Caffe: Problems running test with MKL

[sguada]

It seems that something have change in Makefile that prevents make runtest to work with MKL

Now I'm getting the problem reported in #344

...: /misc/software-lin/x/mkl/lib/intel64/libmkl_intel_thread.so: undefined symbol: omp_get_num_procs

Maybe it is something related to the dynamic use of libraries.

[sguada]

Fixed by setting the right LD_LIBRARY_PATH

# LD_LIBRARY_PATH
export LD_LIBRARY_PATH=/usr/local/lib
# CUDA
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
# MKL
export LD_LIBRARY_PATH=/opt/intel/lib/intel64:/opt/intel/mkl/lib/intel64:$LD_LIBRARY_PATH
# OpenBLAS
export LD_LIBRARY_PATH=/opt/OpenBLAS/lib:$LD_LIBRARY_PATH

[sunsided]

Very nice. For me, the path to libiomp5.a did the trick - it was in /opt/intel/lib/intel64_lin.

[yazdanbakhsh]

I got this error and none of these solutions worked for me. Any solution?

[YerongLi]

I've got this libmkl_intel_thread.so: undefined symbol: omp_get_num_procs error too. Could anyone help me on this?
I added the path setting in my profile:

# MKL
export LD_LIBRARY_PATH=/opt/intel/lib/intel64:/opt/intel/mkl/lib/intel64:$LD_LIBRARY_PATH
# OpenBLAS
export LD_LIBRARY_PATH=/opt/OpenBLAS/lib:$LD_LIBRARY_PATH

But running python -c "import numpy as np;np.test('full')" stiil gives me a mkl error:

Running unit tests for numpy
NumPy version 1.13.3
NumPy relaxed strides checking option: True
NumPy is installed in /home/yerong/local/anaconda3/lib/python3.6/site-packages/numpy
Python version 3.6.2 |Anaconda custom (64-bit)| (default, Sep 30 2017, 18:42:57) [GCC 7.2.0]
nose version 1.3.7
...........................................................................................................................................................................................................................S....................python: symbol lookup error: /home/yerong/local/anaconda3/lib/python3.6/site-packages/numpy/core/../../../../libmkl_intel_thread.so: undefined symbol: omp_get_num_procs

I found I did not even have /opt/intel folder...

[YerongLi]

I reinstalled the anaconda and rebuild the caffe and everything works out fine after that.