Re: Hurd and linux timing differences

Hello,

João Pedro Malhado, le sam. 10 janv. 2026 19:51:39 +0000, a ecrit:
> I was just wondering if anyone would have any insight on what may be causing
> this difference. I note that on linux more information about the CPU cache is
> reported, so could it be that on hurd relevant CPU features for the tests are
> not picked up?

Possibly, yes.

> I notice that the timing numbers on the hurd are usually round
> numbers, so could there be problems with the timers?

It's simply that for more precise timing you need to use the HPET
driver, which is available only with an acpi-enabled kernel. I wouldn't
bother, though, the measurement is still representative, if not very
precise.

> I would think that on this CPU the lack of SMP on gnumach would not
> explain such difference.

These look like single-threaded tests indeed.

Most tests are fine, the only ones that I see odd are

1: BM_ManualTiming/1/real_time                19999981 ns

It's probably simply waiting for a whole clock tick (10ms) for some
reason. Probably worth having some look at what it actually does,
precisely.

Samuel