Split rsqrt into rsqrt{0,1,2}, with increasing cost and precision on ARM

src/core/SkNx.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 SkNx_DEFINED
 #define SkNx_DEFINED
 
@@ skipping 101 matching lines @@
     Nb operator >= (const SkNf& o) const { return Nb(fLo >= o.fLo, fHi >= o.fHi); }
 
     static SkNf Min(const SkNf& l, const SkNf& r) {
         return SkNf(SkNf<N/2,T>::Min(l.fLo, r.fLo), SkNf<N/2,T>::Min(l.fHi, r.fHi));
     }
     static SkNf Max(const SkNf& l, const SkNf& r) {
         return SkNf(SkNf<N/2,T>::Max(l.fLo, r.fLo), SkNf<N/2,T>::Max(l.fHi, r.fHi));
     }
 
     SkNf  sqrt() const { return SkNf(fLo. sqrt(), fHi. sqrt()); }
-    SkNf rsqrt() const { return SkNf(fLo.rsqrt(), fHi.rsqrt()); }
+
+    // Generally, increasing precision, increasing cost.
+    SkNf rsqrt0() const { return SkNf(fLo.rsqrt0(), fHi.rsqrt0()); }
+    SkNf rsqrt1() const { return SkNf(fLo.rsqrt1(), fHi.rsqrt1()); }
+    SkNf rsqrt2() const { return SkNf(fLo.rsqrt2(), fHi.rsqrt2()); }
 
     SkNf       invert() const { return SkNf(fLo.      invert(), fHi.      invert()); }
     SkNf approxInvert() const { return SkNf(fLo.approxInvert(), fHi.approxInvert()); }
 
     template <int k> T kth() const {
         SkASSERT(0 <= k && k < N);
         return k < N/2 ? fLo.template kth<k>() : fHi.template kth<k-N/2>();
     }
 
 private:
@@ skipping 67 matching lines @@
     Nb operator != (const SkNf& o) const { return Nb(fVal != o.fVal); }
     Nb operator  < (const SkNf& o) const { return Nb(fVal  < o.fVal); }
     Nb operator  > (const SkNf& o) const { return Nb(fVal  > o.fVal); }
     Nb operator <= (const SkNf& o) const { return Nb(fVal <= o.fVal); }
     Nb operator >= (const SkNf& o) const { return Nb(fVal >= o.fVal); }
 
     static SkNf Min(const SkNf& l, const SkNf& r) { return SkNf(SkTMin(l.fVal, r.fVal)); }
     static SkNf Max(const SkNf& l, const SkNf& r) { return SkNf(SkTMax(l.fVal, r.fVal)); }
 
     SkNf  sqrt() const { return SkNf(Sqrt(fVal));        }
-    SkNf rsqrt() const { return SkNf((T)1 / Sqrt(fVal)); }
+    SkNf rsqrt0() const { return SkNf((T)1 / Sqrt(fVal)); }
+    SkNf rsqrt1() const { return this->rsqrt0(); }
+    SkNf rsqrt2() const { return this->rsqrt1(); }
 
     SkNf       invert() const { return SkNf((T)1 / fVal); }
     SkNf approxInvert() const { return this->invert();    }
 
     template <int k> T kth() const {
         SkASSERT(k == 0);
         return fVal;
     }
 
 private:
@@ skipping 35 matching lines @@
 typedef SkNf<4,    float> Sk4f;
 typedef SkNf<4,   double> Sk4d;
 typedef SkNf<4, SkScalar> Sk4s;
 
 typedef SkNi<4, uint16_t> Sk4h;
 typedef SkNi<8, uint16_t> Sk8h;
 
 typedef SkNi<4, int> Sk4i;
 
 #endif//SkNx_DEFINED