Revert of Code's more readable when SkPMFloat is an Sk4f. (patchset #3 id:40001 of https://codereview.chromium.org/1061603002/)

Reason for revert:
missed some neon code

Original issue's description:
> Code's more readable when SkPMFloat is an Sk4f.
>  #floats
>
> BUG=skia:
> BUG=skia:3592
>
> Committed: https://skia.googlesource.com/skia/+/6b5dab889579f1cc9e1b5278f4ecdc4c63fe78c9

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

Review URL: https://codereview.chromium.org/1056143004
diff --git a/src/core/SkPMFloat.h b/src/core/SkPMFloat.h
index eb575f2..a327025 100644
--- a/src/core/SkPMFloat.h
+++ b/src/core/SkPMFloat.h
@@ -15,7 +15,7 @@
 
 // A pre-multiplied color storing each component in the same order as SkPMColor,
 // but as a float in the range [0, 255].
-class SkPMFloat : public Sk4f {
+class SK_STRUCT_ALIGN(16) SkPMFloat {
 public:
     static SkPMFloat FromPMColor(SkPMColor c) { return SkPMFloat(c); }
     static SkPMFloat FromARGB(float a, float r, float g, float b) { return SkPMFloat(a,r,g,b); }
@@ -28,17 +28,20 @@
     explicit SkPMFloat(SkPMColor);
     SkPMFloat(float a, float r, float g, float b)
     #ifdef SK_PMCOLOR_IS_RGBA
-        : INHERITED(r,g,b,a) {}
+        : fColors(r,g,b,a) {}
     #else
-        : INHERITED(b,g,r,a) {}
+        : fColors(b,g,r,a) {}
     #endif
 
-    SkPMFloat(const Sk4f& fs) : INHERITED(fs) {}
 
-    float a() const { return this->kth<SK_A32_SHIFT / 8>(); }
-    float r() const { return this->kth<SK_R32_SHIFT / 8>(); }
-    float g() const { return this->kth<SK_G32_SHIFT / 8>(); }
-    float b() const { return this->kth<SK_B32_SHIFT / 8>(); }
+    // Freely autoconvert between SkPMFloat and Sk4f.
+    /*implicit*/ SkPMFloat(const Sk4f& fs) { fColors = fs; }
+    /*implicit*/ operator Sk4f() const { return fColors; }
+
+    float a() const { return fColors.kth<SK_A32_SHIFT / 8>(); }
+    float r() const { return fColors.kth<SK_R32_SHIFT / 8>(); }
+    float g() const { return fColors.kth<SK_G32_SHIFT / 8>(); }
+    float b() const { return fColors.kth<SK_B32_SHIFT / 8>(); }
 
     // N.B. All methods returning an SkPMColor call SkPMColorAssert on that result before returning.
 
@@ -64,7 +67,7 @@
     }
 
 private:
-    typedef Sk4f INHERITED;
+    Sk4f fColors;
 };
 
