case kExclusion_Mode:  p->append(rgb_srcover<exclusion>);  return true;
+        case kHardLight_Mode:  p->append(rgb_srcover<hardlight>);  return true;
+        case kLighten_Mode:    p->append(rgb_srcover<lighten>);    return true;
+        case kOverlay_Mode:    p->append(rgb_srcover<overlay>);    return true;
+        case kSoftLight_Mode:  p->append(rgb_srcover<softlight>);  return true;
+
+        // TODO
+        case kColor_Mode:       return false;
+        case kHue_Mode:         return false;
+        case kLuminosity_Mode:  return false;
+        case kSaturation_Mode:  return false;
+    }
+    return false;
+}
diff --git a/src/core/SkXfermode_proccoeff.h b/src/core/SkXfermode_proccoeff.h
index 357de4324c..1e6cc482a6 100644
--- a/src/core/SkXfermode_proccoeff.h
+++ b/src/core/SkXfermode_proccoeff.h
@@ -53,6 +53,8 @@ public:
     SK_TO_STRING_OVERRIDE()
     SK_DECLARE_PUBLIC_FLATTENABLE_DESERIALIZATION_PROCS(SkProcCoeffXfermode)
 
+    bool onAppendStages(SkRasterPipeline*) const override;
+
 protected:
     void flatten(SkWriteBuffer& buffer) const override;