[gmx-developers] checking CPU set
szilard.pall at cbr.su.se
Tue Oct 30 14:28:07 CET 2012
On Tue, Oct 30, 2012 at 8:49 AM, Sander Pronk <pronk at kth.se> wrote:
> What's the practical difference between the two cases? In both cases, the
> thread is allowed to run on all CPUs in the system, right?
The practical difference is that if we run on an 8-core CPU:
mdrun -ntmpi 1 -ntomp 4
we lock the threads to the first four cores, but if we run
taskset 0xFF mdrun -ntmpi 1 -ntomp 4
without distinguishing between the two cases I've just mentioned, we would
override the affinity set by taskset. Of course, the above does not make
much sense, but I would rather not have mdrun exhibit ambiguous behavior.
> On Oct 30, 2012, at 00:32 , Szilárd Páll <szilard.pall at cbr.su.se> wrote:
> > Hi,
> > In order to be nice and not have mdrun override externally set CPU
> affinities I want to check whether the affinity is default or it has been
> changed (by taskset, OpenMP library, etc.). Initially I want to implement
> this for GNU platforms with sched.h.
> > While I initially though if no affinity is set the cpu set should be
> empty, in fact it in this case CPU_ISSET will give the same result as if
> all CPUs were in the set -- which actually makes sense. However, this left
> me quite clueless about how to distinguish between the case when all CPUs
> are present in the cpu set and when e.g. all CPU are set by taskset, e.g.
> > $ taskset 0xF mdrun # quite useless on a quad-core machine
> > Does anybody have an idea how to distinguish these two cases. It would
> be possible, because e.g. hwloc-ps can do it (but they do it based on their
> own mask-representation which makes is a bit hard to figure out what I need
> from their code).
> > Thanks,
> > --
> > Szilárd
> > --
> > gmx-developers mailing list
> > gmx-developers at gromacs.org
> > http://lists.gromacs.org/mailman/listinfo/gmx-developers
> > Please don't post (un)subscribe requests to the list. Use the
> > www interface or send it to gmx-developers-request at gromacs.org.
> gmx-developers mailing list
> gmx-developers at gromacs.org
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-developers-request at gromacs.org.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the gromacs.org_gmx-developers