 #ifdef SKNX_NO_SIMD
diff --git a/src/core/SkXfermode.cpp b/src/core/SkXfermode.cpp
index 1d502d9..5e9055c 100644
--- a/src/core/SkXfermode.cpp
+++ b/src/core/SkXfermode.cpp
@@ -1236,7 +1236,9 @@
 struct SrcATop4f {
     static SkPMFloat Xfer(const SkPMFloat& src, const SkPMFloat& dst) {
         const Sk4f inv255(gInv255);
-        return check_as_pmfloat(dst + (src * Sk4f(dst.a()) - dst * Sk4f(src.a())) * inv255);
+        Sk4f s4 = src;
+        Sk4f d4 = dst;
+        return check_as_pmfloat(d4 + (s4 * Sk4f(dst.a()) - d4 * Sk4f(src.a())) * inv255);
     }
     static const bool kFoldCoverageIntoSrcAlpha = true;
     static const SkXfermode::Mode kMode = SkXfermode::kSrcATop_Mode;
@@ -1246,7 +1248,9 @@
 struct DstATop4f {
     static SkPMFloat Xfer(const SkPMFloat& src, const SkPMFloat& dst) {
         const Sk4f inv255(gInv255);
-        return check_as_pmfloat(src + (dst * Sk4f(src.a()) - src * Sk4f(dst.a())) * inv255);
+        Sk4f s4 = src;
+        Sk4f d4 = dst;
+        return check_as_pmfloat(s4 + (d4 * Sk4f(src.a()) - s4 * Sk4f(dst.a())) * inv255);
     }
     static const bool kFoldCoverageIntoSrcAlpha = false;
     static const SkXfermode::Mode kMode = SkXfermode::kDstATop_Mode;
@@ -1256,7 +1260,9 @@
 struct Xor4f {
     static SkPMFloat Xfer(const SkPMFloat& src, const SkPMFloat& dst) {
         const Sk4f inv255(gInv255);
-        return check_as_pmfloat(src + dst - (src * Sk4f(dst.a()) + dst * Sk4f(src.a())) * inv255);
+        Sk4f s4 = src;
+        Sk4f d4 = dst;
+        return check_as_pmfloat(s4 + d4 - (s4 * Sk4f(dst.a()) + d4 * Sk4f(src.a())) * inv255);
     }
     static const bool kFoldCoverageIntoSrcAlpha = true;
     static const SkXfermode::Mode kMode = SkXfermode::kXor_Mode;
@@ -1265,7 +1271,9 @@
 //  kPlus_Mode   [Sa + Da, Sc + Dc]
 struct Plus4f {
     static SkPMFloat Xfer(const SkPMFloat& src, const SkPMFloat& dst) {
-        return check_as_pmfloat(clamp_255(src + dst));
+        Sk4f s4 = src;
+        Sk4f d4 = dst;
+        return check_as_pmfloat(clamp_255(s4 + d4));
     }
     static const bool kFoldCoverageIntoSrcAlpha = true;
     static const SkXfermode::Mode kMode = SkXfermode::kPlus_Mode;
@@ -1275,7 +1283,9 @@
 struct Modulate4f {
     static SkPMFloat Xfer(const SkPMFloat& src, const SkPMFloat& dst) {
         const Sk4f inv255(gInv255);
-        return check_as_pmfloat(src * dst * inv255);
+        Sk4f s4 = src;
+        Sk4f d4 = dst;
+        return check_as_pmfloat(s4 * d4 * inv255);
     }
     static const bool kFoldCoverageIntoSrcAlpha = false;
     static const SkXfermode::Mode kMode = SkXfermode::kModulate_Mode;
@@ -1285,7 +1295,9 @@
 struct Screen4f {
     static SkPMFloat Xfer(const SkPMFloat& src, const SkPMFloat& dst) {
         const Sk4f inv255(gInv255);
-        return check_as_pmfloat(src + dst - src * dst * inv255);
+        Sk4f s4 = src;
+        Sk4f d4 = dst;
+        return check_as_pmfloat(s4 + d4 - s4 * d4 * inv255);
     }
     static const bool kFoldCoverageIntoSrcAlpha = true;
     static const SkXfermode::Mode kMode = SkXfermode::kScreen_Mode;
diff --git a/src/effects/SkColorMatrixFilter.cpp b/src/effects/SkColorMatrixFilter.cpp
index 196ebd0..a20c65a 100644
--- a/src/effects/SkColorMatrixFilter.cpp
+++ b/src/effects/SkColorMatrixFilter.cpp
@@ -269,7 +269,7 @@
 
 static Sk4f unpremul(const SkPMFloat& pm) {
     float scale = 255 / pm.a(); // candidate for fast/approx invert?
-    return pm * Sk4f(scale, scale, scale, 1);
+    return Sk4f(pm) * Sk4f(scale, scale, scale, 1);
 }
 
 static Sk4f clamp_0_255(const Sk4f& value) {
diff --git a/src/opts/SkNx_neon.h b/src/opts/SkNx_neon.h
index 4514829..c8580bc 100644
--- a/src/opts/SkNx_neon.h
+++ b/src/opts/SkNx_neon.h
@@ -187,6 +187,7 @@
     typedef SkNi<4, int32_t> Ni;
 public:
     SkNf(float32x4_t vec) : fVec(vec) {}
+    float32x4_t vec() const { return fVec; }
 
