Remove warnings, improve benchmark quality, complicate benchmark results:
 - data on heap instead of stack
 - more closely match operations of actual SkMatrix::setConcat()
 - avoid random perturbations, do_always;
   use perf annotate to verify that assembly isn't getting optimized away



git-svn-id: http://skia.googlecode.com/svn/trunk@1500 2bbb7eff-a529-9590-31e7-b0007b416f81
1 file changed