Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] 1.7.5: OpenShmem symbols are not weak
From: Mike Dubman (miked_at_[hidden])
Date: 2014-04-07 18:37:24


Hi,
it is not required by spec but we have a patch which is in the latest
stages of verification.
hope it will make into trunk this week.

M

On Tue, Apr 8, 2014 at 1:35 AM, Jeff Squyres (jsquyres)
<jsquyres_at_[hidden]>wrote:

> Mike / Mellanox --
>
> Does does OpenShmem define that there should be a profiling interface for
> SHMEM API functions (like MPI does)?
>
>
> On Mar 31, 2014, at 7:21 PM, Mike Dubman <miked_at_[hidden]> wrote:
>
> > thanks, we will fix that.
> >
> >
> > On Mon, Mar 31, 2014 at 5:04 PM, Ralph Castain <rhc_at_[hidden]> wrote:
> > Hmmm...I'll ask the Mellanox folks to fix that, but it will have to be
> for 1.8.1 as we have to release 1.8.0 today for the upstream packagers. I
> expect 1.8.1 will go out in the next 1-2 months based on prior experience
> (we always get a bunch of corrections after initial stable release as more
> people pick those up).
> >
> >
> > On Mar 31, 2014, at 1:14 AM, Bert Wesarg <Bert.Wesarg_at_[hidden]>
> wrote:
> >
> > > Hi all,
> > >
> > > while working on support for OpenShmem in our next Score-P measurement
> [1] release, we noticed that the liboshmem.so library from Open MPI does
> not mark the symbols as weak as one would expect. Here is an output of
> liboshmem.so on my 64bit ubuntu system. Lets try shmem_init:
> > >
> > > $ readelf -W --dyn-syms liboshmem.so.1.0.0 |grep 'shmem_init$'
> > > 1098: 000000000001dac2 141 FUNC GLOBAL DEFAULT 11 shmem_init
> > > 1228: 000000000001c73a 236 FUNC GLOBAL DEFAULT 11
> oshmem_shmem_init
> > >
> > > Oh, no pshmem_init. Maybe missing, but its legacy anyway. Lets try
> shmem_broadcast:
> > >
> > > $ readelf -W --dyn-syms liboshmem.so.1.0.0 |grep 'shmem_broadcast$'
> > > 555: 000000000002769a 92 FUNC GLOBAL DEFAULT 11
> pshmem_broadcast
> > > 577: 000000000001ed82 92 FUNC GLOBAL DEFAULT 11
> shmem_broadcast
> > >
> > > Ok, we see the pshmem_broadcast and the shmem_broadcast symbol, but
> the latter is not WEAK. Btw, this function is not in the standard. Thus
> (finally) take a symbol from the standard:
> > >
> > > $ readelf -W --dyn-syms liboshmem.so.1.0.0 |grep 'shmem_barrier$'
> > > 511: 000000000001d47e 24 FUNC GLOBAL DEFAULT 11
> oshmem_shmem_barrier
> > > 525: 000000000001f250 217 FUNC GLOBAL DEFAULT 11 shmem_barrier
> > > 1099: 0000000000027b68 217 FUNC GLOBAL DEFAULT 11
> pshmem_barrier
> > >
> > > Again, no WEAK for shmem_barrier.
> > >
> > > For reference, the output for MPI_Init:
> > >
> > > $ readelf -W --dyn-syms libmpi.so.1.4.0 |grep 'MPI_Init$'
> > > 1383: 00000000000841ff 484 FUNC WEAK DEFAULT 11 MPI_Init
> > > 2051: 00000000000841ff 484 FUNC GLOBAL DEFAULT 11 PMPI_Init
> > >
> > > HTH,
> > > Bert
> > >
> > > [1] http://www.vi-hps.org/projects/score-p
> > >
> > > --
> > > Dipl.-Inf. Bert Wesarg
> > > wiss. Mitarbeiter
> > >
> > > Technische Universität Dresden
> > > Zentrum für Informationsdienste und Hochleistungsrechnen (ZIH)
> > > 01062 Dresden
> > > Tel.: +49 (351) 463-42451
> > > Fax: +49 (351) 463-37773
> > > E-Mail: Bert.Wesarg_at_[hidden]
> > >
> > > _______________________________________________
> > > devel mailing list
> > > devel_at_[hidden]
> > > Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/devel
> > > Link to this post:
> http://www.open-mpi.org/community/lists/devel/2014/03/14429.php
> >
> > _______________________________________________
> > devel mailing list
> > devel_at_[hidden]
> > Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/devel
> > Link to this post:
> http://www.open-mpi.org/community/lists/devel/2014/03/14430.php
> >
> > _______________________________________________
> > devel mailing list
> > devel_at_[hidden]
> > Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/devel
> > Link to this post:
> http://www.open-mpi.org/community/lists/devel/2014/03/14434.php
>
>
> --
> Jeff Squyres
> jsquyres_at_[hidden]
> For corporate legal information go to:
> http://www.cisco.com/web/about/doing_business/legal/cri/
>
> _______________________________________________
> devel mailing list
> devel_at_[hidden]
> Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/devel
> Link to this post:
> http://www.open-mpi.org/community/lists/devel/2014/04/14465.php
>