Open MPI logo

Hardware Locality Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |  

This web mail archive is frozen.

This page is part of a frozen web archive of this mailing list.

You can still navigate around this archive, but know that no new mails have been added to it since July of 2016.

Click here to be taken to the new web archives of this list; it includes all the mails that are in this frozen archive plus all new mails that have been sent to the list since it was migrated to the new archives.

Subject: Re: [hwloc-devel] release status
From: Samuel Thibault (samuel.thibault_at_[hidden])
Date: 2009-10-02 16:16:39


Jeff Squyres, le Fri 02 Oct 2009 13:53:48 -0400, a écrit :
> On Oct 2, 2009, at 1:39 PM, Samuel Thibault wrote:
>
> >I also think inlining doesn't bring much performance compared to the
> >cost of parsing a cpuset.
> >
> >> You'll get warnings from the linker about how the struct changed
> >size.
> >
> >Ah, in the static case, yes. But in our case cpusets would always be
> >dynamically allocated.
>
> It's not the size of the array in the cpuset struct that matters --
> it's the size of the struct.

Sure, I understood this.

> Are there any global constants akin to MPI_COMM_WORLD? (e.g., an empty
> cpuset, or a full cpuset)

No, that's why I'm saying in our case we don't have the problem.

> I *think* that we only run into problems if we use global constants
> provided by the library.

Yes, and we don't.

> But it might be worth testing the example I sent before, even with
> dynamically allocated handles.

Should work perfectly.

> Linkers are deep, dark voodoo with unexpected corner cases in round
> rooms. :-)

Sure, but linkers don't have anything to do with malloc().

Samuel