[Shootout-list] fannkuch (timer resolution; HZ=1000?)
Einar Karttunen
ekarttun@cs.helsinki.fi
Mon, 23 May 2005 17:26:25 +0300
Jon Harrop <jon@ffconsultancy.com> writes:
> On Monday 23 May 2005 10:39, Bengt Kleberg wrote:
>> if all bechmarks contained one measurement of the startup (for that
>> benchmark) and the posssibility to subtract that value from the other
>> measurments, then we would have a good comproimise. imho.
>
> Alas you can't do this because the startup time for each benchmark is likely
> to be quite different (at least for some languages) because different
> benchmarks will exercise different parts of the run-time which may incur
> their own startup.
This would be quite unfair.
Consider two languages (A and B) and a benchmark using the threading
subsystem. A will initiase the threading subsystem before main, while B
will initialise the threading subsystem when it is first used.
If we measure per-benchmark startup times then it will be likely that
either A or B will be punished. Thus the best way seems to be using hello
for the startup time.
- Einar Karttunen