linear -> sRGB: use fast approximate sqrt()

src/effects/gradients/Sk4fLinearGradient.cpp

 /*
  * Copyright 2016 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #include "Sk4fLinearGradient.h"
 #include "Sk4x4f.h"
 #include "SkXfermode.h"
@@ skipping 35 matching lines @@
 // Planar version of ramp (S32 no-premul only).
 template<>
 void ramp<DstType::S32, ApplyPremul::False>(const Sk4f& c, const Sk4f& dc, SkPMColor dst[], int n) {
     SkASSERT(n > 0);
 
     const Sk4f    dc4 = dc * 4;
     const Sk4x4f dc4x = { Sk4f(dc4[0]), Sk4f(dc4[1]), Sk4f(dc4[2]), Sk4f(dc4[3]) };
     Sk4x4f        c4x = Sk4x4f::Transpose(c, c + dc, c + dc * 2, c + dc * 3);
 
     while (n >= 4) {
-        const Sk4x4f cx4s32 = { c4x.r.sqrt(), c4x.g.sqrt(), c4x.b.sqrt(), c4x.a };
+        const Sk4x4f cx4s32 = {
+            c4x.r.rsqrt().invert(),
+            c4x.g.rsqrt().invert(),
+            c4x.b.rsqrt().invert(),
+            c4x.a
+        };
         cx4s32.transpose((uint8_t*)dst);
 
         c4x.r += dc4x.r;
         c4x.g += dc4x.g;
         c4x.b += dc4x.b;
         c4x.a += dc4x.a;
 
         dst += 4;
         n   -= 4;
     }
@@ skipping 403 matching lines @@
         static_cast<const LinearGradient4fContext*>(state->fCtx);
 
     if (ctx->fColorsArePremul) {
         ctx->shadePremulSpan<DstType::F16, ApplyPremul::False>(
             x, y, dst.writable_addr64(x, y), count);
     } else {
         ctx->shadePremulSpan<DstType::F16, ApplyPremul::True>(
             x, y, dst.writable_addr64(x, y), count);
     }
 }