Building NumPy and SciPy with oneMKL

These steps assume a Linux* or Windows* operating system, Intel® 64 architecture, and ILP64 interface.

1. Get the latest NumPy and SciPy packages from http://www.scipy.org/Download and unpack them

2. Install the latest versions oneMKL and Intel® C++ and Intel® Fortran Compilers

3. Set the environment variables for Intel C++ and Fortran compilers:

   • Linux*:

     Execute the command:

     $source <intel tools installation dir>/bin/compilervars.sh intel64

   • Windows*:

     Launch environment setters to specify the Visual Studio* mode for your Intel64 build binaries:

     1. (Windows 8:) Place the mouse pointer in the bottom-left corner of the screen, click the right mouse button, select Search, and click anywhere in the screen white space.

     2. Navigate to the Intel Parallel Studio 2016 section and select Intel64 Visual Studio 20XX mode.

4. Change directory to <numpy dir>

5. Make a copy of the existing site.cfg.example and save it as site.cfg

6. Open site.cfg, uncomment the [mkl] section, and modify it to look as follows:

   • Linux:

     [mkl]
     library_dirs = /opt/intel/compilers_and_libraries_2016/linux/mkl/lib/intel64
     include_dirs = /opt/intel/compilers_and_libraries_2016/linux/mkl/include
     mkl_libs = mkl_rt
     lapack_libs =
     

   • Windows:

     [mkl]
     library_dirs = C:\Program Files (x86)\IntelSWTools\compilers_and_libraries_2016\windows\mkl\lib\intel64;
     C:\Program Files (x86)\Intel\Composer XE 2015.x.yyy\compiler\lib\intel64
     include_dirs = C:\Program Files (x86)\IntelSWTools\compilers_and_libraries_2016\windows\mkl\include
     mkl_libs = mkl_lapack95_lp64,mkl_blas95_lp64,mkl_intel_lp64,mkl_intel_thread,mkl_core,libiomp5md
     lapack_libs = mkl_lapack95_lp64,mkl_blas95_lp64,mkl_intel_lp64,mkl_intel_thread,mkl_core,libiomp5md
     

7. Modify intelccompiler.py in <numpy dir>/distutils to pass optimization options to Intel C++ Compiler:

   • Linux:

     self.cc_exe = 'icc –O3 –g -xhost –fPIC  
     –fomit-frame-pointer –openmp –DMKL_ILP64'
     

   • Windows:

     self.compile_options = [ '/nologo', '/O3', '/MD', '/W3', '/Qstd=c99', 
     '/QxHost', '/fp:strict', '/Qopenmp']
     

8. Modify intel.py in the <numpy dir>/distutils/fcompiler folder to pass optimization options to Intel Fortran Compiler:

   • Linux:

     ifort –xhost –openmp –i8 –fPIC
     

   • Windows:

     def get_flags(self):
         opt = ['/nologo', '/MD', '/nbs','/names:lowercase', '/assume:underscore']
     

9. Change directory to <numpy dir> and build and install NumPy:

   • Linux:

     $python setup.py config --compiler=intelem build_clib 
     --compiler=intelem 
     build_ext --compiler=intelem install

   • Windows:

     python setup.py config --compiler=intelemw build_clib 
     --compiler=intelemw build_ext --compiler=intelemw install

10. Change directory to <scipy dir> and build and install SciPy:

    • Linux:

      $python setup.py config --compiler=intelem --fcompiler=intelem build_clib  
      --compiler=intelem --fcompiler=intelem build_ext --compiler=intelem 
      --fcompiler=intelem install
      

    • Windows:

      python setup.py config --compiler=intelemw --fcompiler=intelvem build_clib 
      --compiler=intelemw --fcompiler=intelvem build_ext --compiler=intelemw 
      --fcompiler=intelvem install
      

Code Example

import numpy as np
import scipy.linalg.blas as slb
import time

M = 10000
N = 6000
k_list = [64, 128, 256, 512, 1024, 2048, 4096, 8192]

np.show_config()

for K in k_list:
        a = np.array(np.random.random((M, N)), dtype=np.double, order='C', copy=False)
        b = np.array(np.random.random((N, K)), dtype=np.double, order='C', copy=False)
        A = np.matrix(a, dtype=np.double, copy=False)
        B = np.matrix(b, dtype=np.double, copy=False)

        start = time.time()
        C = slb.dgemm(1.0, a=A, b=B)
        end = time.time()

        tm = start - end
        print ('{0:4}, {1:9.7}'.format(K, tm))

Source code: see the dgemm_python folder in the samples archive available at https://software.intel.com/content/dam/develop/external/us/en/documents/mkl-cookbook-samples-120115.zip.