Revert of De-templatize Sk4pxXfermode code a bit. (patchset #2 id:20001 of https://codereview.chromium.org/1242743004/)

Reason for revert:
http://build.chromium.org/p/client.skia.compile/builders/Build-Ubuntu-GCC-Arm7-Debug-Android_NoNeon/builds/1168/steps/build%20most/logs/stdio

Original issue's description:
> De-templatize Sk4pxXfermode code a bit.
>
> This deduplicates a few pieces of code:
>   - we end up with one copy of each xfer32() driver loop instead of one per xfermode;
>   - we end up with two* copies of each xfermode implementation instead of ten**.
>
> * For a given Mode: Mode() itself and xfer_aa<Mode>().
> ** From unrolling: twice at a stride of 8, once at 4, once at 2, and once at 1, then all again for when we have AA.
>
> This decreases the size of SkXfermode.o from 1.5M to 620K on x86-64 and from 1.3M to 680K on ARMv7+NEON.
>
> If we wanted to, we could eliminate the xfer_aa<Mode>() copy by tagging each Mode() function as __attribute__((noinline)) or its equivalent.  This would result in another ~100K space savings.
>
> Performance is affected in proportion to the original xfermode speed:
> fast modes like Plus take the largest proportional hit, and slow modes
> like HardLight or SoftLight see essentially no hit at all.
>
> This adds SK_VECTORCALL to help keep this code fast on ARMv7 and Windows.  I've looked at the ARMv7 generated code... it looks good, even pretty.
>
> For compatibility with SK_VECTORCALL, we now pass the vector-sized arguments by value instead of by reference.  Some refactoring now allows us to declare each mode as just a static function instead of a struct, which simplifies things.
>
> TBR=reed@google.com
> No public API changes.
>
> BUG=skia:
>
> Committed: https://skia.googlesource.com/skia/+/e617e1525916d7ee684142728c0905828caf49da

TBR=msarett@google.com,mtklein@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=skia:

Review URL: https://codereview.chromium.org/1245273005

1 files changed, 64 insertions, 63 deletions

diff --git a/src/core/Sk4pxXfermode.h b/src/core/Sk4pxXfermode.h
index d89dbf8e2d..0c8dcb5302 100644
--- a/src/core/Sk4pxXfermode.h
+++ b/src/core/Sk4pxXfermode.h
@@ -17,7 +17,12 @@
 namespace {
 
 // Most xfermodes can be done most efficiently 4 pixels at a time in 8 or 16-bit fixed point.
-#define XFERMODE(Name) static Sk4px SK_VECTORCALL Name(Sk4px s, Sk4px d)
+#define XFERMODE(Name)                                                    \
+    struct Name {                                                         \
+        static Sk4px Xfer(const Sk4px&, const Sk4px&);                    \
+        static const SkXfermode::Mode kMode = SkXfermode::k##Name##_Mode; \
+    };                                                                    \
+    inline Sk4px Name::Xfer(const Sk4px& s, const Sk4px& d)
 
 XFERMODE(Clear) { return Sk4px::DupPMColor(0); }
 XFERMODE(Src)   { return s; }
@@ -25,13 +30,13 @@ XFERMODE(Dst)   { return d; }
 XFERMODE(SrcIn)   { return     s.approxMulDiv255(d.alphas()      ); }
 XFERMODE(SrcOut)  { return     s.approxMulDiv255(d.alphas().inv()); }
 XFERMODE(SrcOver) { return s + d.approxMulDiv255(s.alphas().inv()); }
-XFERMODE(DstIn)   { return SrcIn  (d,s); }
-XFERMODE(DstOut)  { return SrcOut (d,s); }
-XFERMODE(DstOver) { return SrcOver(d,s); }
+XFERMODE(DstIn)   { return SrcIn  ::Xfer(d,s); }
+XFERMODE(DstOut)  { return SrcOut ::Xfer(d,s); }
+XFERMODE(DstOver) { return SrcOver::Xfer(d,s); }
 
 // [ S * Da + (1 - Sa) * D]
 XFERMODE(SrcATop) { return (s * d.alphas() + d * s.alphas().inv()).div255(); }
-XFERMODE(DstATop) { return SrcATop(d,s); }
+XFERMODE(DstATop) { return SrcATop::Xfer(d,s); }
 //[ S * (1 - Da) + (1 - Sa) * D ]
 XFERMODE(Xor) { return (s * d.alphas().inv() + d * s.alphas().inv()).div255(); }
 // [S + D ]
@@ -81,7 +86,7 @@ XFERMODE(HardLight) {
     auto colors = (both + isLite.thenElse(lite, dark)).div255();
     return alphas.zeroColors() + colors.zeroAlphas();
 }
-XFERMODE(Overlay) { return HardLight(d,s); }
+XFERMODE(Overlay) { return HardLight::Xfer(d,s); }
 
 XFERMODE(Darken) {
     auto sa = s.alphas(),
@@ -112,7 +117,12 @@ XFERMODE(Lighten) {
 #undef XFERMODE
 
