make pm4f be RGBA always, not pmcolor order

src/core/SkPM4f.h

Index: src/core/SkPM4f.h
diff --git a/src/core/SkPM4f.h b/src/core/SkPM4f.h
index fb22783dcc44203a8c3931982f645e3de7dde956..2bf2f23a92e5634dd64132feb47f02d5a16da282 100644
--- a/src/core/SkPM4f.h
+++ b/src/core/SkPM4f.h
@@ -9,29 +9,73 @@
 #define SkPM4f_DEFINED
 
 #include "SkColorPriv.h"
+#include "SkNx.h"
+
+static inline Sk4f swizzle_rb(const Sk4f& x) {
+    return Sk4f(x[2], x[1], x[0], x[3]);

[mtklein, 2016/03/07 16:13:49]

We'll have better opportunity to improve this later if we write this as:
    return SkNx_shuffle<2,1,0,3>(x);

[reed1, 2016/03/07 17:39:07]

Done.

+}
+
+static inline Sk4f swizzle_rgba_to_pmorder(const Sk4f& x) {
+#ifdef SK_PMCOLOR_IS_BGRA
+    return Sk4f(x[2], x[1], x[0], x[3]);

[mtklein, 2016/03/07 16:13:49]

return swizzle_rb(x);  ?

[reed1, 2016/03/07 17:39:07]

Done.

+#else
+    return x;
+#endif
+}
+
+static inline Sk4f swizzle_pmorder_to_rgba(const Sk4f& x) {
+#ifdef SK_PMCOLOR_IS_BGRA
+    return Sk4f(x[2], x[1], x[0], x[3]);

[mtklein, 2016/03/07 16:13:49]

ditto

[reed1, 2016/03/07 17:39:07]

Done.

+#else
+    return x;
+#endif
+}
 
 /*
  *  The float values are 0...1 premultiplied
  */
 struct SkPM4f {
     enum {
+#if 0
         A = SK_A32_SHIFT/8,
         R = SK_R32_SHIFT/8,
         G = SK_G32_SHIFT/8,
         B = SK_B32_SHIFT/8,
+#else
+        R, G, B, A,
+#endif
     };
     float fVec[4];
 
     float a() const { return fVec[A]; }
 
-    SkColor4f unpremul() const;
+    SkPM4f rgba() const { return *this; }

[mtklein, 2016/03/07 16:13:49]

It seems weird for these to return SkPM4f.  Only rgba() really returns an
SkPM4f.

Is it possible to eliminate these entirely in favor of the to4f() versions? 
Those make sense to me, and will probably be faster than something like
Sk4f::Load(pmf.pmorder()).

[reed1, 2016/03/07 17:39:07]

Moved to call-site

+    SkPM4f bgra() const { return{{ fVec[2], fVec[1], fVec[0], fVec[3] }}; }
+    SkPM4f pmorder() const {
+#ifdef SK_PMCOLOR_IS_BGRA
+        return this->bgra();
+#else
+        return this->rgba();
+#endif
+    }
 
+    static SkPM4f From4f(const Sk4f& x) {
+        SkPM4f pm;
+        x.store(pm.fVec);
+        return pm;
+    }
+    static SkPM4f FromF16(const uint16_t[4]);
     static SkPM4f FromPMColor(SkPMColor);
 
-    // half-float routines
+    Sk4f to4f() const { return Sk4f::Load(fVec); }
+    Sk4f to4f_rgba() const { return this->to4f(); }
+    Sk4f to4f_bgra() const { return swizzle_rb(this->to4f()); }
+    Sk4f to4f_pmorder() const { return swizzle_rgba_to_pmorder(this->to4f()); }
+
     void toF16(uint16_t[4]) const;
     uint64_t toF16() const; // 4 float16 values packed into uint64_t
-    static SkPM4f FromF16(const uint16_t[4]);
+
+    SkColor4f unpremul() const;
 
 #ifdef SK_DEBUG
     void assertIsUnit() const;
@@ -42,5 +86,4 @@ struct SkPM4f {
 
 typedef SkPM4f (*SkXfermodeProc4f)(const SkPM4f& src, const SkPM4f& dst);
 
-
 #endif