Is necessary to introduce rdtsc_ordered?

[ZENOTME]

Hi, I notice that this library uses rdtsc to get the cycle. According to Pitfalls of TSC usage and implementation in linux kernel, rdtsc may encounter Out-of-Order Execution Issues. I'm not sure if this has been considered before — is this something that should be fixed?

[tisonkun]

cc @zhongzc @ariesdevil thoughts?

[BewareMyPower]

Yes, it's an issue. But after introducing the fence mechanism, the performance might be similar to or even worse than std::Instant::now(). See my benchmark results in #11.

If the code you want to measure is a very small piece of code, our rdtsc function above might need to be re-implement by LFENCE or RDTSCP.

However, this only affects the case that a very small piece of code is measured. So in most environment, it might not be a real issue.

I suggest documenting the trade-off of the performance improvement and recommend users to use std::Instant::now() if the trade-off is unacceptable