 // Some xfermodes use math like divide or sqrt that's best done in floats 1 pixel at a time.
-#define XFERMODE(Name) static SkPMFloat SK_VECTORCALL Name(SkPMFloat s, SkPMFloat d)
+#define XFERMODE(Name) \
+    struct Name {                                                         \
+        static SkPMFloat Xfer(const SkPMFloat&, const SkPMFloat&);        \
+        static const SkXfermode::Mode kMode = SkXfermode::k##Name##_Mode; \
+    };                                                                    \
+    inline SkPMFloat Name::Xfer(const SkPMFloat& s, const SkPMFloat& d)
 
 XFERMODE(ColorDodge) {
     auto sa = s.alphas(),
@@ -175,15 +185,15 @@ XFERMODE(SoftLight) {
 
 // A reasonable fallback mode for doing AA is to simply apply the transfermode first,
 // then linearly interpolate the AA.
-template <Sk4px (SK_VECTORCALL *Mode)(Sk4px, Sk4px)>
-static Sk4px SK_VECTORCALL xfer_aa(Sk4px s, Sk4px d, Sk4px aa) {
-    Sk4px bw = Mode(s, d);
+template <typename Mode>
+static Sk4px xfer_aa(const Sk4px& s, const Sk4px& d, const Sk4px& aa) {
+    Sk4px bw = Mode::Xfer(s, d);
     return (bw * aa + d * aa.inv()).div255();
 }
 
 // For some transfermodes we specialize AA, either for correctness or performance.
 #define XFERMODE_AA(Name) \
-    template <> Sk4px SK_VECTORCALL xfer_aa<Name>(Sk4px s, Sk4px d, Sk4px aa)
+    template <> Sk4px xfer_aa<Name>(const Sk4px& s, const Sk4px& d, const Sk4px& aa)
 
 // Plus' clamp needs to happen after AA.  skia:3852
 XFERMODE_AA(Plus) {  // [ clamp( (1-AA)D + (AA)(S+D) ) == clamp(D + AA*S) ]
@@ -192,47 +202,44 @@ XFERMODE_AA(Plus) {  // [ clamp( (1-AA)D + (AA)(S+D) ) == clamp(D + AA*S) ]
 
 #undef XFERMODE_AA
 
-class Sk4pxXfermode : public SkProcCoeffXfermode {
+template <typename ProcType>
+class SkT4pxXfermode : public SkProcCoeffXfermode {
 public:
-    typedef Sk4px (SK_VECTORCALL *Proc4)(Sk4px, Sk4px);
-    typedef Sk4px (SK_VECTORCALL *AAProc4)(Sk4px, Sk4px, Sk4px);
-
-    Sk4pxXfermode(const ProcCoeff& rec, SkXfermode::Mode mode, Proc4 proc4, AAProc4 aaproc4)
-        : INHERITED(rec, mode)
-        , fProc4(proc4)
-        , fAAProc4(aaproc4) {}
+    static SkProcCoeffXfermode* Create(const ProcCoeff& rec) {
+        return SkNEW_ARGS(SkT4pxXfermode, (rec));
+    }
 
     void xfer32(SkPMColor dst[], const SkPMColor src[], int n, const SkAlpha aa[]) const override {
         if (NULL == aa) {
             Sk4px::MapDstSrc(n, dst, src, [&](const Sk4px& dst4, const Sk4px& src4) {
-                return fProc4(src4, dst4);
+                return ProcType::Xfer(src4, dst4);
             });
         } else {
             Sk4px::MapDstSrcAlpha(n, dst, src, aa,
                     [&](const Sk4px& dst4, const Sk4px& src4, const Sk4px& alpha) {
-                return fAAProc4(src4, dst4, alpha);
+                return xfer_aa<ProcType>(src4, dst4, alpha);
             });
         }
     }
 
 private:
-    Proc4 fProc4;
-    AAProc4 fAAProc4;
+    SkT4pxXfermode(const ProcCoeff& rec) : INHERITED(rec, ProcType::kMode) {}
+
     typedef SkProcCoeffXfermode INHERITED;
 };
 
-class SkPMFloatXfermode : public SkProcCoeffXfermode {
+template <typename ProcType>
+class SkTPMFloatXfermode : public SkProcCoeffXfermode {
 public:
-    typedef SkPMFloat (SK_VECTORCALL *ProcF)(SkPMFloat, SkPMFloat);
-    SkPMFloatXfermode(const ProcCoeff& rec, SkXfermode::Mode mode, ProcF procf)
-        : INHERITED(rec, mode)
-        , fProcF(procf) {}
+    static SkProcCoeffXfermode* Create(const ProcCoeff& rec) {
+        return SkNEW_ARGS(SkTPMFloatXfermode, (rec));
+    }
 
