Revert of 565 support for SIMD xfermodes

src/core/Sk4px.h

 /*
  * Copyright 2015 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #ifndef Sk4px_DEFINED
 #define Sk4px_DEFINED
 
 #include "SkNx.h"
 #include "SkColor.h"
-#include "SkColorPriv.h"
 
 // This file may be included multiple times by .cpp files with different flags, leading
 // to different definitions.  Usually that doesn't matter because it's all inlined, but
 // in Debug modes the compilers may not inline everything.  So wrap everything in an
 // anonymous namespace to give each includer their own silo of this code (or the linker
 // will probably pick one randomly for us, which is rarely correct).
 namespace {
 
 // 1, 2 or 4 SkPMColors, generally vectorized.
 class Sk4px : public Sk16b {
@@ skipping 17 matching lines @@
     static Sk4px Load1(const SkPMColor[1]);  // PMColor[1] -> ARGB ???? ???? ????
 
     // Ditto for Alphas... Load2Alphas fills the low two lanes of Sk4px.
     static Sk4px Load4Alphas(const SkAlpha[4]);  // AaXx -> AAAA aaaa XXXX xxxx
     static Sk4px Load2Alphas(const SkAlpha[2]);  // Aa   -> AAAA aaaa ???? ????
 
     void store4(SkPMColor[4]) const;
     void store2(SkPMColor[2]) const;
     void store1(SkPMColor[1]) const;
 
-    // Same as above for 565.
-    static Sk4px Load4(const SkPMColor16 src[4]);
-    static Sk4px Load2(const SkPMColor16 src[2]);
-    static Sk4px Load1(const SkPMColor16 src[1]);
-    void store4(SkPMColor16 dst[4]) const;
-    void store2(SkPMColor16 dst[2]) const;
-    void store1(SkPMColor16 dst[1]) const;
-
     // 1, 2, or 4 SkPMColors with 16-bit components.
     // This is most useful as the result of a multiply, e.g. from mulWiden().
     class Wide : public Sk16h {
     public:
         Wide(const Sk16h& v) : Sk16h(v) {}
 
         // Pack the top byte of each component back down into 4 SkPMColors.
         Sk4px addNarrowHi(const Sk16h&) const;
 
         // Rounds, i.e. (x+127) / 255.
@@ skipping 32 matching lines @@
 
     // Generally faster than (*this * o).div255().
     // May be incorrect by +-1, but is always exactly correct when *this or o is 0 or 255.
     Sk4px approxMulDiv255(const Sk16b& o) const {
         // (x*y + x) / 256 meets these criteria.  (As of course does (x*y + y) / 256 by symmetry.)
         return this->widenLo().addNarrowHi(*this * o);
     }
 
     // A generic driver that maps fn over a src array into a dst array.
     // fn should take an Sk4px (4 src pixels) and return an Sk4px (4 dst pixels).
-    template <typename Fn, typename Dst>
-    static void MapSrc(int n, Dst* dst, const SkPMColor* src, const Fn& fn) {
+    template <typename Fn>
+    static void MapSrc(int n, SkPMColor* dst, const SkPMColor* src, const Fn& fn) {
         // This looks a bit odd, but it helps loop-invariant hoisting across different calls to fn.
         // Basically, we need to make sure we keep things inside a single loop.
         while (n > 0) {
             if (n >= 8) {
                 Sk4px dst0 = fn(Load4(src+0)),
                       dst4 = fn(Load4(src+4));
                 dst0.store4(dst+0);
                 dst4.store4(dst+4);
                 dst += 8; src += 8; n -= 8;
                 continue;  // Keep our stride at 8 pixels as long as possible.
             }
             SkASSERT(n <= 7);
             if (n >= 4) {
                 fn(Load4(src)).store4(dst);
                 dst += 4; src += 4; n -= 4;
             }
             if (n >= 2) {
                 fn(Load2(src)).store2(dst);
                 dst += 2; src += 2; n -= 2;
             }
             if (n >= 1) {
                 fn(Load1(src)).store1(dst);
             }
             break;
         }
     }
 
