[gmx-developers] Wallcycle bug?

[Szilárd Páll]

On Tue, Sep 15, 2015 at 5:16 PM, Berk Hess <hess at kth.se> wrote:

> Hi,
>
> We didn't realize that modern x86 cpus no longer synchronize clocks
> between cores.


Actually, having read up a bit more, the _opposite_ seems to be the case:
on older CPUs there could be a problem, but on newer ones the TSC should be:
- "constant", i.e. not affected by frequency scaling,
- "nonstop", i.e. unaffected by CPU ACPI states.
The "constant_tsc" and "nonstop_tsc" cpuid flags indicate whether these are
supported (most AMD and Nehalem+ on Intel).

Intel seems to call both these cases "invariant TSC" (dev manual
section 17.14.1 [1]). Apparently a number sources claim [2,3] that
 invariant TSC implies that TSC is synchronized (or can be if the BIOS
syncs at boot) and the Intel docs seems to suggest this too by claiming
that the OS can use TSC for wall clock.


Based on this the issue you observed could only be caused by something
funny with your BIOS (or OS/kernel?). Thoughts?


Cheers,
--
Szilárd


[1]
http://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-software-developer-manual-325462.pdf

[2] http://www.gossamer-threads.com/lists/xen/devel/185419

[3]
https://software.intel.com/en-us/forums/intel-isa-extensions/topic/289368

We should get the fix for this merged asap:
> https://gerrit.gromacs.org/#/c/5091/
>
> Cheers,
>
> Berk
>
> On 09/15/2015 05:06 PM, David van der Spoel wrote:
>
>> Hi,
>>
>> my patch stumbled over this issue in Jenkins:
>>
>> Program:     gmx mdrun, VERSION 5.2-dev-20150915-ef00eb4-local
>> Source file: src/gromacs/timing/wallcycle.cpp (line 398)
>> Function:    void subtract_cycles(wallcc_t *, int, int)
>>
>> Assertion failed:
>> Condition: wcc[ewc_main].c >= wcc[ewc_sub].c
>> Subcounter cannot have more ticks than parent
>>
>> For more information and tips for troubleshooting, please check the
>> GROMACS
>> website at http://www.gromacs.org/Documentation/Errors
>>
>> Any clue? Or is this assertion just overly pedantic?
>>
>
> --
> Gromacs Developers mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers
> or send a mail to gmx-developers-request at gromacs.org.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20150915/879a0240/attachment-0001.html>