     void xfer32(SkPMColor dst[], const SkPMColor src[], int n, const SkAlpha aa[]) const override {
         for (int i = 0; i < n; i++) {
             SkPMFloat s(src[i]),
                       d(dst[i]),
-                      b(fProcF(s,d));
+                      b(ProcType::Xfer(s,d));
             if (aa) {
                 // We do aa in full float precision before going back down to bytes, because we can!
                 SkPMFloat a = Sk4f(aa[i]) * Sk4f(1.0f/255);
@@ -243,46 +250,40 @@ public:
     }
 
 private:
-    ProcF fProcF;
+    SkTPMFloatXfermode(const ProcCoeff& rec) : INHERITED(rec, ProcType::kMode) {}
+
     typedef SkProcCoeffXfermode INHERITED;
 };
 
 static SkProcCoeffXfermode* SkCreate4pxXfermode(const ProcCoeff& rec, SkXfermode::Mode mode) {
 #if !defined(SK_CPU_ARM32) || defined(SK_ARM_HAS_NEON)
     switch (mode) {
-    #define CASE(Mode) case SkXfermode::k##Mode##_Mode: \
-        return SkNEW_ARGS(Sk4pxXfermode, (rec, mode, &Mode, &xfer_aa<Mode>))
-        CASE(Clear);
-        CASE(Src);
-        CASE(Dst);
-        CASE(SrcOver);
-        CASE(DstOver);
-        CASE(SrcIn);
-        CASE(DstIn);
-        CASE(SrcOut);
-        CASE(DstOut);
-        CASE(SrcATop);
-        CASE(DstATop);
-        CASE(Xor);
-        CASE(Plus);
-        CASE(Modulate);
-        CASE(Screen);
-        CASE(Multiply);
-        CASE(Difference);
-        CASE(Exclusion);
-        CASE(HardLight);
-        CASE(Overlay);
-        CASE(Darken);
-        CASE(Lighten);
-    #undef CASE
-
-    #define CASE(Mode) case SkXfermode::k##Mode##_Mode: \
-        return SkNEW_ARGS(SkPMFloatXfermode, (rec, mode, &Mode))
-        CASE(ColorDodge);
-        CASE(ColorBurn);
-        CASE(SoftLight);
-    #undef CASE
-
+        case SkXfermode::kClear_Mode:      return SkT4pxXfermode<Clear>::Create(rec);
+        case SkXfermode::kSrc_Mode:        return SkT4pxXfermode<Src>::Create(rec);
+        case SkXfermode::kDst_Mode:        return SkT4pxXfermode<Dst>::Create(rec);
+        case SkXfermode::kSrcOver_Mode:    return SkT4pxXfermode<SrcOver>::Create(rec);
+        case SkXfermode::kDstOver_Mode:    return SkT4pxXfermode<DstOver>::Create(rec);
+        case SkXfermode::kSrcIn_Mode:      return SkT4pxXfermode<SrcIn>::Create(rec);
+        case SkXfermode::kDstIn_Mode:      return SkT4pxXfermode<DstIn>::Create(rec);
+        case SkXfermode::kSrcOut_Mode:     return SkT4pxXfermode<SrcOut>::Create(rec);
+        case SkXfermode::kDstOut_Mode:     return SkT4pxXfermode<DstOut>::Create(rec);
+        case SkXfermode::kSrcATop_Mode:    return SkT4pxXfermode<SrcATop>::Create(rec);
+        case SkXfermode::kDstATop_Mode:    return SkT4pxXfermode<DstATop>::Create(rec);
+        case SkXfermode::kXor_Mode:        return SkT4pxXfermode<Xor>::Create(rec);
+        case SkXfermode::kPlus_Mode:       return SkT4pxXfermode<Plus>::Create(rec);
+        case SkXfermode::kModulate_Mode:   return SkT4pxXfermode<Modulate>::Create(rec);
+        case SkXfermode::kScreen_Mode:     return SkT4pxXfermode<Screen>::Create(rec);
+        case SkXfermode::kMultiply_Mode:   return SkT4pxXfermode<Multiply>::Create(rec);
+        case SkXfermode::kDifference_Mode: return SkT4pxXfermode<Difference>::Create(rec);
+        case SkXfermode::kExclusion_Mode:  return SkT4pxXfermode<Exclusion>::Create(rec);
+        case SkXfermode::kHardLight_Mode:  return SkT4pxXfermode<HardLight>::Create(rec);
+        case SkXfermode::kOverlay_Mode:    return SkT4pxXfermode<Overlay>::Create(rec);
+        case SkXfermode::kDarken_Mode:     return SkT4pxXfermode<Darken>::Create(rec);
+        case SkXfermode::kLighten_Mode:    return SkT4pxXfermode<Lighten>::Create(rec);
+
+        case SkXfermode::kColorDodge_Mode: return SkTPMFloatXfermode<ColorDodge>::Create(rec);
+        case SkXfermode::kColorBurn_Mode:  return SkTPMFloatXfermode<ColorBurn>::Create(rec);
+        case SkXfermode::kSoftLight_Mode:  return SkTPMFloatXfermode<SoftLight>::Create(rec);
         default: break;
     }
 #endif