__rdtscp vs __rdtsc in logger_benchmarks

[DanielekW]

I have a question to logger_benchmarks. In ns_per_rdtsc_tick() to convert rdtsc to wall time __rdtsc() is used while next in benchmark tests __rdtscp(&aux) is used. Why in ns_per_rdtsc_tick() you do not use __rdtscp?

[odygrd]

__rdtsc() should have lower overhead. __rdtscp() has an additional serialising barrier, I believe I used it in benchmarks for more accurate results.