Library Link Error

[Toshiyuki_N_]

Hi: I tried to link Theano with the your python library.

When I tried to test Theano

 python -c "import theano; theano.test()"

I got the following error

Exception: Compilation failed (return status=1): /usr/bin/ld: /home/t-nakata/intel/python/2.7.20150803_184913/lib/libpython2.7.a(abstract.o): relocation R_X86_64_32 against `.rodata.str1.4' can not be used when making a shared object; recompile with -fPIC. /home/t-nakata/intel/python/2.7.20150803_184913/lib/libpython2.7.a: could not read symbols: Bad value. collect2: error: ld returned 1 exit status.

I think the libraries need to be compiled with -fPIC option. (My two cents' worth)

Best Regards

Toshiyuki Nakata

[Anton_M_Intel]

Hi Toshiyuki,

Thank you for this report! I've reproduced the issue and filled the bug report.

Another possible resolution is to build libpython2.7.so which is not provided along with our distribution.

Thanks
// Anton
 

[Michael_L_4]

Hi Anton,

I ran into the same problem, is there a way to get the distribution compiled with the flag --enable-shared?

-Michael

[Ricardo_C_Intel]

Hello Michael,
The Tech Preview release does not provide the necessary libraries, but we are working to fix it in the next release.

Thank you.

[glenn_n_]

This is a workaround I just stumbled upon:

# download Python 2.7.10 (same version as the Intel bundle)
wget http://python.org/ftp/python/2.7.10/Python-2.7.10.tar.xz

# extract it and recompile it in a shared manner
tar xf Python-2.7.10.tar.xz && cd cd Python-2.7.10
./configure --enable-shared
make

# COPY the shared libraries over into your intel python lib directory
cp libpython2.7.a /<intel directory>/lib

# don't know whether this one is actually needed here
cp libpython2.7.so /<intel directory>/lib

@Intel, maybe this helps you too :)

[Toshiyuki_N_]

Hi Glenn: Thank you for the workaround.

Unfortunately, I had wanted to test the Intel optimized one.

Perhaps people in Intel could compule their library with the --enable-shared option ?

Best Regards

[Rob_H_]

Just want to add a +1 to this.  I'd love to be able to test Caffe with the Intel-optimized Python, but right now I'm getting the exact same error as Toshiyuki when compiling Caffe against the Intel distribution.

Sounds like it's in the works to make a release with libpython2.7.so included.  Is there a timeline on that release?

[Robert_C_Intel]

We can't promise a date for a fix but ...

If you registered for the tech preview you will get a email about the next release. I expect it will happen by April 1.

Robert

[Rob_H_]

Awesome.  Thanks Robert.

[Toshiyuki_N_]

Two likes to Rob and Robert ;-)

[Toshiyuki_N_]

Just one more comment both Theano and Caffe are very important libraries for Deep Learning. So, to be able to prove that the

performance of these libraries is a very important point..

[Robert_C_Intel · Accepted solution]

Caffe and Theano are important for Intel. In the next release any improvements will only come from having better BLAS in numpy. Over time, we want to do more.

[Toshiyuki_N_]

There is Chainer as well..  http://chainer.org/

Best Regards

TN

[hussain__zafar]

Neon by Nervana is also a good Neural network toolkit, hopefully it will be included in future Intel Python releasse

cheers

zh