     SkNf() {}
     explicit SkNf(float val)           : fVec(vdupq_n_f32(val)) {}
@@ -251,7 +252,7 @@
         return vgetq_lane_f32(fVec, k&3);
     }
 
-protected:
+private:
     float32x4_t fVec;
 };
 
diff --git a/src/opts/SkNx_sse.h b/src/opts/SkNx_sse.h
index c1ae1f0..cae2d45 100644
--- a/src/opts/SkNx_sse.h
+++ b/src/opts/SkNx_sse.h
@@ -146,6 +146,7 @@
     typedef SkNi<4, int32_t> Ni;
 public:
     SkNf(const __m128& vec) : fVec(vec) {}
+    __m128 vec() const { return fVec; }
 
     SkNf() {}
     explicit SkNf(float val)           : fVec( _mm_set1_ps(val) ) {}
@@ -181,7 +182,7 @@
         return pun.fs[k&3];
     }
 
-protected:
+private:
     __m128 fVec;
 };
 
diff --git a/src/opts/SkPMFloat_SSE2.h b/src/opts/SkPMFloat_SSE2.h
index 88a3803..e10328a 100644
--- a/src/opts/SkPMFloat_SSE2.h
+++ b/src/opts/SkPMFloat_SSE2.h
@@ -17,7 +17,7 @@
     __m128i fix8    = _mm_set_epi32(0,0,0,c),
             fix8_16 = _mm_unpacklo_epi8 (fix8,    _mm_setzero_si128()),
             fix8_32 = _mm_unpacklo_epi16(fix8_16, _mm_setzero_si128());
-    fVec = _mm_cvtepi32_ps(fix8_32);
+    fColors = _mm_cvtepi32_ps(fix8_32);
     SkASSERT(this->isValid());
 }
 
@@ -27,7 +27,7 @@
 
 inline SkPMColor SkPMFloat::roundClamp() const {
     // We don't use _mm_cvtps_epi32, because we want precise control over how 0.5 rounds (up).
-    __m128i fix8_32 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), fVec)),
+    __m128i fix8_32 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), fColors.vec())),
             fix8_16 = _mm_packus_epi16(fix8_32, fix8_32),
             fix8    = _mm_packus_epi16(fix8_16, fix8_16);
     SkPMColor c = _mm_cvtsi128_si32(fix8);
@@ -37,7 +37,7 @@
 
 inline SkPMColor SkPMFloat::trunc() const {
     // Basically, same as roundClamp(), but no rounding.
-    __m128i fix8_32 = _mm_cvttps_epi32(fVec),
+    __m128i fix8_32 = _mm_cvttps_epi32(fColors.vec()),
             fix8_16 = _mm_packus_epi16(fix8_32, fix8_32),
             fix8    = _mm_packus_epi16(fix8_16, fix8_16);
     SkPMColor c = _mm_cvtsi128_si32(fix8);
@@ -66,10 +66,10 @@
         SkPMColor colors[4]) {
     // Same as _SSSE3.h's.  We use 3 _mm_packus_epi16() where the naive loop uses 8.
     // We don't use _mm_cvtps_epi32, because we want precise control over how 0.5 rounds (up).
-    __m128i c0 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), a.fVec)),
-            c1 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), b.fVec)),
-            c2 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), c.fVec)),
-            c3 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), d.fVec));
+    __m128i c0 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), a.fColors.vec())),
+            c1 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), b.fColors.vec())),
+            c2 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), c.fColors.vec())),
+            c3 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), d.fColors.vec()));
     __m128i c3210 = _mm_packus_epi16(_mm_packus_epi16(c0, c1),
                                      _mm_packus_epi16(c2, c3));
     _mm_storeu_si128((__m128i*)colors, c3210);
diff --git a/src/opts/SkPMFloat_SSSE3.h b/src/opts/SkPMFloat_SSSE3.h
index 9ff7356..c1c6d67 100644
--- a/src/opts/SkPMFloat_SSSE3.h
+++ b/src/opts/SkPMFloat_SSSE3.h
@@ -18,13 +18,13 @@
     const int _ = 255;  // _ means to zero that byte.
     __m128i fix8    = _mm_set_epi32(0,0,0,c),
             fix8_32 = _mm_shuffle_epi8(fix8, _mm_set_epi8(_,_,_,3, _,_,_,2, _,_,_,1, _,_,_,0));
-    fVec = _mm_cvtepi32_ps(fix8_32);
+    fColors = _mm_cvtepi32_ps(fix8_32);
     SkASSERT(this->isValid());
 }
 
