One thing the author of this post is doing here is calling gettimeofday() in every iteration through the loops. That's an awful way to benchmark something. gett

One thing the author of this post is doing here is calling gettimeofday() in every iteration through the loops. That's an awful way to benchmark something. gettimeofday() issues a serializing instruction, which can potentially flush the entire pipeline. It can also cause cores to sync their clocks. Either way, the net result of calling gettimeofday() too often is awful multithreaded performance. Calling it much less often or using a profiler would give very different results than reported.