Numpy + MKL only using 1 core of cpu on Ubuntu 16.04

[xiao__yuanzheng]

Hi,

I am using Ubuntu 16.04 on my pc with "Intel® Core™ i7-7700K CPU @ 4.20GHz × 8".

My numpy and scipy only use one cpu when I try to do some element calculation for my numpy ndarray. 

Something like:

numpy.power(matrix, 1.5)

I compiled the numpy and scipy following https://software.intel.com/en-us/articles/numpyscipy-with-intel-mkl?page=1

The numpy configurations are as following.

blas_opt_info:

    include_dirs = ['/opt/intel/compilers_and_libraries_2018/linux/mkl/include']

    library_dirs = ['/opt/intel/compilers_and_libraries_2018/linux/mkl/lib/intel64']

    define_macros = [('SCIPY_MKL_H', None), ('HAVE_CBLAS', None)]

    libraries = ['mkl_rt', 'pthread']

lapack_opt_info:

    include_dirs = ['/opt/intel/compilers_and_libraries_2018/linux/mkl/include']

    library_dirs = ['/opt/intel/compilers_and_libraries_2018/linux/mkl/lib/intel64']

    define_macros = [('SCIPY_MKL_H', None), ('HAVE_CBLAS', None)]

    libraries = ['mkl_rt', 'pthread']

blas_mkl_info:

    include_dirs = ['/opt/intel/compilers_and_libraries_2018/linux/mkl/include']

    library_dirs = ['/opt/intel/compilers_and_libraries_2018/linux/mkl/lib/intel64']

    define_macros = [('SCIPY_MKL_H', None), ('HAVE_CBLAS', None)]

    libraries = ['mkl_rt', 'pthread']

lapack_mkl_info:

    include_dirs = ['/opt/intel/compilers_and_libraries_2018/linux/mkl/include']

    library_dirs = ['/opt/intel/compilers_and_libraries_2018/linux/mkl/lib/intel64']

    define_macros = [('SCIPY_MKL_H', None), ('HAVE_CBLAS', None)]

    libraries = ['mkl_rt', 'pthread']

I tried to modify environment variables like MKL_NUM_THREADS, OMP_NUM_THREADS, MKL_DOMAIN_NUM_THREADS, MKL_DYNAMIC, but they do nothing to my situation. 

Thanks for your help. 

[xiao__yuanzheng]

May I bring some more information.

The intel precompiled python runs well in my computer with multicores. 

And the numpy I built can also do matrix product using multicores, but for element-wise product only one core is going to use (intel python will use at least 3).

Any suggestions? Thanks.

[Ying_H_Intel]

Hi Yuanzheng,

​Your observation give the hint for the answer of your question :) 
​The build Numpy, Scipy with MKL is  based on  configuration:

   define_macros = [('SCIPY_MKL_H', None), ('HAVE_CBLAS', None)]

So it adopts  part of MKL,   here CBLAS function  http://www.netlib.org/blas/​, which only have functions like matrix product gemm, matrix*vector etc. It don't include element-wise functions.  So you can't see MKL acceleration or multi-core for such operation in your build.

And Intel distributed Python did more optimization,  including MKL, DAAL, vector math, multithreading etc.  So you will see more performance benefits .

Best Regards
​Ying

[xiao__yuanzheng]

Hi Ying,

Do you have any suggestions for how to build numpy/scipy with those optimization functions?

Thanks.

[Ying_H_Intel]

Hi Yuanzheng,

​What i can suggest,

maybe easier way , to use Intel distributed Python, which is free and compatible with other conda package.

second way, do change the numpy/scipy source code manually , for example change their Vector implementation with MKL VML function etc. As Intel python developer did.

​third way, specific for your project,  for example, evaluate the hot functions list you will use, then manually optimize them by your function to replace the numpy/scipy implementation.

​Best Regards,
​Ying