 inline SkPMColor SkPMFloat::trunc() const {
     const int _ = 255;  // _ means to zero that byte.
-    __m128i fix8_32 = _mm_cvttps_epi32(fVec),
+    __m128i fix8_32 = _mm_cvttps_epi32(fColors.vec()),
             fix8    = _mm_shuffle_epi8(fix8_32, _mm_set_epi8(_,_,_,_, _,_,_,_, _,_,_,_, 12,8,4,0));
     SkPMColor c = _mm_cvtsi128_si32(fix8);
     SkPMColorAssert(c);
@@ -37,7 +37,7 @@
 
 inline SkPMColor SkPMFloat::roundClamp() const {
     // We don't use _mm_cvtps_epi32, because we want precise control over how 0.5 rounds (up).
-    __m128i fix8_32 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), fVec)),
+    __m128i fix8_32 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), fColors.vec())),
             fix8_16 = _mm_packus_epi16(fix8_32, fix8_32),
             fix8    = _mm_packus_epi16(fix8_16, fix8_16);
     SkPMColor c = _mm_cvtsi128_si32(fix8);
@@ -69,10 +69,10 @@
         SkPMColor colors[4]) {
     // Same as _SSE2.h's.  We use 3 _mm_packus_epi16() where the naive loop uses 8.
     // We don't use _mm_cvtps_epi32, because we want precise control over how 0.5 rounds (up).
-    __m128i c0 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), a.fVec)),
-            c1 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), b.fVec)),
-            c2 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), c.fVec)),
-            c3 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), d.fVec));
+    __m128i c0 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), a.fColors.vec())),
+            c1 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), b.fColors.vec())),
+            c2 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), c.fColors.vec())),
+            c3 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), d.fColors.vec()));
     __m128i c3210 = _mm_packus_epi16(_mm_packus_epi16(c0, c1),
                                      _mm_packus_epi16(c2, c3));
     _mm_storeu_si128((__m128i*)colors, c3210);
diff --git a/src/opts/SkPMFloat_neon.h b/src/opts/SkPMFloat_neon.h
index fcb9ba8..baad190 100644
--- a/src/opts/SkPMFloat_neon.h
+++ b/src/opts/SkPMFloat_neon.h
@@ -22,7 +22,7 @@
 }
 
 inline SkPMColor SkPMFloat::trunc() const {
-    uint32x4_t  fix8_32  = vcvtq_u32_f32(fVec);  // vcvtq_u32_f32 truncates
+    uint32x4_t  fix8_32  = vcvtq_u32_f32(fColors.vec());  // vcvtq_u32_f32 truncates
     uint16x4_t  fix8_16  = vmovn_u32(fix8_32);
     uint8x8_t   fix8     = vmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0)));
     SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0);
@@ -35,7 +35,7 @@
 }
 
 inline SkPMColor SkPMFloat::roundClamp() const {
-    float32x4_t add_half = vaddq_f32(fVec, vdupq_n_f32(0.5f));
+    float32x4_t add_half = vaddq_f32(fColors.vec(), vdupq_n_f32(0.5f));
     uint32x4_t  fix8_32  = vcvtq_u32_f32(add_half);  // vcvtq_u32_f32 truncates, so round manually
     uint16x4_t  fix8_16  = vqmovn_u32(fix8_32);
     uint8x8_t   fix8     = vqmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0)));