I think that sub-ms times are not really reliable. If you're interested in 
accurate benchmarks, you might want to consider either running the code many 
times in a loop (but make sure the optimizer doesn't eliminate the actual 
calculation!), or better, using a larger problem instance.

Reply via email to