Fix for nexus 5 crashing in GL benches

bench/Benchmark.h

 /*
  * Copyright 2011 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #ifndef Benchmark_DEFINED
 #define Benchmark_DEFINED
 
@@ skipping 57 matching lines @@
     }
 
     // Call before draw, allows the benchmark to do setup work outside of the
     // timer. When a benchmark is repeatedly drawn, this should be called once
     // before the initial draw.
     void preDraw();
 
     // Called once before and after a series of draw calls to a single canvas.
     // The setup/break down in these calls is not timed.
     void perCanvasPreDraw(SkCanvas*);
+    void perCanvasLastCallBeforeDraw(SkCanvas*);

[mtklein_C, 2015/09/30 17:26:29]

Having names this similar to the perCanvas-flavored calls seems error prone,
especially if we document that we call them once but actually call them for each
sample.

I'm inclined to write it like this:
  // Called just before and after each call to draw().  Not timed.
  void preDraw(SkCanvas*)
  void postDraw(SkCanvas*)

Now, this does have a slight problem, which is that we've already (mis-) used
the name "preDraw()".  So I propose this sequence of global edits to your patch
here:
   1)  s/preDraw/delayedSetup/g
   2) s/onPreDraw/onDelayedSetup/g
   3) s/perCanvasLastCallBeforeDraw/preDraw/g
   4) s/onPerCanvasLastCallBeforeDraw/onPreDraw/g
   5) s/perCanvasFirstCallAfterDraw/postDraw/g
   6) s/onPerCanvasFirstCallAfterDraw/onPostDraw/g
   7) Separate preDraw / postDraw from perCanvasPreDraw and perCanvasPostDraw,
and add comments.

I'd also be fine if you made these directly virtual rather than jumping through
the foo / onFoo hoop.  There's no need to do that for internal tools.  It's also
fine by me if you want to leave it.

[joshualitt, 2015/09/30 17:57:17]

done
Meh, I agree with you, but lets just keep it consistent.

+    void perCanvasFirstCallAfterDraw(SkCanvas*);
     void perCanvasPostDraw(SkCanvas*);
 
     // Bench framework can tune loops to be large enough for stable timing.
     void draw(const int loops, SkCanvas*);
 
     void setForceAlpha(int alpha) {
         fForceAlpha = alpha;
     }
 
     void setDither(SkTriState::State state) {
@@ skipping 19 matching lines @@
      */
     virtual bool isVisual() { return false; }
 
 protected:
     virtual void setupPaint(SkPaint* paint);
 
     virtual const char* onGetName() = 0;
     virtual const char* onGetUniqueName() { return this->onGetName(); }
     virtual void onPreDraw() {}
     virtual void onPerCanvasPreDraw(SkCanvas*) {}
+    virtual void onPerCanvasLastCallBeforeDraw(SkCanvas*) {}
+    virtual void onPerCanvasFirstCallAfterDraw(SkCanvas*) {}
     virtual void onPerCanvasPostDraw(SkCanvas*) {}
     // Each bench should do its main work in a loop like this:
     //   for (int i = 0; i < loops; i++) { <work here> }
     virtual void onDraw(const int loops, SkCanvas*) = 0;
 
     virtual SkIPoint onGetSize();
 
 private:
     int     fForceAlpha;
     SkTriState::State  fDither;
     uint32_t    fOrMask, fClearMask;
 
     typedef SkRefCnt INHERITED;
 };
 
 typedef SkTRegistry<Benchmark*(*)(void*)> BenchRegistry;
 
 #endif