     // As above, but with dst4' = fn(dst4, src4).
-    template <typename Fn, typename Dst>
-    static void MapDstSrc(int n, Dst* dst, const SkPMColor* src, const Fn& fn) {
+    template <typename Fn>
+    static void MapDstSrc(int n, SkPMColor* dst, const SkPMColor* src, const Fn& fn) {
         while (n > 0) {
             if (n >= 8) {
                 Sk4px dst0 = fn(Load4(dst+0), Load4(src+0)),
                       dst4 = fn(Load4(dst+4), Load4(src+4));
                 dst0.store4(dst+0);
                 dst4.store4(dst+4);
                 dst += 8; src += 8; n -= 8;
                 continue;  // Keep our stride at 8 pixels as long as possible.
             }
             SkASSERT(n <= 7);
             if (n >= 4) {
                 fn(Load4(dst), Load4(src)).store4(dst);
                 dst += 4; src += 4; n -= 4;
             }
             if (n >= 2) {
                 fn(Load2(dst), Load2(src)).store2(dst);
                 dst += 2; src += 2; n -= 2;
             }
             if (n >= 1) {
                 fn(Load1(dst), Load1(src)).store1(dst);
             }
             break;
         }
     }
 
     // As above, but with dst4' = fn(dst4, src4, alpha4).
-    template <typename Fn, typename Dst>
-    static void MapDstSrcAlpha(int n, Dst* dst, const SkPMColor* src, const SkAlpha* a,
+    template <typename Fn>
+    static void MapDstSrcAlpha(int n, SkPMColor* dst, const SkPMColor* src, const SkAlpha* a,
                                const Fn& fn) {
         while (n > 0) {
             if (n >= 8) {
                 Sk4px dst0 = fn(Load4(dst+0), Load4(src+0), Load4Alphas(a+0)),
                       dst4 = fn(Load4(dst+4), Load4(src+4), Load4Alphas(a+4));
                 dst0.store4(dst+0);
                 dst4.store4(dst+4);
                 dst += 8; src += 8; a += 8; n -= 8;
                 continue;  // Keep our stride at 8 pixels as long as possible.
             }
@@ skipping 10 matching lines @@
                 fn(Load1(dst), Load1(src), DupAlpha(*a)).store1(dst);
             }
             break;
         }
     }
 
 private:
     typedef Sk16b INHERITED;
 };
 
-// TODO: specialize these per-backend
-
-inline Sk4px Sk4px::Load4(const SkPMColor16 src[4]) {
-    SkPMColor src32[4];
-    for (int i = 0; i < 4; i++) { src32[i] = SkPixel16ToPixel32(src[i]); }
-    return Load4(src32);
-}
-inline Sk4px Sk4px::Load2(const SkPMColor16 src[2]) {
-    SkPMColor src32[2];
-    for (int i = 0; i < 2; i++) { src32[i] = SkPixel16ToPixel32(src[i]); }
-    return Load2(src32);
-}
-inline Sk4px Sk4px::Load1(const SkPMColor16 src[1]) {
-    SkPMColor src32 = SkPixel16ToPixel32(src[0]);
-    return Load1(&src32);
-}
-
-inline void Sk4px::store4(SkPMColor16 dst[4]) const {
-    SkPMColor dst32[4];
-    this->store4(dst32);
-    for (int i = 0; i < 4; i++) { dst[i] = SkPixel32ToPixel16(dst32[i]); }
-}
-inline void Sk4px::store2(SkPMColor16 dst[2]) const {
-    SkPMColor dst32[2];
-    this->store2(dst32);
-    for (int i = 0; i < 2; i++) { dst[i] = SkPixel32ToPixel16(dst32[i]); }
-}
-inline void Sk4px::store1(SkPMColor16 dst[1]) const {
-    SkPMColor dst32;
-    this->store1(&dst32);
-    dst[0] = SkPixel32ToPixel16(dst32);
-}
-
 }  // namespace
 
 #ifdef SKNX_NO_SIMD
     #include "../opts/Sk4px_none.h"
 #else
     #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSE2
         #include "../opts/Sk4px_SSE2.h"
     #elif defined(SK_ARM_HAS_NEON)
         #include "../opts/Sk4px_NEON.h"
     #else
         #include "../opts/Sk4px_none.h"
     #endif
 #endif
 
 #endif//Sk4px_DEFINED