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

src/opts/SkNx_sse.h

Index: src/opts/SkNx_sse.h
diff --git a/src/opts/SkNx_sse.h b/src/opts/SkNx_sse.h
index 2608525bd0189530b16ca470de56683831358803..cbe624ba2dbaaaa6edca19b32989b3a8a7a3c4d4 100644
--- a/src/opts/SkNx_sse.h
+++ b/src/opts/SkNx_sse.h
@@ -82,7 +82,9 @@ public:
     static SkNf Max(const SkNf& l, const SkNf& r) { return _mm_max_ps(l.fVec, r.fVec); }
 
     SkNf  sqrt() const { return _mm_sqrt_ps (fVec);  }
-    SkNf rsqrt() const { return _mm_rsqrt_ps(fVec); }
+    SkNf rsqrt0() const { return _mm_rsqrt_ps(fVec); }
+    SkNf rsqrt1() const { return this->rsqrt0(); }
+    SkNf rsqrt2() const { return this->rsqrt1(); }
 
     SkNf       invert() const { return SkNf(1) / *this; }
     SkNf approxInvert() const { return _mm_rcp_ps(fVec); }
@@ -126,7 +128,9 @@ public:
     static SkNf Max(const SkNf& l, const SkNf& r) { return _mm_max_pd(l.fVec, r.fVec); }
 
     SkNf  sqrt() const { return _mm_sqrt_pd(fVec);  }
-    SkNf rsqrt() const { return _mm_cvtps_pd(_mm_rsqrt_ps(_mm_cvtpd_ps(fVec))); }
+    SkNf rsqrt0() const { return _mm_cvtps_pd(_mm_rsqrt_ps(_mm_cvtpd_ps(fVec))); }
+    SkNf rsqrt1() const { return this->rsqrt0(); }
+    SkNf rsqrt2() const { return this->rsqrt1(); }
 
     SkNf       invert() const { return SkNf(1) / *this; }
     SkNf approxInvert() const { return _mm_cvtps_pd(_mm_rcp_ps(_mm_cvtpd_ps(fVec))); }
@@ -210,7 +214,9 @@ public:
     static SkNf Max(const SkNf& l, const SkNf& r) { return _mm_max_ps(l.fVec, r.fVec); }
 
     SkNf  sqrt() const { return _mm_sqrt_ps (fVec);  }
-    SkNf rsqrt() const { return _mm_rsqrt_ps(fVec); }
+    SkNf rsqrt0() const { return _mm_rsqrt_ps(fVec); }
+    SkNf rsqrt1() const { return this->rsqrt0(); }
+    SkNf rsqrt2() const { return this->rsqrt1(); }
 
     SkNf       invert() const { return SkNf(1) / *this; }
     SkNf approxInvert() const { return _mm_rcp_ps(fVec); }