"Undefined symbol: GOMP_parallel" on CentOS 6.5

[Pontus Stenetorp]

Hello everyone,

I am currently running Torch under CentOS 6.5 and I am having some
GOMP related issues that I have been unable to resolve. My personal
experience with OpenMP/GOMP is rather limited, but, as far as I
understand, it comes shipped with modern versions of GCC and should
just work for most cases. However, after running for example `th -e
'require "nn"'`, Torch goes down in flames with "Undefined symbol:
GOMP_parallel". The only workaround that I have come up with is to
compile without OpenMP support, by adding `-DWITH_OPENMP=OFF` to the
`.rockspec`. While this of course will get things working, it would
be great to actually be able to use OpenMP. Below are some of my
system details.

> cat /etc/centos-release
CentOS release 6.5 (Final)

> uname -a
Linux 2.6.32-431.11.2.el6.x86_64 #1 SMP Tue Mar 25 19:59:55 UTC
2014 x86_64 x86_64 x86_64 GNU/Linux

> gcc --version
gcc (GCC) 4.4.7 20120313 (Red Hat 4.4.7-4)

> locate libgomp.so
/usr/lib64/libgomp.so.1
/usr/lib64/libgomp.so.1.0.0

Have anyone had similar issues, if so, how did you work around them?
I am sure that there is something reasonably simple that I am missing,
but I am not familiar enough with the Torch code base yet to be able
to know where to look.

Pontus Stenetorp

[soumith]

hey.

does it work with:

LD_PRELOAD="/usr/lib64/libgomp.so.1" th -e "require 'nn'"

    Pontus Stenetorp

--
You received this message because you are subscribed to the Google Groups "torch7" group.
To unsubscribe from this group and stop receiving emails from it, send an email to torch7+un...@googlegroups.com.
To post to this group, send email to tor...@googlegroups.com.
Visit this group at http://groups.google.com/group/torch7.
For more options, visit https://groups.google.com/d/optout.

[Pontus Stenetorp]

On 29 July 2015 at 17:38, torch7 on behalf of smth chntla

<tor...@googlegroups.com> wrote:
>
> does it work with:
>
> LD_PRELOAD="/usr/lib64/libgomp.so.1" th -e "require 'nn'"

Sadly, no. Debugging linking errors is really not my forte, but I
would hate to have to give up running Torch with OpenMP on our
cluster.

Pontus

[soumith]

yea I really have no hints as to what's happening. for what it's worth, if you are using GPUs, using OpenMP shouldn't matter much for you.

--
You received this message because you are subscribed to the Google Groups "torch7" group.

[Pontus Stenetorp]

On 30 July 2015 at 00:17, torch7 on behalf of smth chntla

<tor...@googlegroups.com> wrote:
>
> yea I really have no hints as to what's happening. for what it's worth, if
> you are using GPUs, using OpenMP shouldn't matter much for you.

This was for our heterogeneous cluster, which we use for CPU
computation. Good news though, well, in a way. As I was about to
take another stab at getting things to work I cloned the latest code
for Torch and it now just works.

Pontus