Timing routines about benchmarks

[bgin]

Hi @ivan-pi

You should use a timing method of the highest achievable precision and that means either the RDTSCP instruction or the fixed performance event: CPU_CLK_UNHALTED.THREAD.
You may try to use RDTSC instruction, but you should check if its RDTSC invariant version is supported by the specific CPU.
THis will require a check of MSR register.
The usage of aforementioned instructions will require a call to C wrapper.
P.s.
I would advice against usage of complex ADT timers, because of emission of additional machine code instructions before and after the measured code.
Simplicity is the most important here.

Below is an example (not Fortran) of timing methodology (loc: 317)
https://github.com/bgin/Guided-Missile-Modeling-Simulation/blob/master/ReleaseGuidedMissileSimPerfTests/GMS_perf_test_ComputeGrassParamEq_zmm16r4_call_scope_looped100x10000_non_instr_autocor.cpp

Regards

Bernard Gingold